Don’t pity me, because I remember hearing stories as a kid about someone who decided to urinate on one of these fences, which always struck me as an especially ambitious way to learn about electrical conductivity. Urine contains plenty of electrolytes and conducts electricity pretty well, though there may be easier ways to discover this.

With that context, watch this demonstration of the G.L.O.V.E.

The officer in the video seems to know what’s coming. Another man, wearing a pair of black gloves, takes hold of him. He cries out, then drops onto the padded floor. Afterward, he explains that he tried to pull away and found that he had “no control.” As he puts it, “there was just nothing there.” Similar demonstrations show officers yelling or falling to the ground while the glove is in contact with them (Anderson, 2026). Everything here is consistent with receiving an electric jolt.

The gloves in the demo are called the CT-G5 G.L.O.V.E., short for Generated Low Output Voltage Emitter, and are sold by Compliant Technologies. U.S. Immigration and Customs Enforcement (ICE) is preparing to buy them for Homeland Security Investigations and Enforcement and Removal Operations. A Department of Homeland Security acquisition may be completed as soon as March 31, 2027 (U.S. Department of Homeland Security, 2026), and ICE is apparently developing training protocols and policy, so it may be a while before officers begin carrying them operationally in our communities (Lovan & Foley, 2026).

“Less-lethal” methods are used with the justification that law-enforcement officers sometimes have to take physical control of people who are frightened, intoxicated, or determined not to be arrested. The history of policing includes such measures, including batons, hands-on restraint, chemical irritants, contact electrical devices and eventually conducted-energy weapons such as the TASER. Research on less-lethal technologies has found that properly used conducted-energy devices can reduce injuries to both officers and suspects in some use-of-force encounters (National Institute of Justice, 2011).

ICE encounters are a bit different than general law enforcement. In some ICE encounters, the original reason for taking someone into custody may be civil rather than criminal. The justification for escalating force can therefore arise during the encounter itself, such as when officers perceive resistance or a threat. Any device capable of eliciting involuntary movement has the potential to influence the very behavioral cues that may trigger such an escalation.

What electrical parameters are at play here? Voltage describes the electrical potential that drives charge through a circuit. Current tells us how quickly that charge is moving. A coulomb measures the amount of electrical charge itself (technically it’s the integral of current over time). One coulomb represents about 6.24 billion billion elementary charges (National Institute of Standards and Technology [NIST], 2018). To put that scale in perspective, NIST estimates that an average lightning bolt transfers about 5 coulombs of charge. A more appropriate scale for comparing the effects of less-lethal measures on biological tissues is the microcoulomb (μC), or millionths of a coulomb. For a pulsed electrical device, this is an especially helpful number because it incorporates both the strength of the current and how long each pulse lasts. A very high-voltage pulse can carry relatively little charge if it’s brief or severely current-limited. A lower-voltage device might deliver more. When comparing cattle fences, stun guns, TASERs, and the G.L.O.V.E., the charge delivered in each pulse can then tell us more about the electrical stimulus reaching excitable tissue than just the rated voltage.

The effect on the body also depends on the area through which that charge enters the body. The same 100 μC delivered through a fingertip-sized contact produces a much greater local charge density than 100 μC distributed across a broad electrode. The manufacturer publishes the G.L.O.V.E.’s total pulse charge. I couldn’t find a published specification for the effective electrode area or charge density during use.

An electric fence delivers its charge in brief pulses, generally separated by about a second. This is good for a slow-moving cow that might brush up against the fence to produce a zap. In one laboratory measurement of a modern electric-fence energizer, a pulse delivered about 600 μC of charge through a standard 500-ohm test load (Kroll et al., 2020). The shock gets a cow’s attention (and it got mine). There’s also plenty of time before the next one to move away from it (or in my case, to let go).

The G.L.O.V.E. is a bit different. A 3.7-volt lithium-ion battery and electronics are housed in the glove. Once the device is armed, conductive contacts deliver electrical stimulation when the conductive palm makes contact with skin. The CT-G5 produces about 324 to 362 volts, with an upper limit of 380 volts. The individual pulses last approximately 114 to 116 microseconds and arrive at a rate of 29 to 30 pulses per second (Compliant Technologies, 2024). This isn’t meant to convey that the glove provides a larger biological “dose”, just that the temporal pattern of stimulation, and the total charge delivered over a given interval, are different.

Those voltage numbers sound modest next to consumer stun guns, some of which advertise hundreds of thousands or even millions of volts. But comparing μC tells a different story.

The glove delivers between 72.4 and 111.7 μC with every pulse. But remember, this is repeated about 30 times each second (Compliant Technologies, 2024). Independently tested consumer contact stun guns from SABRE deliver roughly 0.8 to 2.5 μC (SABRE, n.d.). (A published study of four G.L.O.V.E. devices reported broadly similar electrical characteristics across physiologically relevant loads (Jiang & Panescu, 2019)).

The approximately 600 μC delivered by a cattle fence energizer tested by Kroll and colleagues was substantially greater than the charge in one G.L.O.V.E. pulse. The key difference is that the fence delivers that pulse as a blip, only about once a second. The G.L.O.V.E. delivers thirty pulses in that same interval. A person in sustained contact with the glove can therefore receive in total roughly 2,100 to 3,350 μC of charge each second, depending on where the device falls within its specified output range. The temporal structure of the two stimuli is very different (Compliant Technologies, 2024; Kroll et al., 2020).

How does the G.L.O.V.E. compare with a TASER? Technical documents describe a conducted-energy waveform of about 63 μC per pulse at 22 pulses per second as sufficient to produce neuromuscular incapacitation when the electrodes have adequate separation (Axon Enterprise, 2022). At first blush, the charge per pulse for the glove is in the same general range as a TASER.

But relative charge is only part of the problem. The contact area and spacing of the electrodes determine current density and the volume and location of tissue exposed to the electric field. TASER probes are deliberately spread across the body, allowing the electrical field to recruit motor nerves controlling substantial groups of muscles. The G.L.O.V.E. has closely spaced contacts, which would presumably create a much more local current path. So while comparing the charge of these devices is imperfect, the amount of charge in an individual G.L.O.V.E. pulse is in the range used by law-enforcement conducted-energy weapons.

The manufacturer describes the G.L.O.V.E.’s mechanism as stimulation of the sensory portion of the peripheral nervous system. The company calls the resulting effect “Neuro-Peripheral Interference” and says that overstimulation of touch interferes with coordinated movement (Compliant Technologies, 2024).

I’m skeptical that a stimulus like this is limited to sensory nerve fibers. Peripheral nerves contain axons with different diameters, functions, and activation thresholds. Many nerves close to the skin are mixed nerves, carrying sensory information toward the spinal cord alongside motor axons carrying commands to muscles. Electrical stimulation delivered through the skin can recruit motor nerves directly.

In fact, this is the basic principle underlying neuromuscular electrical stimulation, which has been used clinically for decades to evoke muscle contraction (Nussbaum et al., 2017). For example, functional electrical stimulators used therapeutically for foot drop (a neurological condition making it hard to lift the front of the foot while walking) deliver maximum pulse charges of about 24 μC for the NESS L300 Plus and 41.2 μC for WalkAide, at frequencies overlapping the G.L.O.V.E.’s approximately 30 pulses per second (U.S. Food and Drug Administration [FDA], 2015). At 72–112 μC per pulse, the glove exceeds the maximum pulse charge reported for these two representative foot-drop stimulators, although their waveforms, electrode geometry, and charge densities differ.

Add to this that a strong electrical stimulus can send an abrupt volley of afferent activity into the spinal cord. Noxious electrical stimulation is routinely used experimentally to evoke the nociceptive withdrawal reflex, a coordinated spinal response that activates muscles to pull a threatened limb away from the stimulus (Serrao et al., 2006; Linde et al., 2021).

Maybe you see the potential problem.

The glove delivers its pulses about thirty times each second. Repeated stimulation of motor axons could produce summation of muscle contraction, while repeated sensory volleys engage spinal and supraspinal responses. Add the startle reaction and the very powerful voluntary desire to get away from something that is actively hurting, and the resulting movement can become quite complicated.

A person being shocked may experience involuntary movements, and reflexively struggle to move away. Some of that movement may be voluntary. But some may arise from spinal reflexes, or from direct electrical recruitment of motor axons. All of it is movement, open to interpretation. And movement is one of the principal things an officer watches when deciding whether someone is resisting.

Imagine several officers trying to restrain one person. One officer has an activated G.L.O.V.E. around an exposed arm. The other officers are holding other parts of the body. The gloved officer establishes electrical contact and the person suddenly twists or pushes against the restraint.

The manufacturer states that the glove allows other officers to continue touching and handcuffing the person without receiving electrical stimulation themselves (Compliant Technologies, 2024, p. 12). That feature may create a concerning human-factors problem. The officer wearing the glove knows what has just happened. Another officer holding the suspect may feel only the sudden increase in movement. Will this movement be interpreted as a withdrawal reflex, or resistance?

The manufacturer has anticipated at least part of this problem. Its training manual advises having another officer present during deployment and instructs the operator to announce “G.L.O.V.E. ON!” so that fellow officers know stimulation is occurring (Compliant Technologies, 2024, p. 15). TASERs have similar protocols. Other officers need to know that their colleague has introduced an electrical stimulus into the nervous system of the person they are attempting to restrain.

A shouted warning like that might help, although real arrests are often noisy and confusing, and officers may arrive at the scene at different times, with differing interpretations of what they are seeing. Even when everyone hears “G.L.O.V.E. ON,” they still have to interpret what happens next.

This problem is already familiar enough with TASERs that some police departments have written it directly into policy. The Johns Hopkins Police Department warns that the electrical energy from a conducted-energy weapon “may result in involuntary muscle movement that may appear as voluntary resistance.” Its policy also instructs officers to give a person sufficient time to regain the ability to respond before treating failure to follow commands as grounds for another electrical application (Johns Hopkins Police Department, 2024).

If an officer perceives resistance and applies the glove, it could produce additional movement. Other officers might then interpret that movement as increasing resistance and apply greater physical force. The restrained person generates still more movement in response to the stimulation and restraint. Continued movement encourages additional electrical stimulation, or possible escalation beyond less-lethal means.

The stakes rise considerably if an involuntary movement looks threatening, and DHS policy allows deadly force when an officer reasonably believes someone poses an imminent threat of death or serious bodily injury (U.S. Department of Homeland Security, 2023). Officers sometimes have only fractions of a second to decide what a movement means. Electrical stimulation introduces another possible source for the movement they are trying to interpret.

In Ohio, David Nall was tased while officers ordered him to stop resisting and put his hands behind his back, even though the Sixth Circuit concluded that his apparent noncompliance could have been caused by TASER-induced involuntary convulsions (Goodwin v. City of Painesville, 2015). He later went into cardiac arrest and survived with severe cognitive impairment. And a decade later in Tennessee, Melissa Wooden was tased during a mental-health crisis and then shot twice after moving toward an officer. In Chrestman ex rel. Wooden v. Metropolitan Government of Nashville & Davidson County (2025), the Sixth Circuit found that the video evidence couldn’t establish whether that movement was intentional or an involuntary response to the TASER. Wooden survived after several surgeries and a two-month hospitalization. It’s clear that electrically-altered movement has the potential to be interpreted first as resistance and, in some circumstances, a threat.

Something else about the glove demo raises another question. At least some demos I reviewed were performed with the volunteer standing on thick foam mats, presumably to make the falls safer. But it may also partly electrically isolate the volunteer from the floor.

The intended G.L.O.V.E. circuit appears to be predominantly local between contacts on the glove, so insulation from the floor should have little effect on its normal stimulation. An actual arrest can create a very different electrical environment. A person might be face down on wet concrete, lying against a metal surface, soaked by rain, restrained at several points, or contacted by two G.L.O.V.E.s simultaneously. The manufacturer says that the device remains effective when the user or subject is wet, can work through soaked thin clothing, and permits as many as two gloves to be applied to one person (Compliant Technologies, 2024).

I’m speculating, but wet or conductive surfaces and multiple points of contact create possible current paths that are absent from a demonstration performed on a dry, insulating mat. Their importance depends on details of circuit isolation, leakage, capacitance and electrode geometry. This is an engineering question with an achievable, empirical answer.

The restraint of someone lying on the ground raises a more immediate neurophysiological concern. A standing volunteer can respond with a withdrawal reflex by moving the stimulated limb or falling away from the source. A person lying face down with immobilized arms and legs has fewer available movements. The nervous system can still respond to shock by muscle contractions. To the people restraining that person, it may even feel like resistance.

Once electrical stimulation begins, an officer observing a struggling person has to decide how much of the movement reflects intention and how much is an involuntary result being generated by the stimulated nervous system. This may be a more difficult inference to draw than it might first appear, and we are preparing to ask law enforcement officers to make this call in seconds, during an arrest.

Detailed specifications for the electrical output of the G.L.O.V.E. are available online, as are demonstrations showing its effects on volunteers. I couldn’t find an independent physiological study measuring EMG during application, any determination whether motor axons are directly recruited, or examining the device under realistic restraint conditions. Data like that seem worth having before the technology is deployed on the streets.

Before we add more uncertainty to thousands of encounters between law enforcement and the public, I think we should understand these devices better. And as a society, we should decide how much uncertainty we are willing to tolerate when the price of getting it wrong could end or forever alter a human life.

[Dwayne Godwin is a neuroscientist who researches PTSD, TBI, and alcohol use disorder. He is the co-author of “Out of Your Mind: The Biggest Mysteries of the Human Brain”. The views expressed here are his own and do not necessarily reflect those of the National Institutes of Health or his employer.]

For Further Exploration:

Anderson, M. (2026, August 12). ICE plans to outfit officers with electric shock gloves. NPR.

Axon Enterprise, Inc. (2022). Safety switch for a conducted electrical weapon (International Patent Publication No. WO2022251574A1).

Chrestman ex rel. Wooden v. Metropolitan Government of Nashville & Davidson County, 156 F.4th 694 (6th Cir. 2025).

Compliant Technologies LLC. (2024). GEN 4 & GEN 5 G.L.O.V.E. user manual (Version 1.4).

Goodwin v. City of Painesville, 781 F.3d 314 (6th Cir. 2015).

Jiang, Z., & Panescu, D. (2019). Electrical safety and performance of the G.L.O.V.E relative to relevant requirements of international electrical standards. In 2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) (pp. 6149–6154). IEEE. https://doi.org/10.1109/EMBC.2019.8857344

Johns Hopkins Police Department. (2024). Directive 405: Conducted energy weapon.

Kroll, M. W., Perkins, P. E., Pratt, H., Stuart, E., Bury, J., & Panescu, D. (2020). Safety of a high-efficiency electrical fence energizer. In 2020 42nd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC) (pp. 5016–5020). IEEE. https://doi.org/10.1109/EMBC44109.2020.9176351

Linde, L. D., Duarte, F. C. K., Esmaeili, H., Hamad, A., Masani, K., & Kumbhare, D. A. (2021). The nociceptive flexion reflex: A scoping review and proposed standardized methodology for acquisition in those affected by chronic pain. British Journal of Pain, 15(1), 102–113. https://doi.org/10.1177/2049463720913289

Lovan, D., & Foley, R. J. (2026, August 13). White House border czar defends ICE electric shock gloves as tools to avoid deadly force. Associated Press.

National Institute of Justice. (2011). Police use of force: The impact of less-lethal weapons and tactics. U.S. Department of Justice.

National Institute of Standards and Technology. (2018). Ampere: Introduction. U.S. Department of Commerce.

Nussbaum, E. L., Houghton, P., Anthony, J., Rennie, S., Shay, B. L., & Hoens, A. M. (2017). Neuromuscular electrical stimulation for treatment of muscle impairment: Critical review and recommendations for clinical practice. Physiotherapy Canada, 69(5), 1–76. https://doi.org/10.3138/ptc.2015-88

SABRE. (n.d.). Stun gun voltage vs. microcoulombs.

Serrao, M., Pierelli, F., Don, R., Ranavolo, A., Cacchio, A., Currà, A., Sandrini, G., Frascarelli, M., & Santilli, V. (2006). Kinematic and electromyographic study of the nociceptive withdrawal reflex in the upper limbs during rest and movement. Journal of Neuroscience, 26(13), 3505–3513. https://doi.org/10.1523/JNEUROSCI.5160-05.2006

Stålberg, E., van Dijk, H., Falck, B., Kimura, J., Neuwirth, C., Pitt, M., Podnar, S., Rubin, D. I., Rutkove, S., Sanders, D. B., Sonoo, M., Tankisi, H., & Zwarts, M. (2019). Standards for quantification of EMG and neurography. Clinical Neurophysiology, 130(9), 1688–1729. https://doi.org/10.1016/j.clinph.2019.05.008

U.S. Department of Homeland Security. (2023). Department policy on the use of force.

U.S. Department of Homeland Security. (2026). Acquisition Planning Forecast System: F2026074524, CTG-5 G.L.O.V.E.

U.S. Food and Drug Administration. (2015). MyGait Stimulation System: 510(k) summary (K141812). Center for Devices and Radiological Health.

© 2026 Dwayne W. Godwin.

– Dylan Thomas

It’s a normal human impulse to want to win at the game of life. It’s also understandable that people want to keep playing. Entire industries have sprung up around supplements and extending the aging clock, or even turning it back.

Aging is partly etched in our genes, and some of the experimental gains to control aging by manipulating them are impressive, especially if you are a mouse. Gene-editing tools such as CRISPR are rapidly improving our ability to manipulate genes involved in cellular aging and repair. It’s easy to imagine a strategy that suppresses the processes that cause our decline, while promoting those that repair cellular damage, giving us a greatly extended life. Researchers have prolonged life or reduced age-related disease in animals by manipulating genes, clearing senescent (“worn out”) cells, and partially resetting some of the epigenetic markers associated with aging.

Human cells in culture have also been pushed toward younger states in the laboratory. Basically, we are getting better at “Benjamin Button-ing” cells, though turning that into a safe treatment for an aging person is another matter.

Despite these advances, extending our lives far beyond the current human record remains beyond the immediate capabilities of medical science. Moreover, science may somehow stave off the natural aging process, but disease, trauma, and other killers would remain. Death might be delayed without ultimately being denied.

What would it mean to continue living, in some form, after our last neuron has fired? Science fiction loves this idea. Usually it involves transferring our essence into electronic form, typically followed by installation in a realistic android body. Some believe this sort of digital immortality will eventually be possible, encouraged by neuroscience advances in understanding brain organization, increasing computing power, leaps in artificial intelligence, and massive data storage.

In Greg Egan’s 1990 short story “Learning to Be Me,” a child grows up with a small jewel implanted inside his skull. The device receives the same sensory and physiological information as his brain and is continually adjusted whenever its responses diverge from his own. By adulthood, the jewel “knows” him well enough to take over.

The technical barriers to this concept are becoming easier to describe, even as the prospect of uploading a human mind remains far out of reach. Researchers can identify some of the necessary steps: reconstruct the connections, record the activity moving through them, build a computational model constrained by both, and test whether that model responds as the biological system did. Whether those steps would preserve the person whose brain supplied the data is a different question.

How would you actually upload your consciousness into a computer?

Step 1: Use the Right Hardware

Uploading a brain will require more than your laptop. Estimates of the storage capacity required to hold the contents of a human brain range from hundreds to thousands of terabytes, depending on hazy assumptions about what it takes to store a memory. Let’s simply assume it’s a large number. Storage itself is unlikely to be the most serious limitation, since distributed systems can already hold immense quantities of information and their capacity will only continue to grow.

The computer needed to “hold” a human brain and simulate its functions might be engineered. Neuromorphic chips are one promising platform because they allow computations that mimic some of the operating principles of neurons. In silico mimicry of biological neurons may offer a route toward a neurocomputing system that is modular, expandable, and capable of plasticity.

Perhaps we could grow such a platform. After all, it’s hard to imagine a better repository for human consciousness than a human brain, possibly one grown from your own cells. Scientists can already produce cerebral organoids from stem cells, creating small brain-like tissues containing several cortical neuron types. These structures are developmental models, far removed from complete brains, but it is reasonable to imagine increasingly complex biological or hybrid systems.

Biological solutions would need many of the same support systems as our own brains, including circulation, metabolism, immune protection, and regulation of the chemical environment (perhaps even a cloned body to support it). Growing a brain-like organ, though, also raises ethical questions that become increasingly serious as the tissue becomes more organized, responsive, and capable. Imagine how hard it is to know what’s going on in other human beings who can communicate their thoughts and signal when they are in pain or discomfort. A brain in a vat would lack this basic capability, so if it somehow gained some vestige of consciousness or pain perception, would it be ethical to do it?

Predicting the right platform is a bit like traveling back to the 1970s and deciding whether Betamax or VHS will become the video medium of the future. VHS won that battle, only to give way to digital media. So the best we can do is identify the properties required for success, knowing that an unexpected technological advance may radically change how the problem is approached.

Whatever platform is chosen will need to operate according to the same basic rules as real brains, including some analog of neuroplasticity. Without that capacity, we would be missing an essential part of what makes an individual brain feel like itself.

Step 2: Reconstruct the Connections

Some representation of the brain’s connections may be needed on which to “run” the mind. We’d therefore need a way to determine the relevant, or effective, synaptic connectome of an individual.

Sebastian Seung has compared the connectome to the bed of a stream. The riverbed determines the flow of the water, while the water gradually erodes and reshapes the bed. It’s a nifty metaphor for the plastic changes that continually rewire our brains.

Current brain imaging can estimate functional connections, but its resolution is limited to volumes containing the aggregate activity of many thousands of neurons. Each neuron may make thousands of connections with other cells, so the fine details that contribute to making you “you” disappear into the average.

We’ve come a long way from treating this idea solely as the realm of science fiction. In 2024, the FlyWire Consortium published the first complete wiring diagram of an adult fruit-fly brain, tracing 139,255 neurons and 54.5 million synapses. The accomplishment is remarkable, especially considering that the brain belongs to an animal capable of some of the things humans do: navigating, finding food, avoiding danger, courting, making decisions, and learning. Researchers can now follow synaptic pathways across an entire adult brain instead of reconstructing a small circuit in isolation. But it’s a fly, and 139,255 neurons is a far cry from the 86 billion neurons of the human brain.

Even a complete connectome only tells us where signals might travel, rather than what the living system is doing. It doesn’t fully reveal how strongly one neuron affects another, how those effects change with experience or internal state, or how the network gives rise to behavior. The fly connectome is an unusually complete road map, but without the details that make it functional.

A project in mouse visual cortex has begun connecting the map with the traffic. The MICrONS Consortium recorded activity from about 75,000 neurons while an awake mouse viewed visual stimuli, then aligned those recordings with an electron-microscopy reconstruction containing more than 200,000 cells and roughly half a billion synapses. The result attempts to link what neurons did while the animal was alive with the connections found after the tissue was preserved. Even this extraordinary dataset represents only a small part of the visual system of a single mouse. It brings us closer to understanding how a wiring diagram becomes a functioning circuit while also revealing the distance between reconstructing a circuit and reconstructing a person.

Anatomical methods such as electron microscopy can resolve individual synapses, though the process generally requires that the brain be dead, infused with preservatives, sliced into extremely thin sheets, and imaged piece by piece. Not fun. Even then, a picture of the connections would leave out much of what determines how those connections behave.

Synapses are worlds unto themselves. Thousands of proteins populate them and modify how information is transmitted. These proteins can assume multiple configurations, move within the cell, form complexes with other proteins, and change their behavior in response to patterns of activity. Synaptic terminals function differently depending on what has just happened to them, sometimes accelerating and sometimes putting the brakes on neurotransmitter release. Add to this uncertainty astrocytes and other glial cells, which constitute a large volume of brain tissue and participate in synaptic regulation, network dynamics, and perhaps even the cellular machinery of memory.

The resulting system contains a hum of continuous activity beneath, between, and sometimes independent of the electrical spikes we commonly associate with the nervous system. This molecular and biochemical activity undergirds those spikes like the submerged part of an iceberg.

Computational neuroscientists often boil the function of a synapse down to a connection weight and whether it exerts an excitatory or inhibitory effect. This simplifying assumption allows them to focus on the other immensely complex interactions across brain networks. But it may also shortchange the local complexity of the synapse and get the flow of information through some circuits wrong.

As Alfred Korzybski famously observed, “the map is not the territory.” A map may get us from one city to another while leaving out potholes, construction zones, and that great diner where we stopped once along the way. We should be no more surprised when a computational model of the brain leaves out something important.

Step 3: Capture the Brain as a Dynamic System

Dynamic and highly nonlinear relationships can be detected in the electrical activity of the brain, but they remain invisible even to high-resolution electron microscopy. Just as taking a picture of a Ferrari doesn’t provide information about what it would be like to punch the gas pedal, an anatomical image of a synapse would not necessarily capture cognition.

Imagine that you were an alien who had never seen a bird. Presented with a still photograph of an old, dead robin lying on the ground, you might never conclude that it had once possessed the capacity for flight. A preserved brain, however exquisitely imaged, would also be like that dead bird: a still photograph of a once dynamic system.

When viewing brain tissue prepared for electron microscopy, one feels the static nature of the images. Properly done, they appear to be permanent monuments to a particular preserved brain. A purely anatomical connectome would preserve one moment of an individual’s life, most likely the last one. In a dynamic system, it might be nearly impossible to estimate the state of the network in enough detail to determine how the system should function when reinitialized.

We tend to think of our brains as rather static objects. I blame the ubiquitous “brain as a computer” metaphor for this, because if you think of computers, you think of the innards as a motherboard and power supply. The logic gates in a chip don’t move. So surely, nothing is moving around in our head?

Well actually, during development projections span significant intracranial distances. Even as adults, dendrites change their shape in response to inputs. Synapses strengthen and weaken, proteins are modified and replaced, and neuromodulators alter the responsiveness of whole networks. The same anatomical circuit may behave very differently depending on factors like sleep, stress, hunger, intoxication, attention, hormones, and even its own recent history of activity.

An electron micrograph is therefore a bit like a really good still frame from a movie, at best a moment frozen in time. So in addition to the wiring diagram, we’d need to capture the brain in action to obtain the rest of the movie. So far, we have limited ways to do this at the resolution required to understand which parts of your unique connectome become active when you think of your lover, or Benjamin Franklin, or being chased by a bear. Current functional magnetic resonance imaging and magnetoencephalography methods are probably insufficient. Future improvements in these technologies, or entirely new methods, might record and read out brain activity at several relevant scales.

That’s an enormous problem, but let’s pretend that we solve it. Once the activity is captured, a computational framework would still be needed to run the simulation. Here, we might count on some efficiency. It may be possible to simplify complex synaptic functions in much the same way that digital compression preserves the important information in an image while reducing its size. Computational shortcuts might approximate the effects of ion channels, neurotransmitter release, neuromodulation, and brain rhythms without needing to reproduce every molecular interaction.

Of course, we don’t want to discard the complexity altogether. That would be like bulldozing the city and starting from scratch because the street map had become inconvenient. The difficulty is in knowing which details can be safely omitted, and right now we have no idea.

Another shortcut might involve focusing on the similarities among brains, or among regions within the same brain. Cortical areas share broad organizational principles. Their inputs, outputs, and histories of plasticity help determine whether a particular patch of cortex becomes visual, tactile, auditory, or something else.

This might mean that we do not need a unique wiring diagram for every person. Perhaps we need one very good, highly plastic wiring diagram that can be trained by the dynamic patterns of activity distinguishing one personality from another. This “protobrain” might be shaped by experience, carving out the riverbed that defines the characteristic flow of information and thought for a particular person.

But even a complete record of the brain in action may be insufficient to capture the uniqueness of consciousness. Discussions about uploading brains often leave out another essential requirement: some continuity in the experience of being the person whose brain has supposedly been transferred. This may be the biggest hurdle of all.

Step 4: Preserve the Feeling of Being You

If we wish to realize that we made it across the digital divide of death, rather than finding a befuddled clone sitting on the other side, some continuity of consciousness should be the desired result. This may seem straightforward, because when you go to sleep and awaken, you still have the sense that you are you.

People awakening from comas generally remember who they are, though some are shocked to discover how much time has passed and how much they have aged. Frozen wood frogs that hibernate through the winter presumably don’t forget that they are frogs. Researchers have even frozen and thawed worms in ways that preserve evidence of simple learned behavior. Are death, preservation, uploading, and reanimation simply a more complicated version of this kind of interruption?

Sleep, coma, and hibernation occur within a continuing organism. The same brain remains in place, maintaining some degree of metabolic, structural, and causal continuity. Uploading proposes that activity begins in another physical system, and that key difference brings us back to the question of what consciousness actually is.

In an earlier essay, I described consciousness simply as “the feeling of being you.” It is the feeling of inhabiting a particular body, receiving its sensations, storing its memories, and occupying its position in the world.

The systems supporting these feelings are partly understood. Interoception tracks the state of the organs and viscera, while proprioception maps the position and movement of the body. Emotional and motivational circuits assign value to those signals, marking some states as safe, threatening, pleasurable, or urgent.

Antonio Damasio describes this architecture as a hierarchy. At its base is the proto-self, the brain’s continuous, largely nonconscious map of the body’s condition. From this emerges core consciousness, the moment-to-moment feeling of being a subject having an experience. The autobiographical self adds the remembered past and an anticipated future.

Autobiographical memory matters, though it may not be what makes the story feel as if it belongs to you. A person can lose portions of autobiographical memory, while continuing to possess an immediate sense of self. Henry Molaison, the famous memory patient H.M., had profound difficulty forming new declarative memories while still retaining a sense of self bound to his immediate present.

People with severe aphasia may lose the ability to translate their thoughts into words while remaining conscious of their predicament. Language, memory, and consciousness overlap without being identical.

This distinction has become increasingly important in the age of large language models. An LLM can maintain a conversational persona, and produce moving descriptions of experiences it has never had. It learned these patterns from the language of embodied, conscious people. The result can sound like an autobiography even though there is no evidence of a proto-self, a body to regulate, or a core subject to whom the story belongs.

A reconstructed person would be vastly more sophisticated. It could contain your memories, preferences, habits, and characteristic patterns of thought. Its declaration that “I am still here” would demonstrate preservation of the information required to tell your story. Whether the feeling of being you had also survived would remain uncertain. To an outside observer, the uploaded person might seem the same as the old version. Philosophers have worried about variants of this problem for decades, most famously in Derek Parfit’s thought experiments about teletransportation and divided selves.

Imagine that scientists create a perfect reconstruction while you are still alive. The electronic brain awakens and remembers entering the laboratory. It recognizes your family and colleagues, recalls agreeing to the procedure, and remembers wondering what the transition would feel like. When asked whether the upload worked, it says yes, and it isn’t lying. You, meanwhile, are still lying in the scanner, still in the old body.

Both versions would feel that they are the original. Each has the same memories up to the moment of duplication, though from that point forward their experiences diverge and they become separate conscious agents.

If your biological body died the next day, the digital version would continue. It may finish your work, care about the people you cared about, and mourn the loss of what it regards as its original body, and feel grateful that it “cheated” death. Your family may experience this as your survival, and the digital person will almost certainly experience it that way. But the consciousness generated by your biological brain will still wink out.

You may have provided immortality to a digital clone who becomes, in effect, a separate person. Similarity between brains does not establish shared identity. Identical twins, after all, don’t believe themselves to be each other.

Greg Egan entertained this problem in “Learning to Be Me.” In Egan’s imagined future, children receive a device called a jewel while they are still young. The jewel monitors the activity of the brain and receives the same sensory and physiological information. A teaching system continually adjusts it whenever its responses diverge from those of the biological brain. It spends years learning to reproduce the person who carries it, and eventually it is meant to take control.

Egan’s premise is more interesting than a conventional brain scan because the jewel shares the person’s life. It has seen what they have seen, received signals from the same body, and spent decades learning to be them. It’s essentially the proto-brain to which I alluded.

The unsettling question in Egan’s story is whether that history makes the jewel the same person or produces a second consciousness trained to believe it is the same person. Modern neuroscience suggests an alternative: during those years, the person might also learn that the jewel is part of them.

Sometimes a person who loses an arm or leg continues to feel that the limb is present. The missing limb may itch, move, or become painfully clenched. The brain’s representation of the missing limb persists as a haunting phantom of what was lost.

V. S. Ramachandran famously used a mirror to make the reflected image of an intact hand appear in the position of a missing one. By moving the intact hand and watching its reflection, some patients could experience the phantom hand opening and relaxing. It was as if the brain needed a place to put the intended movement, or a way to reconcile the command with an image of a body capable of carrying it out.

A different illusion can persuade the brain that a rubber hand belongs to the body. The person’s real hand is hidden behind a screen while a rubber hand is placed in view. When the real and artificial hands are stroked in the same places at the same time, something very interesting happens. Over the course of a few minutes, many people begin to experience the rubber hand as part of themselves.

Touch delivered to the hidden hand seems to occur at the rubber one. Asked to point to the location of their hand, some participants drift toward the artificial hand they can see, the one that has become bound to the sensation. Some react with alarm when the rubber hand is struck with a hammer, so tangible is the illusion that the rubber hand is part of them.

The living brain remains the source of the experience, while its definition of the body to which that consciousness belongs begins to change. If consciousness is “the feeling of being you,” and the brain can renegotiate which objects and signals count as parts of “you,” perhaps the substrate supporting consciousness is also somewhat negotiable.

In monkeys, pairing touches to the real arm with visible touches to a virtual arm can alter the responses of neurons in primary somatosensory and motor cortex. After training, some of these neurons respond when the virtual arm alone is touched. The avatar has tapped into neural systems ordinarily devoted to representing the biological body.

Experiments in humans suggest that body representation can expand as well as substitute. Under carefully controlled conditions, people can experience an artificial limb as a third arm while retaining ownership of both biological arms. The brain can, at least temporarily, enlarge the category of things included within the concept of “me.”

Neuroprosthetics provide a practical version of the same principle. Artificial limbs feel more like parts of the body when users control them intuitively and receive appropriately timed sensory feedback. With long-term use, the relationship between person and device can change, so that what begins as a tool becomes incorporated into the continuous loop connecting intention, movement, sensation, and correction. Some neuroengineers describe this as co-adaptation. This reciprocal process, in which biological and artificial systems achieve a sort of symbiosis, suggests another path across Immortality Valley.

Suppose an artificial memory system is connected to the living brain. At first, the person consults it deliberately, much as one might consult a notebook. With repeated use, the interface becomes faster and less noticeable. The system begins to anticipate what information will be needed and supplies it directly, while some memories become distributed across biological and artificial circuits.

Eventually, the person may experience the process simply as remembering. The same progression might occur with vision, language, movement, other cognitive functions, and internal sensations. An artificial system might first assist a function, then share it, and eventually assume most of it. Each change would occur while the person remained conscious and able to learn what the new system did and how its activity felt.

In such a process, the machinery included within the word “me” might gradually change. Body ownership and the ownership of experience may not be identical. The brain can revise which limb or device feels like it belongs to you, while the same subject continues to have the experience. Whether the subject itself could migrate with a gradually changing body remains the part of the experiment that is, so far, out of reach.

Egan’s jewel learns to be its person, while a more complete system would involve reciprocal learning. The jewel would learn the brain, and the brain would learn to treat the jewel as part of itself. If such a jewel learned “the feeling of being you” gradually, providing a kind of continuity of consciousness, then Immortality Valley might be bridged.

The idea of such gradual replacement isn’t new. Hans Moravec imagined neuron-by-neuron “transmigration” decades ago. What newer embodiment research adds is a possible mechanism for making that transition subjective as well as functional: the artificial component might become part of “me” before the biological component disappears.

Whether this process could continue until the original brain had been completely replaced is unknown. At some point, the Ship of Theseus problem becomes a problem for the nervous system itself. If every biological part has been replaced gradually, is the remaining system the same conscious person, or a successor who never noticed when the original disappeared?

Not everyone thinks gradual replacement buys us anything. Others have argued that replacing a brain piece by piece is, with respect to personal identity, metaphysically equivalent to scanning and copying it. That objection might be correct if gradualism mattered only because it avoided a dramatic moment of replacement. The embodiment literature suggests that gradual incorporation may allow the living nervous system to change what it experiences as part of itself while the transition is occurring.

Such gradual incorporation differs meaningfully from copying. A simple copy of you would awaken with your memories in a body it has never inhabited. A gradually extended version of you would learn the new substrate while still alive, while the person remained conscious through each stage of the negotiation. This may be the most plausible route by which the feeling of being you could cross from one kind of machinery into another.

Slide, Don’t Jump

Is the vision of an immortal afterlife, in which we can expand and hack our own consciousness, achievable? Skepticism remains appropriate, but many of the technical challenges are definable and therefore may be solvable in principle, given enough time and the solution of a million engineering challenges, small and large. Success would require a much deeper understanding of the brain at all its levels of operation, including the computationally inconvenient ones. It would also require agreement about what outcomes would count as success.

A digital reconstruction might think, speak, and remember exactly as you do. It might be conscious in its own right and possess an unshakable conviction that it is you. That would be a remarkable scientific achievement, though it would leave open the question of whether the person who entered the procedure had actually achieved this sort of immortality.

Gradual incorporation would suggest that an artificial system might become part of the living brain’s body model, participate in its cognitive functions, and share its ongoing history. Then, the gradually extended person could remain conscious while learning the new substrate, allowing the existing self to participate in each stage of the negotiation.

The tiny chance that this sort of process could work should probably have little effect on one’s current quality of life. We’re still much closer to biological interventions that may extend healthy life in the bodies we have, than we are to transferring a conscious self into another substrate. Still, Egan’s jewel gives us a useful way to think about the problem.

A familiar childhood prayer imagines that death might arrive during sleep:

If I should die before I wake,
I pray the Lord my soul to take.

Digital immortality proposes that something could still awaken with our memories, recognize our world, and stumble through the lives we left behind. The enduring question is whether we would really be there in the morning.

If you’re interest in this topic, my friend Jorge Cham and I discuss what it means to upload brains on his podcast, Science Stuff. We also wrote a very accessible book that touches on this and other themes.

[Dwayne Godwin is a neuroscientist who researches PTSD, TBI, and alcohol use disorder. He is the co-author of “Out of Your Mind: The Biggest Mysteries of the Human Brain”. The views expressed here are his own and do not necessarily reflect those of the National Institutes of Health or his employer.]

For Further Exploration

Azevedo, F. A. C., Carvalho, L. R. B., Grinberg, L. T., Farfel, J. M., Ferretti, R. E. L., Leite, R. E. P., Jacob Filho, W., Lent, R., & Herculano-Houzel, S. (2009). Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain. Journal of Comparative Neurology, 513(5), 532-541. https://doi.org/10.1002/cne.21974

Botvinick, M., & Cohen, J. (1998). Rubber hands “feel” touch that eyes see. Nature, 391, 756. https://doi.org/10.1038/35784

Cappuccio, M. L. (2017). Mind-upload: The ultimate challenge to the embodied mind theory. Phenomenology and the Cognitive Sciences, 16(3), 425-448.

Chalmers, D. J. (1995). Absent qualia, fading qualia, dancing qualia. In T. Metzinger (Ed.), Conscious experience (pp. 309-328). Ferdinand Schoningh.

Corkin, S. (2002). What’s new with the amnesic patient H.M.? Nature Reviews Neuroscience, 3, 153-160. https://doi.org/10.1038/nrn726

Damasio, A. R. (1999). The feeling of what happens: Body and emotion in the making of consciousness. Harcourt Brace.

Damasio, A. R. (2010). Self comes to mind: Constructing the conscious brain. Pantheon Books.

Dorkenwald, S., Matsliah, A., Sterling, A. R., Schlegel, P., Yu, S.-C., McKellar, C. E., Lin, A., Costa, M., Eichler, K., Yin, Y., Silversmith, W., Schneider-Mizell, C., Jordan, C. S., Brittain, D., Halageri, A., Kuehner, K., Ogedengbe, O., Morey, R., Gager, J., et al. (2024). Neuronal wiring diagram of an adult brain. Nature, 634, 124-138. https://doi.org/10.1038/s41586-024-07558-y

Egan, G. (1990, July). Learning to be me. Interzone, 37.

Guterstam, A., Petkova, V. I., & Ehrsson, H. H. (2011). The illusion of owning a third arm. PLOS ONE, 6(2), e17208. https://doi.org/10.1371/journal.pone.0017208

Lancaster, M. A., Renner, M., Martin, C.-A., Wenzel, D., Bicknell, L. S., Hurles, M. E., Homfray, T., Penninger, J. M., Jackson, A. P., & Knoblich, J. A. (2013). Cerebral organoids model human brain development and microcephaly. Nature, 501, 373-379. https://doi.org/10.1038/nature12517

Liang, C., Lin, W.-H., Chang, T.-Y., Chen, C.-H., Wu, C.-W., Chen, W.-Y., Huang, H.-C., & Lee, Y.-T. (2021). Experiential ownership and body ownership are different phenomena. Scientific Reports, 11, 10602. https://doi.org/10.1038/s41598-021-90014-y

Lopez-Otin, C., Blasco, M. A., Partridge, L., Serrano, M., & Kroemer, G. (2023). Hallmarks of aging: An expanding universe. Cell, 186(2), 243-278. https://pubmed.ncbi.nlm.nih.gov/36599349/

MICrONS Consortium. (2025). Functional connectomics spanning multiple areas of mouse visual cortex. Nature, 640, 435-447. https://doi.org/10.1038/s41586-025-08790-w

Moravec, H. (1986). Dualism from reductionism. In Proceedings of the International Symposium on AI and the Human Mind (Vol. 2).

Olova, N., Simpson, D. J., Marioni, R. E., & Chandra, T. (2019). Partial reprogramming induces a steady decline in epigenetic age before loss of somatic identity. Aging Cell, 18(1), e12877. https://doi.org/10.1111/acel.12877

Parfit, D. (1984). Reasons and persons. Oxford University Press.

Pigliucci, M. (2014). Mind uploading: A philosophical counter-analysis. In R. Blackford & D. Broderick (Eds.), Intelligence unbound: The future of uploaded and machine minds (pp. 119-130). Wiley-Blackwell.

Ramachandran, V. S., Rogers-Ramachandran, D., & Cobb, S. (1995). Touching the phantom limb. Nature, 377, 489-490. https://doi.org/10.1038/377489a0

Sanchez, J. C., Mahmoudi, B., DiGiovanna, J., & Principe, J. C. (2009). Exploiting co-adaptation for the design of symbiotic neuroprosthetic assistants. Neural Networks, 22(3), 305-315. https://pubmed.ncbi.nlm.nih.gov/19403263/

Sandberg, A., & Bostrom, N. (2008). Whole brain emulation: A roadmap. Future of Humanity Institute, University of Oxford.

Schofield, J. S., Shell, C. E., Beckler, D. T., Thumser, Z. C., & Marasco, P. D. (2020). Long-term home-use of sensory-motor-integrated bidirectional bionic prosthetic arms promotes functional, perceptual, and cognitive changes. Frontiers in Neuroscience, 14, 120. https://doi.org/10.3389/fnins.2020.00120

Seung, H. S. (2012). Connectome: How the brain’s wiring makes us who we are. Houghton Mifflin Harcourt.

Shokur, S., O’Doherty, J. E., Winans, J. A., Bleuler, H., Lebedev, M. A., & Nicolelis, M. A. L. (2013). Expanding the primate body schema in sensorimotor cortex by virtual touches of an avatar. Proceedings of the National Academy of Sciences of the United States of America, 110(37), 15121-15126. https://doi.org/10.1073/pnas.1308459110

Vita-More, N., & Barranco, D. (2015). Persistence of long-term memory in vitrified and revived Caenorhabditis elegans. Rejuvenation Research, 18(5), 458-463. https://pmc.ncbi.nlm.nih.gov/articles/PMC4620520/

Wiley, K. B., & Koene, R. A. (2016). The fallacy of favoring gradual replacement mind uploading over scan-and-copy. Journal of Consciousness Studies, 23(3-4), 212-235.

Zanichelli, N., Schons, M., Freeman, I., Shiu, P., & Arkhipov, A. (2025). State of brain emulation report 2025. arXiv.

© Copyright 2026 Dwayne Godwin. This essay substantially revises and expands my 2015 BrainFacts.org essay, “Heads in the Cloud.”

I’m glad Helen Keller isn’t aware that there are people who doubt she ever accomplished the things history records. She spent enough of her life explaining herself. I don’t need to defend her here. Her life is its own monument to her accomplishments.

She was born in Tuscumbia, Alabama, in 1880. At nineteen months of age, an acute illness (hypothesized to be meningitis or scarlet fever) left her profoundly deaf and blind. For several years, she lived without formal language before Anne Sullivan arrived at the Keller home in March 1887. The breakthrough at the familiar water pump followed soon afterward, when Keller grasped that the patterns Sullivan spelled into her hand symbolized the water flowing over her other hand.

From there, her life became one of the most thoroughly documented educational journeys in history. She attended Perkins Institution for the Blind, the Wright-Humason School for the Deaf, the Cambridge School for Young Ladies, and eventually Radcliffe College, graduating cum laude in 1904. She published fourteen books and hundreds of essays.

While born into a white Southern family whose property and social position had been built partly through slavery, as an adult, she rejected the racial order she inherited and publicly supported the NAACP, opposed lynching and segregation, and condemned the continuing economic and political subordination of Black Americans. She was also an outspoken and persistent advocate for women’s suffrage and labor reform. She corresponded with presidents, scientists, writers, and political leaders.

She appeared in newsreels, documentaries, and countless photographs. Her communication through tactile fingerspelling survives on film for anyone to see. Keller lived in the age of photography, newspapers, and mass public appearances. Her life left an unassailable documentary trail. Yet disbelief persists, out there among the trolls, flat earthers, and Apollo moon landing deniers.

It’s natural to assume that another person’s mind is organized like our own until evidence forces us to reconsider. People with vivid visual imagery often struggle to imagine people who can’t form mental images (aphantasia). Conversely, people with aphantasia may doubt that others truly “see” images in their minds. Helen Keller conspiracists possess a similar cognitive bias on a larger scale. People imagine their adult selves suddenly deprived of sight and hearing, then try to picture the silence and darkness that would follow. They can’t imagine living as Helen Keller lived, so they conclude that Helen Keller could not have lived that way, either.

We tend to simulate her experience through our own adult frame of reference, instead of asking how her brain might have developed differently. An adult who abruptly loses vision and hearing carries decades of neural organization built around those senses. The brain’s wiring is so established that it’s nearly impossible to imagine what it would be without them at all. Helen Keller lost them before her second birthday, while her cerebral cortex was still assembling itself. The architecture of her mind developed under entirely different constraints than for those who are sighted and able to hear. That’s a key distinction.

The historical evidence of her life is substantial. She generated school records, college examinations, letters, manuscripts, interviews, films, newspaper accounts, and thousands of independent observations extending over more than eighty years. People who had never met Anne Sullivan marveled at Keller’s intellectual range. Her political writings often frustrated some who preferred a more curated public image, which hardly resembles the output of a carefully managed puppet.

Keller’s abilities were repeatedly tested under conditions in which her teacher, Anne Sullivan, could transmit neither the questions nor the answers freely. At Radcliffe, independent proctors presented the material, and Keller typed her own responses. She also communicated through multiple partners and continued her intellectual work for more than thirty years after Sullivan’s death.

The language of her books has also attracted skeptics. How could a deafblind woman write about sunsets, music, birds, or color? Doesn’t one need to “see” those things?

Keller’s use of language reveals something more interesting than her occasional use of visual verbs. Throughout The World I Live In, she repeatedly returns to the theme of touch. Her remaining modalities become recurring themes as she explains how the external world entered her mind through touch, vibration, and smell. Objects emerge through successive exploration rather than a single glance. Places acquire their identity through accumulated tactile and spatial experience. Even abstract ideas are often grounded in physical contact with the world. The frequency with which she returns to these tactile descriptions reflects her conscious awareness of the sensory pathways through which her understanding had been constructed.

“Every object is associated in my mind with tactual qualities which, combined in countless ways, give me a sense of power, of beauty, or of incongruity.” – Helen Keller, The World I Live In.

Familiar scents identified places and people. Rather than limiting her understanding, these sensory pathways provided sufficient information for the developing brain to construct stable representations of objects, space, language, and eventually abstract ideas. Modern readers who dismiss Keller’s achievements often underestimate how much information is available through these channels, particularly when they become the principal sources of experience during early brain development.

How could a brain, deprived of both vision and hearing, construct such a rich internal world? When Keller described exploring a horse, she explained that her hands encountered only separate parts: the muzzle, the mane, the shoulder, the flank. Her fingers never touched the entire animal at once. Yet the completed horse existed in her mind as a unified whole. “My fingers feel the parts,” she wrote, “and my mind puts them together.”

Research over the past several decades has transformed our understanding of sensory cortices we use to assemble the world. During the 1990s, brain imaging studies showed that congenitally blind people reading Braille activated primary visual cortex. At first, many investigators wondered whether the activation was merely incidental. In a clever experiment, researchers applied transcranial magnetic stimulation (TMS is a powerful electromagnet that, when applied to the head, can locally disrupt brain activity) over occipital cortex while blind participants read Braille. The stimulation disrupted their ability to read. It seems the “visual” cortex had become functionally necessary for tactile language.

In some expert Braille readers, stimulating occipital cortex produced tactile sensations in the fingers rather than flashes of light. Deprived occipital cortex had in effect joined a different functional network.

Parallel discoveries emerged from studies of deaf individuals. Regions normally associated with auditory processing became engaged during visual motion, sign language, and other tasks. Across both literatures, the emerging picture grew steadily clearer. The cortex is shaped by developmental constraints, while its mature role also depends upon the information it receives and the computations it performs.

Animal experiments by Mriganka Sur’s laboratory uncovered some of the potential mechanisms underlying this reorganization. Sur and his colleagues rerouted retinal projections into the auditory thalamus of newborn ferrets. Visual information therefore arrived in auditory cortex. As the animals matured, those cortical regions developed visual receptive fields, maps of visual space, orientation selectivity, and the ability to guide visually-based behavior. The rewired cortex retained traces of its original architecture, but also acquired properties normally associated with vision.

While they were not shown to possess normal object vision through the rewired pathway, Sur’s rewired ferrets demonstrated behaviorally useful visual perception through auditory cortex, including light detection, visual–auditory categorization, and spatial-frequency discrimination. The experiment revealed that early cortical identity emerges through a dialogue between inherited structure and patterned experience.

Helen Keller’s own descriptions were a harbinger of this neuroscience in an unexpected way. She portrayed her inner world as coherent spatial forms assembled from successive exploration. Her fingers acquired local information: curvature, texture, resistance, temperature, movement. She integrated those fragments into complete representations that could later be recalled, manipulated, and imagined.

Keller’s descriptions have a modern technological analogue. When 3-D scanner apps on a mobile device scan a three-dimensional object, it doesn’t capture the entire form at once. As the camera moves, it collects thousands of local measurements and continuously updates an internal model until a complete object emerges. Keller’s description is similar, except instead of an image, her fingers explored one region after another while her brain preserved each tactile impression, related it to those that came before, and gradually knit together a coherent representation of the whole. “I feel the parts,” she wrote, “and my mind puts them together.” It’s only an analogy, but it captures an important idea that her perception constructed a model of the world from successive observations.

Words then supplied relationships extending beyond the reach of her hands: mountains she had never climbed, or stars she could never reach. Those descriptions entered a brain already grounded in touch, movement, smell, vibration, and embodied experience, allowing concepts to grow beyond immediate sensation. Was something like a mountain represented within her mind in the same way it is in your mind? It’s doubtful. But the representation was no less tangible to her.

Keller sometimes described this process in her own terms, tracing a different route toward the same understanding. Her accomplishments arose from a rare convergence of intelligence, sustained effort, inspired teaching, and opportunity. Her life also revealed something fundamental about the developing brain: its capacity to construct a coherent world from whatever channels of experience remain open.

Helen Keller skeptics ask us to believe that generations of teachers, classmates, editors, physicians, journalists, filmmakers, and readers collectively sustained an elaborate deception. And that luminaries like Alexander Graham Bell, Mark Twain, Albert Einstein, Andrew Carnegie, Eleanor Roosevelt, Henry Ford, and many other leading intellectual and public figures of her time were either all in on it, or duped. Instead, they knew her as a correspondent, conversationalist, writer, and friend.

The truth of Helen Keller is more interesting, and I think, more human. A young brain, confronted with an altered sensory world, reorganized itself around the information still available. Cortical regions may have been recruited for new purposes: touch acquired the structure of language and opened a channel for conversation. The sequential exploration of objects was the way she assembled her world into spatial understanding. The symbolism of words then carried thought beyond the boundaries of her immediate experience.

An ancient parable from India tells of several blind men encountering an elephant. One feels the trunk and concludes it is a snake. Another grasps a leg and insists it is a tree. A third examines the tusk and imagines a spear, and the last detects, in feeling the ear, a fan. The story has long served as a warning about drawing sweeping conclusions from limited experience. Helen Keller’s life suggests a complementary lesson. The blind men of the parable stop exploring, but Helen Keller didn’t.

She once met an elephant herself. During a visit to the Bronx Zoo, an Indian elephant named Alice examined Keller with her trunk, gently touching her face, shoulder, and foot while Keller returned the exploration with her own hands. For her, the elephant never remained a trunk, an ear, or a leg. Each new observation refined the one before until a coherent whole emerged.

We’ve begun to explain how such integration is possible, and have even designed technologies that solve comparable problems. Keller understood this process through lived experience. Her life reminds us that perception requires seeking, remembering, and assembling. The world yields itself gradually to minds willing to keep exploring.

[Dwayne Godwin is a neuroscientist who researches PTSD, TBI, and alcohol use disorder. He is the co-author of “Out of Your Mind: The Biggest Mysteries of the Human Brain”. The views expressed here are his own and do not necessarily reflect those of the National Institutes of Health or his employer.]

For Further Exploration:

Amedi, A., Malach, R., Hendler, T., Peled, S., & Zohary, E. (2001). Visuo-haptic object-related activation in the ventral visual pathway. Nature Neuroscience, 4(3), 324–330. https://doi.org/10.1038/85201

Bavelier, D., & Neville, H. J. (2002). Cross-modal plasticity: Where and how? Nature Reviews Neuroscience, 3(6), 443–452. https://doi.org/10.1038/nrn848

Bedny, M., Pascual-Leone, A., Dravida, S., & Saxe, R. (2012). A sensitive period for language in the visual cortex: Distinct patterns of plasticity in congenitally versus late blind adults. Brain and Language, 122(3), 162–170. https://doi.org/10.1016/j.bandl.2011.10.005

Cohen, L. G., Celnik, P., Pascual-Leone, A., Corwell, B., Falz, L., Dambrosia, J., Honda, M., Sadato, N., Gerloff, C., Catalá, M. D., & Hallett, M. (1997). Functional relevance of cross-modal plasticity in blind humans. Nature, 389(6647), 180–183. https://doi.org/10.1038/38278

Keller, H. (1903). The story of my life. Doubleday, Page & Company.

Keller, H. (1908). The world I live in. The Century Company.

Keller, H. (1923, September). My animal friends. Zoological Society Bulletin, 26(5).

Kral, A. (2007). Unimodal and cross-modal plasticity in the “deaf” auditory cortex. International Journal of Audiology, 46(9), 479–493. https://doi.org/10.1080/14992020701383027

Kupers, R., Pappens, M., de Noordhout, A. M., Schoenen, J., Ptito, M., & Fumal, A. (2007). rTMS of the occipital cortex abolishes Braille reading and repetition priming in blind subjects. Neurology, 68(9), 691–693. https://doi.org/10.1212/01.wnl.0000255948.60515.11

Lewis, L. B., Saenz, M., & Fine, I. (2010). Mechanisms of cross-modal plasticity in early-blind subjects. Journal of Neurophysiology, 104(6), 2995–3008. https://doi.org/10.1152/jn.00983.2009

Lomber, S. G., Meredith, M. A., & Kral, A. (2010). Cross-modal plasticity in specific auditory cortices underlies visual compensations in the deaf. Nature Neuroscience, 13(11), 1421–1427. https://doi.org/10.1038/nn.2653

Merabet, L. B., & Pascual-Leone, A. (2010). Neural reorganization following sensory loss: The opportunity of change. Nature Reviews Neuroscience, 11(1), 44–52. https://doi.org/10.1038/nrn2758

Pallas, S. L., & Sur, M. (1993). Visual projections induced into the auditory pathway of ferrets. II. Corticocortical connections of primary auditory cortex. Journal of Comparative Neurology, 337(2), 317–333. https://doi.org/10.1002/cne.903370212

Pallas, S. L., Roe, A. W., & Sur, M. (1990). Visual projections induced into the auditory pathway of ferrets. I. Novel inputs to primary auditory cortex (AI) from the LP/pulvinar complex and the topography of the MGN–AI projection. Journal of Comparative Neurology, 298(1), 50–68. https://doi.org/10.1002/cne.902980105

Pascual-Leone, A., Amedi, A., Fregni, F., & Merabet, L. B. (2005). The plastic human brain cortex. Annual Review of Neuroscience, 28, 377–401. https://doi.org/10.1146/annurev.neuro.27.070203.144216

Pietrini, P., Furey, M. L., Ricciardi, E., Gobbini, M. I., Wu, W.-H. C., Cohen, L., Guazzelli, M., & Haxby, J. V. (2004). Beyond sensory images: Object-based representation in the human ventral pathway. Proceedings of the National Academy of Sciences of the United States of America, 101(15), 5658–5663. https://doi.org/10.1073/pnas.0400707101

Ptito, M., Fumal, A., Martens de Noordhout, A., Schoenen, J., Gjedde, A., & Kupers, R. (2008). TMS of the occipital cortex induces tactile sensations in the fingers of blind Braille readers. Experimental Brain Research, 184(2), 193–200. https://doi.org/10.1007/s00221-007-1091-0

Roe, A. W., Pallas, S. L., Kwon, Y. H., & Sur, M. (1992). Visual projections routed to the auditory pathway in ferrets: Receptive fields of visual neurons in primary auditory cortex. Journal of Neuroscience, 12(9), 3651–3664. https://doi.org/10.1523/JNEUROSCI.12-09-03651.1992

Sadato, N., Pascual-Leone, A., Grafman, J., Ibañez, V., Deiber, M.-P., Dold, G., & Hallett, M. (1996). Activation of the primary visual cortex by Braille reading in blind subjects. Nature, 380(6574), 526–528. https://doi.org/10.1038/380526a0

Sadato, N., Pascual-Leone, A., Grafman, J., Deiber, M.-P., Ibañez, V., & Hallett, M. (1998). Neural networks for Braille reading by the blind. Brain, 121(7), 1213–1229. https://doi.org/10.1093/brain/121.7.1213

Sharma, J., Angelucci, A., & Sur, M. (2000). Induction of visual orientation modules in auditory cortex. Nature, 404(6780), 841–847. https://doi.org/10.1038/35009043

Sur, M., Garraghty, P. E., & Roe, A. W. (1988). Experimentally induced visual projections into auditory thalamus and cortex. Science, 242(4884), 1437–1441. https://doi.org/10.1126/science.2462279

Sur, M., & Rubenstein, J. L. R. (2005). Patterning and plasticity of the cerebral cortex. Science, 310(5749), 805–810. https://doi.org/10.1126/science.1112070

von Melchner, L., Pallas, S. L., & Sur, M. (2000). Visual behaviour mediated by retinal projections directed to the auditory pathway. Nature, 404(6780), 871–876. https://doi.org/10.1038/35009102

© 2026 Dwayne W. Godwin.

Emily Dickinson wrote those lines around 1862. The stanza is usually read as a meditation on the vastness of human imagination, the ability of the mind to contain the experience of the boundless sky. But the brain holds more than space. It can also hold time.

If I call your attention to it, you can perceive and estimate the passage of time required to read this sentence. I remember what I did this morning, yesterday, a year ago. I even have an early childhood memory of lying on an ironing board as my mother changed my diaper. I learned to trust—tentatively as a child, more firmly as I grew older—that the next moment in the flow of consciousness would arrive without incident, and would be followed by another, then another, too many to count. I can remember the past with pride or regret. I can imagine the future with joy or anxiety.

We can extend this trick beyond the boundaries of our own lives. We can mentally travel billions of years into the past, toward the beginning of the universe, or forward to imagine the death of our sun as it consumes the Earth. We can go further still, stretching across an ocean of time until the universe grows cold and winks out. The brain is wider than the sky, and more expansive than our own existence.

Physics has made time a strange companion. In everyday life, time seems like a river: one present moment flowing into the next, sometimes a lazy river, sometimes roaring rapids. But relativity made time local and observer-dependent. Clocks moving at different speeds, or occupying different gravitational fields, don’t agree perfectly with one another. Cosmology stretches time across nearly fourteen billion years of expansion. Thermodynamics gives time a direction: entropy increases and our bodies age. Living things resist entropy only temporarily.

Our sense of time may be the brain’s perception of the physics of time unspooling around us. The mind has no meter for entropy, but it can experience the consequences of irreversible change: growth, decay, loss, and memory. The physical world supplies succession and direction. Biology makes that direction intimate. Memory turns it into a life.

The brain seems to build a model of time that is tied to the narrative structure of human existence. In this way, time is the page upon which our stories are written. The components of this biographical model include memory, attention, bodily state, and imagination. Some timing is clock-like: the few milliseconds required to coordinate a movement, the beat of a rhythm, the interval between a flash and a sound. Those forms of timing depend heavily on brain areas including the cerebellum, striatum, thalamus, prefrontal, and parietal circuits. But “felt time” (to borrow Marc Wittmann’s phrase) seems different. It is the subjective continuity that allows us to experience a passing moment as part of a life story. In a prior post I explored consciousness as “the feeling of being you”. That feeling is knit together across moments that provide continuity to your consciousness. You feel that you are the same “you” you were yesterday, today, and who you will be tomorrow.

Memory ties most naturally to the elements of time contained in the past. Whether yesterday or a moment ago, the past exists for us because the memory of it was encoded by our senses, laid down on the imperfect recording medium of human neural circuitry, then accessed and reconstructed at a later moment. It seems obvious that those personal experiences we label “past” are also labeled “memories.” It is less intuitive to say that our sense of time itself may depend on memory.

The brain takes a few milliseconds to convert stimulus to response, so even the present we experience is already slightly behind the world—a wave of existence always frozen just before it crashes, like the helpless boatmen featured in the Great Wave off Kanagawa. The gap is so brief that we rarely notice it. From an evolutionary perspective, that blink of delay is usually small enough not to matter, and our brains are good enough prediction machines that we get along well enough.

The future is more complicated. As we are poised there, helpless before the rogue wave of time, we see the inevitability of falling water and imagine that we hear the crash before it happens. We are Hokusai’s boatmen, frozen for eternity at the instant past becomes future. The predictive power of the brain, immersed in a hospitable pocket of a seemingly orderly universe, has taught us what to expect, even to the point that expectations become routine. We tend to ignore time because it seems both predictable and familiar. That familiarity is due to remembered experience and the ability to weave experience into a model of what was, what is happening, and what may yet come to be.

Has anyone escaped the confines of this mental model of time? None have done so completely, but a few have come close. Their journeys provide clues to our personal connection to time.

Henry Molaison, one of neuroscience’s most consequential research participants, is best known for the profound memory deficits that followed a bilateral medial temporal lobe surgery intended to treat his intractable epilepsy. The operation damaged brain regions essential for forming new declarative memories, including hippocampal and neighboring medial temporal structures. Henry possessed a profound anterograde amnesia, and he took part in one of the longest-running human studies in brain science as “Patient H.M.”.

Most of the tests Henry took were tests of memory. But a lesser-known paper published by Whitman Richards in 1973 tested his perception of time. Richards asked Henry to estimate intervals in a simple task. The experimenter would say “go” and then, after different timed intervals, “stop.” Henry was then asked to reproduce the interval he had just heard, up to a duration of 300 seconds.

Most people can do this fairly well. If their estimated time is plotted against actual time, the data fall roughly along a diagonal line. But Henry’s data looked different.

Henry did well at repeating short intervals. But for longer intervals, his sense of time began to fray. His errors could have gone in either direction, yet he consistently underestimated longer durations. Henry’s memory window (the longest span over which he could retain newly acquired declarative information) was on the order of 20 seconds. His timing errors increased as the task pushed him beyond that window.

You or I can still tap a rhythm, blink at the right instant, or coordinate movement without building a rich autobiographical record of the moment. But Henry’s performance suggested something more subtle: the felt duration of extended experience, the ability to know what a minute has been like as it passes, depends partly on memory systems that stitch moments together.

The short intervals remained within reach of Henry’s mind. As the interval lengthened, time itself began to lose continuity. He could still stand in the moment, but he could no longer hold enough of the passing moment to feel its full duration.

This notion of memory’s involvement in time perception first seems at odds with time flying “when you’re having fun” and the apparent slowness of watched pots coming to a boil. A fun activity with simple demands may be no less enjoyable than a complex activity. The difference is that, in the watched-pot case, attention is fixed on elapsed time instead of being carried away by a more engaging object. Tasks that take the mind away from the passage of time distort perception. We become too distracted to plant the flags that mark the moments.

That may be one reason novelty stretches experience. New places, new people, unfamiliar problems, and emotionally charged events create more landmarks in memory. Routine compresses time. We look back on a period with few signposts and find little to measure. A day can be full while it is happening and strangely thin in retrospect. A week of travel can feel longer and richer than a month of repetition.

If memory is essential to the sense of lived time, do diseases that rob us of memory alter our relationship with time?

Alzheimer’s disease (AD) does not only erase the recall of names, appointments, and recent conversations, it alters a person’s location in time. Early on, the disruption may seem ordinary: a misplaced object or a repeated question. But these are also temporal failures. The recent past no longer holds its shape. The mind may have difficulty placing the self in the sequence of the day. More disturbing, the AD patient may find themselves feeling unmoored from the current moment, and displaced to one from the past. Time becomes less like a continuous road and more like a set of disconnected rooms. Some memories remain with striking clarity: of childhood, old music, or the face of a long-deceased parent. Others dissolve almost as soon as the moment passes through them.

Reminiscent of H.M., but owing to more extensive circuit involvement, even in dementia some short-duration timing may remain partly intact, especially when the task depends on rhythm, motor coordination, or immediate attention. But longer time is different. Longer time must be bound together with memory, context, and expectation.

Korsakoff syndrome offers a related but distinct example. Often associated with chronic thiamine deficiency and alcohol use disorder, it can produce profound anterograde amnesia, confabulation, and temporal disorientation. People with Korsakoff syndrome may speak fluently and appear to be socially present, yet be unable to retain the recent past long enough to keep themselves oriented in time. In retrospective estimates, elapsed time may be underestimated (similar to H.M.), as though periods without durable memory traces collapse when the mind tries to reconstruct them.

These disorders suggest that memory is more than an archive of lived time. It is one of the instruments by which lived time is perceived. It also raises the possibility that changes in subjective duration, temporal orientation, and the ability to order recent events may become useful clues in detecting cognitive decline.

“Tomorrow” appears to be a mental model based on an extrapolation of yesterday into the future, with ourselves placed inside the imagined scene. Endel Tulving and Daniel Schacter examined the role of memory in this process. They studied another amnesic patient known by the initials K.C., whose hippocampus and other brain regions were damaged in a motorcycle accident. K.C. provided an opportunity to examine how profound deficits in episodic memory could erase the ability to imagine the future. When asked to imagine his future, he drew a blank.

Studies of individuals with more circumscribed hippocampal damage support the point. In 2007, Hassabis and colleagues asked subjects with bilateral hippocampal damage to imagine future scenes. The subjects could often produce fragments—objects, details, isolated facts—but had difficulty arranging those fragments into a coherent whole. They could summon pieces of a world, but they struggled to inhabit it. This is what Tulving called mental time travel, later formalized by Schacter and Addis as the constructive episodic simulation hypothesis: memory doesn’t just store the past, it supplies the raw material the mind recombines into imagined futures.

The hippocampus is no longer viewed only as a memory archive, a place where finished experiences are stored after they occur. Human recordings now show neurons in the hippocampus and entorhinal cortex that fire at particular moments in an unfolding experience. These “time cells” help organize sequences, binding events to their temporal location. Animal studies have also shown that the lateral entorhinal cortex carries information about the temporal structure of experience across seconds, minutes, and even longer spans. It’s the brain’s attempt to bind what happened, where it happened, and when it happened into a usable map.

It is dangerous to extrapolate too much from special cases, because the hippocampus is extensively connected to the rest of the brain. Damage or removal of hippocampal tissue not only affects the hippocampus proper, but far-ranging networks connected to it as well. Still, neuroimaging studies have implicated many of the same hippocampal and prefrontal networks when subjects recall prior events and when they imagine future events. The right hippocampus appears especially engaged during future imagining.

This hippocampal involvement has implications beyond what brain regions “light up” during the complex activity of imagining the future. One of the major functions of the hippocampal formation is to encode the memory of place. Arrays of cells within it help represent specific positions in space. Strong hippocampal activation during future imagining raises a natural question. Could one consequence of amnesia be a deficit in the ability to place ourselves in a future narrative frame? Older adults also show changes in the precision of episodic memory, which appear to coincide with reduced fidelity of future imagination. Is one consequence of age-related memory loss a subtle weakening in the ability to imagine ourselves in our own future? Perhaps.

If the hippocampus is like a compass defining our place in the immediate world, these studies, and others, suggest it may also serve as the GPS for mental time travel.

In a parallel with Emily Dickinson, Albert Einstein once said, “Imagination is more important than knowledge.” It is always treacherous to disagree with Einstein, but I will suggest that this famous quote presents a false choice. The key elements of imagination appear to depend intimately on the retention and reworking of prior knowledge. The imagined future is assembled from fragments of our past memories. They’re inextricable.

Age begins to sneak time from us with each passing year. Just as the rise of the world’s oceans may be imperceptible in the short term, our sense of time may shift gradually as we age. In youth, a passing year is a large percentage of the total time we have existed. At age four, a year is fully a quarter of one’s life. The events that populate young lives are infused with novelty and meaning as the brain learns to make sense of an alien planet. They are marked by firsts: the first birthday remembered, the first friend, the first touch of grief, the first time standing at the edge of an ocean, the first realization that adults don’t know everything.

As we age, it becomes easier to feel we’ve “been there, done that”, and harder to encounter the unfamiliar. That expansive youthful year becomes a tiny fraction of a life approaching eighty, closing in on a single percent. The calendar still contains the same twelve months, but the mind measures lived time by change, memory, and event boundaries. This may underlie the cognitive benefits of travel, new friendships, new work, and absorbing hobbies. These things may matter because they provide the signposts of living and change, helping us build and preserve our place in time.

The trick to a longer-seeming life, and perhaps to the ability to imagine a richer future, may be to create wonderful memories and dwell on them long enough to make them durable markers of a lived life. Could this mean that a little social media narcissism is not such a bad thing? Perhaps taking the picture, telling the story, or calling an old friend are all small acts of resistance against time’s compression. They say: this happened; I was there; I remember it, and it mattered.

Our reliance on memory to structure our sense of time, including our future sense, also suggests that imagination is bounded by the experiences we remember, even when those memories are of imaginings rather than direct events. We can imagine the beginning of the universe, though we never witnessed it. We can imagine the death of the sun, though none of us will see it. We can think past the span of our own lives and place ourselves, however briefly, inside the majesty of cosmic time.

Time is stranger than it feels. Relativity suggests there is no one cosmic “now” dividing all events into past and future. Thermodynamics gives direction to time’s arrow in the form of entropy, but we can only guess at the arrow’s path. Cosmology reminds us that a single human life is a blip in the more expansive lifetime of the universe. None of these models explains why time feels like something we live through. That may be the work of memory, storing the now.

As Dickinson also wrote:

“Forever - is composed of Nows

‘Tis not a different time

Except for Infiniteness

And Latitude of Home”

The brain turns successive moments into events, events into stories, and stories into a felt sense of self in time. It gathers the irreversible changes of the world—growth, aging, loss, recovery, and decay—and transmutes them into the experience of a life moving forward. The moments we experience in the here and now are held long enough to become memories, and those memories are recombined richly enough to imagine what may come next.

Thanks for your time.

Dwayne Godwin is a Professor of Translational Neuroscience at the Wake Forest University School of Medicine. He is the co-Author of “Out of Your Mind: The Biggest Mysteries of the Human Brain.”

This commentary should not be interpreted as reflecting the views of his employer.

For Further Exploration

Addis, D. R., Wong, A. T., & Schacter, D. L. (2007). Remembering the past and imagining the future: Common and distinct neural substrates during event construction and elaboration. Neuropsychologia, 45(7), 1363–1377. https://doi.org/10.1016/j.neuropsychologia.2006.10.016

Annese, J., Schenker-Ahmed, N. M., Bartsch, H., Maechler, P., Sheh, C., Thomas, N., Kayano, J., Ghatan, A., Bresler, N., Frosch, M. P., Klaming, R., & Corkin, S. (2014). Postmortem examination of patient H.M.’s brain based on histological sectioning and digital 3D reconstruction. Nature Communications, 5, Article 3122. https://doi.org/10.1038/ncomms4122

Buhusi, C. V., & Meck, W. H. (2005). What makes us tick? Functional and neural mechanisms of interval timing. Nature Reviews Neuroscience, 6(10), 755–765. https://doi.org/10.1038/nrn1764

Carroll, S. (2010). From eternity to here: The quest for the ultimate theory of time. Dutton.

Coull, J. T., Cheng, R.-K., & Meck, W. H. (2011). Neuroanatomical and neurochemical substrates of timing. Neuropsychopharmacology, 36(1), 3–25. https://doi.org/10.1038/npp.2010.113

Dickinson, E. (1998). The poems of Emily Dickinson: Variorum edition (R. W. Franklin, Ed.). Belknap Press of Harvard University Press. Original work composed ca. 1862.

Dickinson, E. (2024, January 26). Forever – is composed of Nows –. Wikisource. Retrieved July 9, 2026, from wikisource.org

Eddington, A. S. (1928). The nature of the physical world. The Macmillan Company; Cambridge University Press.

El Haj, M., Kapogiannis, D., & Antoine, P. (2016). Time distortions in Alzheimer’s disease: A systematic review and theoretical integration. npj Aging and Mechanisms of Disease, 2, Article 16016. https://doi.org/10.1038/npjamd.2016.16

El Haj, M., Nandrino, J.-L., Kessels, R. P. C., Matton, C., Bacquet, J.-E., Urso, L., & Antoine, P. (2017). Retrospective time perception in Korsakoff’s syndrome. The Journal of Neuropsychiatry and Clinical Neurosciences, 29(3), 245–251. https://doi.org/10.1176/appi.neuropsych.16100206

Godwin, D. (2015). Remembering the future. BrainFacts.org. https://www.brainfacts.org/neuroscience-in-society/the-arts-and-the-brain/2015/remembering-the-future

Hassabis, D., Kumaran, D., Vann, S. D., & Maguire, E. A. (2007). Patients with hippocampal amnesia cannot imagine new experiences. Proceedings of the National Academy of Sciences, 104(5), 1726–1731. https://doi.org/10.1073/pnas.0610561104

Hokusai, K. (ca. 1830–1832). Under the Wave off Kanagawa (Kanagawa oki nami ura), also known as The Great Wave, from the series Thirty-six Views of Mount Fuji (Fugaku sanjūrokkei) [Woodblock print; ink and color on paper]. The Metropolitan Museum of Art. https://www.metmuseum.org/art/collection/search/45434

Korkki, S. M., Richter, F. R., Jeyarathnarajah, P., & Simons, J. S. (2020). Healthy ageing reduces the precision of episodic memory retrieval. Psychology and Aging, 35(1), 124–142. https://doi.org/10.1037/pag0000432

MacDonald, C. J., Lepage, K. Q., Eden, U. T., & Eichenbaum, H. (2011). Hippocampal “time cells” bridge the gap in memory for discontiguous events. Neuron, 71(4), 737–749. https://doi.org/10.1016/j.neuron.2011.07.012

Matell, M. S., & Meck, W. H. (2004). Cortico-striatal circuits and interval timing: Coincidence detection of oscillatory processes. Cognitive Brain Research, 21(2), 139–170. https://doi.org/10.1016/j.cogbrainres.2004.06.012

Pastalkova, E., Itskov, V., Amarasingham, A., & Buzsáki, G. (2008). Internally generated cell assembly sequences in the rat hippocampus. Science, 321(5894), 1322–1327. https://doi.org/10.1126/science.1159775

Reddy, L., Zoefel, B., Possel, J. K., Peters, J. C., Dijksterhuis, D., Poncet, M., van Straaten, E. C. W., Baayen, J. C., Idema, S., & Self, M. W. (2021). Human hippocampal neurons track moments in a sequence of events. The Journal of Neuroscience, 41(31), 6714–6725. https://doi.org/10.1523/JNEUROSCI.3157-20.2021

Richards, W. (1973). Time reproductions by H.M. Acta Psychologica, 37(4), 279–282. https://doi.org/10.1016/0001-6918(73)90020-6

Rosenbaum, R. S., Köhler, S., Schacter, D. L., Moscovitch, M., Westmacott, R., Black, S. E., Gao, F., & Tulving, E. (2005). The case of K.C.: Contributions of a memory-impaired person to memory theory. Neuropsychologia, 43(7), 989–1021. https://doi.org/10.1016/j.neuropsychologia.2004.10.007

Rovelli, C. (2018). The order of time (E. Segre & S. Carnell, Trans.). Riverhead Books. Original work published 2017.

Schacter, D. L., Addis, D. R., Hassabis, D., Martin, V. C., Spreng, R. N., & Szpunar, K. K. (2012). The future of memory: Remembering, imagining, and the brain. Neuron, 76(4), 677–694. https://doi.org/10.1016/j.neuron.2012.11.001

Schacter, D. L., Gaesser, B., & Addis, D. R. (2013). Remembering the past and imagining the future in the elderly. Gerontology, 59(2), 143–151. https://doi.org/10.1159/000342198

Scoville, W. B., & Milner, B. (1957). Loss of recent memory after bilateral hippocampal lesions. Journal of Neurology, Neurosurgery & Psychiatry, 20(1), 11–21. https://doi.org/10.1136/jnnp.20.1.11

Tacikowski, P., Kalender, G., Ciliberti, D., & Fried, I. (2024). Human hippocampal and entorhinal neurons encode the temporal structure of experience. Nature, 635, 160–167. https://doi.org/10.1038/s41586-024-07973-1

Teghil, A., Boccia, M., Bonavita, A., Guariglia, C., & Giannini, A. M. (2023). Multidimensional assessment of time perception along the Alzheimer’s disease continuum. Scientific Reports, 13, Article 22908. https://doi.org/10.1038/s41598-023-49222-x

Tsao, A., Sugar, J., Lu, L., Wang, C., Knierim, J. J., Moser, M.-B., & Moser, E. I. (2018). Integrating time from experience in the lateral entorhinal cortex. Nature, 561(7721), 57–62. https://doi.org/10.1038/s41586-018-0459-6

Tulving, E. (1985). Memory and consciousness. Canadian Psychology/Psychologie canadienne, 26(1), 1–12. https://doi.org/10.1037/h0080017

Tulving, E. (2002). Episodic memory: From mind to brain. Annual Review of Psychology, 53, 1–25. https://doi.org/10.1146/annurev.psych.53.100901.135114

Tulving, E., Schacter, D. L., McLachlan, D. R., & Moscovitch, M. (1988). Priming of semantic autobiographical knowledge: A case study of retrograde amnesia. Brain and Cognition, 8(1), 3–20. https://doi.org/10.1016/0278-2626(88)90035-8

Umbach, G., Kantak, P., Jacobs, J., Kahana, M., Pfeiffer, B. E., Sperling, M., & Lega, B. (2020). Time cells in the human hippocampus and entorhinal cortex support episodic memory. Proceedings of the National Academy of Sciences, 117(45), 28463–28474. https://doi.org/10.1073/pnas.2013250117

Viereck, G. S. (1929, October 26). What life means to Einstein: An interview by George Sylvester Viereck. The Saturday Evening Post, 17, 110, 113–114, 117.

Wittmann, M. (2016). Felt time: The psychology of how we perceive time (E. Butler, Trans.). MIT Press. (Original work published 2015)

© 2026 Dwayne Godwin. Developed from an essay originally published at BrainFacts.org (2015).

In the spring of 2025, aides of the Secretary of Defense used polygraphs on their own coworkers as part of an investigation of alleged leaks. As recently as May 2026, FBI Director Kash Patel reportedly ordered polygraphs for more than two dozen current and former staff members in an effort to identify people who had spoken with reporters.

Polygraphs are a trope of police and espionage dramas, so much so that they have accrued a rather strange reputation as a means of getting at the truth, when conventional means seem inadequate. But do they work? In this essay, I’ll make the case that their reputation is undeserved.

Police departments have long used polygraphs to narrow suspects and push interrogations toward confession. Intelligence and national security agencies use them for things like counterintelligence and leak hunts. For years, private employers used them so freely that Congress eventually had to rein them in, by passing the Employee Polygraph Protection Act of 1988. The practice continues because federal agencies still find it useful. It has the trappings of science and measurement, of certainty from data. And, it can be intimidating.

A polygraph records things like respiration, sweating, blood pressure, and pulse rate.

Polygraphs live in a grey area between pseudoscience and actual science. They do seem plausible in the abstract. Lying can be stressful, and stress changes the body. But the physiological signals polygraphs measure don’t actually tell you that a person is lying. Fear can produce them. So can shame, anger, confusion, exhaustion, medication, trauma, stimulants, or the ordinary stress of having a government official stare at you while implying that you are disloyal. The responses produced by a polygraph examination simply don’t map well enough onto deception to make the test reliable.

That’s because many of the body’s stress signals are driven by the sympathetic nervous system, the system that prepares the body for fight or flight. When it’s activated, your heart rate rises, breathing changes, blood pressure may increase, and sweat glands become more active. Those changes can happen when someone is lying, but they can also happen whenever a person feels threatened, or unfairly accused.

In formal logic, polygraphs are an embodiment of a fallacy called, “affirming the consequent”:

If a person is lying, they may show arousal.

This person shows arousal.

Therefore, this person is lying.

But of course, “arousal” can have many causes, not just lying.

There is no little lie meter tucked away in your brain. There are only millions of years of evolutionary pressure leading us to respond to stress in specific ways that may be adaptive if you were living in the wild, but not when strapped to a machine.

William Moulton Marston, a psychologist, and eventually the creator of Wonder Woman, developed a blood-pressure based approach to lie detection and sold the broader idea with flair. Later, in creating Wonder Woman, he gave her the Lasso of Truth, able to compel those under its spell to spill the beans. It’s almost too on the nose as a cultural echo of the machine he envisioned.

Others reasoned that if blood pressure was useful, perhaps additional measures would be even better. John Larson helped develop the first practical polygraph in the 1920s, while working with Berkeley police chief August Vollmer, hoping that simultaneous recordings of blood pressure, pulse, and respiration could make interrogation more disciplined and less dependent on coercion. The public and the press soon gave it a more powerful and misleading name: the “lie detector.”

Police departments used it with growing confidence, even while Larson’s own experience showed that many results were ambiguous and heavily dependent on the examiner’s interpretation. Over time, he came to see the polygraph as an invention that had failed its original purpose: a tool meant to temper rough interrogation had become an instrument that was more psychologically abusive, often treated as if it could reveal truth directly. Near the end of his life, he reportedly compared it to a “Frankenstein’s monster,” a creation he rejected and had spent decades trying, and failing, to control.

One reason that polygraphs endure is that under tightly controlled conditions, certain testing formats can perform above chance. But “above chance” looks worse where the stakes are high, such as when someone’s job or freedom is at stake.

The track record of polygraphs isn’t reassuring. False negatives (failures to detect actual lies) include Aldrich Ames, who passed polygraphs while spying for the Soviet Union and then Russia. Gary Ridgway, the Green River Killer (responsible for dozens of murders), reportedly passed one, too. These examples make the stark point that for a physiological response to be measured, the liar has to care that they are lying. Individuals with profound psychopathology may not be wired that way.

Some studies suggest that individuals high in certain traits, particularly callous or affectively cold traits, may show blunted responses to fear, punishment, or aversive anticipation. That means the basic premise on which a polygraph is based is unstable. The same test may overread fear in an innocent person and miss it in someone who feels little fear at all. Ironically, the traits investigators most want to uncover may be among the traits that make a person less physiologically readable by a test built around stress.

The fact that sympathetic activity can be engaged independently of lying can lead to false positives, which have dire consequences. Jeffrey Deskovic was a teenager when he was told he had failed a polygraph during the investigation of a rape and murder in New York. After prolonged interrogation, he confessed to a crime he didn’t commit. DNA later exonerated him, but only after he had spent years in prison.

Internal mental states are not read out in the way polygraph enthusiasts want them to be. Deception is not a single physiological state, or even a single psychological state. Lying can involve combinations of planning, memory, conflict, emotional flatness, pleasure, fear, or none of the above. The same visible signal can arise from very different internal causes. The brain is itself a dynamic system. It’s distributed, noisy, context-sensitive, and often opaque even to the person living inside it. The simple readouts provided by polygraphs are no less noisy.

What if what it means to lie doesn’t remain fixed in the brain? What if it’s a moving target? A phenomenon known as the illusory truth effect holds that repeated claims can begin to feel truer simply because they become familiar. Studies of deception and memory suggest that telling lies can blur later belief about what actually happened.

Repeated dishonesty may also become less emotionally aversive over time. One brain-imaging study found that self-serving lies escalated as amygdala responses to dishonesty declined. A practiced liar, then, may not experience the binary internal conflict imagined by the examiner. The lie may become partly absorbed into the liar’s own feeling of what is true.

Then there’s the problem of whether an event is perceived as a lie in the first place. In a prior post, I explored how two people can witness the same event, yet walk away with a very different interpretation based on their cognitive biases. If the person strapped into a polygraph believes they are telling the truth, how can a system detect a lie?

A polygraph can’t accurately detect deception if the body has stopped reacting to the lie as a stressful transgression.

If peripheral measures are inadequate, might looking deeper, into the brain itself, produce a better result? That’s the notion behind fMRI-based lie detection. Functional MRI does not measure deception directly; it measures blood-oxygen-level-dependent, or BOLD, changes that are used as an indirect marker of regional brain activity. In deception studies, subjects are usually asked to lie or tell the truth while in the scanner, and investigators then look for patterns of activity that differ between truthful and deceptive responses. Early studies produced encouraging results, with deception often associated with greater activity in prefrontal and parietal regions involved in cognitive control, response inhibition, and conflict monitoring. Some individual-subject studies reported accuracy in the range of roughly 76% to more than 90% under controlled laboratory conditions, which made fMRI seem, for a time, like a more sophisticated successor to the polygraph.

One problem is that lying is different in a lab than in the real world. Many studies rely on instructed lies, mock crimes, or forced-choice tasks in which the stakes are low and the experimenter already knows what kind of lie is being tested. Reviews have emphasized that these designs don’t establish that fMRI can identify deception in ordinary legal or security settings, where memories and motives may be uncertain, and fear or self-protection inseparable from the act of answering.

The method is also vulnerable to countermeasures: one study found that covert mental strategies could disrupt fMRI-based deception detection, and legal commentators have noted that courts have generally regarded the technology as insufficiently validated for real-world adjudication. The current state of the art is therefore scientifically interesting but forensically fragile. Brain patterns can show that deception is cognitively demanding under some experimental conditions, but they don’t yet provide an accurate enough test of truth to stake a person’s liberty or reputation upon.

Another challenge inherent to both polygraphs and brain imaging is that they’re indirect. After all, what is the basic unit of a lie? Is it a physiological response, the activity of a cluster of cells, a region of the brain, or (more likely) multiple regions? What is the relationship between the decision to lie and the responses being measured? In the case of brain imaging, does a given area activating indicate fluid deception, or an internal struggle to tell the truth? Until these questions are answered, how can these methods be relied upon?

The Daubert standard and Federal Rule of Evidence 702 govern expert testimony in federal court. A judge must ask whether a method is scientifically sound, whether it rests on sufficient data, whether its error rate is known, whether it has been validated outside artificial laboratory settings, and whether it will help the jury rather than confuse it. The Supreme Court reached a similar practical conclusion in United States v. Scheffer, upholding a military rule that categorically excluded polygraph results from courts-martial and noting the continuing dispute over polygraph reliability. Neither polygraphs nor fMRI lie detection has yet surmounted this barrier in a way that would justify treating them as instruments of truth.

A polygraph readout or a fancy brain image can make an uncertain inference look like a fact. Studies of science communication have found that brain images can make accompanying explanations seem more credible, even when the image adds little or no useful information; neuroscience technobabble (Neurobabble?) itself could also make weak explanations feel more satisfying to nonexperts. The effect is not uniform across all settings, and some aspects have failed to be replicated, but the concern is strong enough to explain why courts worry that a colorful brain scan may carry more persuasive force than its underlying inference deserves, and obscure an underlying messiness.

At best, these methods pick up signals related to things like conflict, attention, cognitive effort, or recognition. But a person can recognize details for innocent reasons. A brain may work hard while telling the truth. Laboratory studies may be able to separate groups under carefully staged conditions, but real investigations are full of noise.

Whenever institutions become paranoid enough, the response can be aggressive. Technology offers an easy and quick answer to complex problems. But the aggressiveness may only serve to weed out those who are most upset at being suspected of potentially illegal activities. It may also be a convenient pseudoscientific tool to intimidate employees who are deemed disloyal, or to assign guilt when other data points are ambiguous.

There’s a reason polygraphs are typically excluded from criminal trials. In a prior essay, I describe how artificial intelligence fools people into thinking that it possesses consciousness. Polygraphs are different than AI, but some of the same human frailties come into play: the desire for easy answers, our need to confirm our own biases, and a trust of technology whose workings produce an answer, even when we can’t cleanly connect either objective or subjective truth to the response of the body.

The search for certainty through ever-finer measurement produces its own uncertainty. In physics, it might be couched as the observer effect, where the instrument doesn’t sit outside the system it studies. A polygraph or an fMRI scanner is no exception. Neither offers a completely neutral window into the mind, or the lies we tell ourselves.

Dwayne Godwin is a Professor of Translational Neuroscience at the Wake Forest University School of Medicine. He is the co-Author of “Out of Your Mind: The Biggest Mysteries of the Human Brain.”

This commentary should not be interpreted as reflecting the views of his employer.

For Further Exploration

Aftergood, S. (2000) Polygraph Testing and the DOE National Laboratories. Science 290, 939-940. DOI:10.1126/science.290.5493.939

Alder, K. (2002). A social history of untruth: Lie detection and trust in twentieth-century America. Representations, 80(1), 1–33. https://doi.org/10.1525/rep.2002.80.1.1

Ames, A. H. (2000). A letter from Aldrich Ames on polygraph testing. Federation of American Scientists. https://sgp.fas.org/othergov/polygraph/ames.html

AntiPolygraph.org. (2020, May 7). “Green River Killer” Gary Leon Ridgway’s polygraph charts. https://antipolygraph.org/blog/2020/05/07/green-river-killer-gary-leon-ridgways-polygraph-charts/

Brennen, T., & Magnussen, S. (2023). Lie detection: What works? Current Directions in Psychological Science, 32(5), 395–401. https://doi.org/10.1177/09637214231173095

Bunn, G. C. (1997). The lie detector, Wonder Woman and liberty: The life and work of William Moulton Marston. History of the Human Sciences, 10(1), 91–119. https://doi.org/10.1177/095269519701000105

Bunn, G. C. (2019). “Supposing that truth is a woman, what then?”: The lie detector, the love machine, and the logic of fantasy. History of the Human Sciences, 32(5), 135–163. https://doi.org/10.1177/0952695119867022

Cook, L. G., & Mitschow, C. (2019). Beyond the polygraph: Deception detection and the autonomic nervous system. Federal Practitioner, 36(7), 316–321. https://www.mdedge.com/fedprac/article/204072/mental-health/beyond-polygraph-deception-detection-and-autonomic-nervous

Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993). https://supreme.justia.com/cases/federal/us/509/579/

Dechêne, A., Stahl, C., Hansen, J., & Wänke, M. (2010). The truth about the truth: A meta-analytic review of the truth effect. Personality and Social Psychology Review, 14(2), 238–257. https://doi.org/10.1177/1088868309352251

Employee Polygraph Protection Act of 1988, 29 U.S.C. §§ 2001–2009. https://uscode.house.gov/view.xhtml?path=/prelim@title29/chapter22&edition=prelim

Farah, M. J., Hutchinson, J. B., Phelps, E. A., & Wagner, A. D. (2014). Functional MRI-based lie detection: Scientific and societal challenges. Nature Reviews Neuroscience, 15(2), 123–131. https://doi.org/10.1038/nrn3665

Fazio, L. K., Brashier, N. M., Payne, B. K., & Marsh, E. J. (2015). Knowledge does not protect against illusory truth. Journal of Experimental Psychology: General, 144(5), 993–1002. https://doi.org/10.1037/xge0000098

Federal Rule of Evidence 403. https://www.law.cornell.edu/rules/fre/rule_403

Federal Rule of Evidence 702. https://www.law.cornell.edu/rules/fre/rule_702

Frye v. United States, 293 F. 1013 (D.C. Cir. 1923). https://law.justia.com/cases/district-of-columbia/court-of-appeals/1923/no-3968.html

Ganis, G., Rosenfeld, J. P., Meixner, J., Kievit, R. A., & Schendan, H. E. (2011). Lying in the scanner: Covert countermeasures disrupt deception detection by functional magnetic resonance imaging. NeuroImage, 55(1), 312–319. https://doi.org/10.1016/j.neuroimage.2010.11.025

Garrett, N., Lazzaro, S. C., Ariely, D., & Sharot, T. (2016). The brain adapts to dishonesty. Nature Neuroscience, 19(12), 1727–1732. https://doi.org/10.1038/nn.4426

Hasher, L., Goldstein, D., & Toppino, T. (1977). Frequency and the conference of referential validity. Journal of Verbal Learning and Verbal Behavior, 16(1), 107–112. https://doi.org/10.1016/S0022-5371(77)80012-1

Hinkle, C. (2021). The modern lie detector: AI-powered affect screening and the Employee Polygraph Protection Act. Georgetown Law Journal, 109, 1201–1240. https://www.law.georgetown.edu/georgetown-law-journal/wp-content/uploads/sites/26/2021/06/Hinkle-The_Modern_Law_Detector.pdf

Im, S.-H., Varma, K., & Varma, S. (2017). Extending the seductive allure of neuroscience explanations effect to popular articles about educational topics. British Journal of Educational Psychology, 87(4), 518–534. https://doi.org/10.1111/bjep.12162

Johnson, M. B., & Drucker, J. (2009). Two recently confirmed false confessions: Byron A. Halsey and Jeffrey M. Deskovic. The Journal of Psychiatry & Law, 37(1), 25–46. https://doi.org/10.1177/009318530903700105

Jones, R. P., & Singh, K. (2025, April 29). FBI starts using polygraph tests in internal leak investigations. Reuters. https://www.reuters.com/world/us/fbi-starts-using-polygraph-tests-internal-leak-investigations-2025-04-29/

Knappe, J., Ullsperger, M., & Kirschner, H. (2025). Decoding deception with the P300: A meta-analysis of the Concealed Information Test. International Journal of Psychophysiology, 215, Article 113236. https://doi.org/10.1016/j.ijpsycho.2025.113236

Kozel, F. A., Johnson, K. A., Mu, Q., Grenesko, E. L., Laken, S. J., & George, M. S. (2005). Detecting deception using functional magnetic resonance imaging. Biological Psychiatry, 58(8), 605–613. https://doi.org/10.1016/j.biopsych.2005.07.040

Kozel, F. A., Padgett, T. M., & George, M. S. (2004). A replication study of the neural correlates of deception. Behavioral Neuroscience, 118(4), 852–856. https://doi.org/10.1037/0735-7044.118.4.852

Lorber, M. F. (2004). Psychophysiology of aggression, psychopathy, and conduct problems: A meta-analysis. Psychological Bulletin, 130(4), 531–552. https://doi.org/10.1037/0033-2909.130.4.531

Lowell, H. (2025, July 31). Senior aides to Hegseth gave polygraphs to fellow Pentagon colleagues. The Guardian. https://www.theguardian.com/us-news/2025/jul/31/hegseth-defense-secretary-polygraphs

McCabe, D. P., & Castel, A. D. (2008). Seeing is believing: The effect of brain images on judgments of scientific reasoning. Cognition, 107(1), 343–352. https://doi.org/10.1016/j.cognition.2007.07.017

Meijer, E. H., & Verschuere, B. (2017). Deception detection based on neuroimaging: Better than the polygraph? Journal of Forensic Radiology and Imaging, 8, 17–21. https://doi.org/10.1016/j.jofri.2017.03.003

Michael, R. B., Newman, E. J., Vuorre, M., Cumming, G., & Garry, M. (2013). On the (non)persuasive power of a brain image. Psychonomic Bulletin & Review, 20(4), 720–725. https://doi.org/10.3758/s13423-013-0391-6

National Research Council. (2003). The polygraph and lie detection. The National Academies Press. https://doi.org/10.17226/10420

Otgaar, H., & Baker, A. (2018). When lying changes memory for the truth. Memory, 26(1), 2–14. https://doi.org/10.1080/09658211.2017.1340286

Polage, D. C. (2017). The effect of telling lies on belief in the truth. Europe’s Journal of Psychology, 13(4), 633–644. https://doi.org/10.5964/ejop.v13i4.1422

Polage, D. C. (2019). Liar, liar: Consistent lying decreases belief in the truth. Applied Cognitive Psychology, 33(4), 527–536. https://doi.org/10.1002/acp.3489

Reichert, D. G. (2004). Chasing the devil: My twenty-year quest to capture the Green River Killer. Little, Brown.

Saulny, S. (2024, October 22). Why the creator of one of the first “lie detectors” lived to regret his invention. Smithsonian Magazine. https://www.smithsonianmag.com/smithsonian-institution/why-creator-first-lie-detectors-regret-invention-polygraph-machine-inventor-180985222/

Synnott, J., Dietzel, D., & Ioannou, M. (2015). A review of the polygraph: History, methodology and current status. Crime Psychology Review, 1(1), 59–83. https://doi.org/10.1080/23744006.2015.1060080

United States v. Scheffer, 523 U.S. 303 (1998). https://supreme.justia.com/cases/federal/us/523/303/

United States v. Semrau, 693 F.3d 510 (6th Cir. 2012). https://www.opn.ca6.uscourts.gov/opinions.pdf/12a0312p-06.pdf

U.S. Department of Labor, Wage and Hour Division. (n.d.). Employee Polygraph Protection Act. https://www.dol.gov/agencies/whd/polygraph

U.S. Press Freedom Tracker. (2026, May 7). FBI Director Kash Patel targets press. https://pressfreedomtracker.us/all-incidents/fbi-director-kash-patel-targets-press/

U.S. Senate Select Committee on Intelligence. (1994). An assessment of the Aldrich H. Ames espionage case and its implications for U.S. intelligence (Committee Print 103-90). U.S. Government Printing Office. https://www.intelligence.senate.gov/wp-content/uploads/2024/08/sites-default-filesations-10390.pdf

Veit, R., Flor, H., Erb, M., Hermann, C., Lotze, M., Grodd, W., & Birbaumer, N. (2013). Deficient fear conditioning in psychopathy as a function of interpersonal and affective disturbances. Frontiers in Human Neuroscience, 7, Article 706. https://doi.org/10.3389/fnhum.2013.00706

Weisberg, D. S., Keil, F. C., Goodstein, J., Rawson, E., & Gray, J. R. (2008). The seductive allure of neuroscience explanations. Journal of Cognitive Neuroscience, 20(3), 470–477. https://doi.org/10.1162/jocn.2008.20040

© 2026 Dwayne Godwin. All rights reserved. No part of this publication may be reproduced or transmitted without permission.

WONDER WOMAN and all related characters, names, logos, distinctive likenesses, and elements are trademarks and copyrights of DC Comics. All rights reserved.

Students of moral philosophy will recognize this as a real-world example of the classic trolley problem.

In the classroom version, a runaway trolley heads toward five people. Pulling a lever diverts the path, sacrificing one person for the several who would otherwise die. Students argue about numbers, duty, innocence, and whether refusing to act is itself a choice. It recalls Mr. Spock’s line from The Wrath of Khan: “The needs of the many outweigh the needs of the few…or the one.”

Now, replace you in this scenario with an AI. That’s not so far-fetched, because it’s actually happening.

The version of the trolley problem I posed is more complicated because the five are jaywalkers, while the lone person is simply minding their own business on a sidewalk. But it could be even more complex. The calculation driving your decision might change further if you knew the lone pedestrian held the cure for cancer in their mind.

When you’re the driver, these complexities are invisible. Your decision is based on who you are, and what you see, in that fateful moment. The complications with an AI behind the wheel extend to those who trained or oversee its operation, or regulators who permitted it to drive the streets. Essentially, the issue is who has the agency to make the decision at all, and who is accountable once the decision is made.

Legal responsibility usually starts with a person. Criminal law then asks what that person did, what they knew, what they intended, and what a reasonable person would have done: actus reus (literally, the guilty act) and mens rea (the “guilty mind”). The Latin phrasing for someone not of sound mind, is “non compos mentis.”

With AI, the more apt framing is, perhaps, “non compos machinae.” The guilty act is visible but, at least right now, there’s no mind to be found. The question of responsibility disperses across training sets, deployment decisions, and any humans who might be in the loop. An AI system itself has neither conscience nor consciousness to examine. But it can still be difficult to know where the humans who trained it stop, and the machine begins.

AI is a unique case. It exists in the world as a product, a “brain” without mind, that is put into the decision space of a person, but as I’ve suggested elsewhere, it currently lacks the inner life that would make accountability coherent and just. Current systems don’t have the architecture we would expect for consciousness: no body, no developmental history, and no valuation circuitry shaped by human experiences. Consequences for actions are a symbolic construct without a tangible reality. There is no one “there” to be held accountable.

So, when the law needs to assign responsibility, it still looks for a human being.

Elaine Herzberg was 49 years old when she was struck and killed while crossing the road with her bicycle. She had grown up in Apache Junction, Arizona, graduated high school in 1985, outlived her husband, and by her daughter Christine’s account was kind and generous. She wrote poems and was a grandmother.

In March 2018, an Uber self-driving vehicle struck and killed Herzberg as she walked her bicycle across a four-lane road in Tempe. She has the tragic distinction of being the first pedestrian killed by an autonomous vehicle on a public road. The car’s sensors detected her six seconds before impact. The system classified her, sequentially, as an unknown object, a vehicle, and then a bicycle, each reclassification resetting the threat calculation. She had almost made the crossing when she was hit. There was no trolley problem, because in the AI’s calculation there wasn’t a human in its path, and no recognized alternative path. It never braked. A human backup driver, employed specifically to monitor the system and intervene if something went wrong, was behind the wheel, but was looking at her phone.

But there’s more to it than simply, “The AI did it.” Internal documents showed Uber had disabled the Volvo’s factory emergency braking system to reduce false positives during testing. The company had cut the number of required human backup drivers from two to one. There were known concerns about the system’s handling of jaywalkers. The safety driver was charged with negligent homicide, the only person to face criminal consequences for a death that resulted from years of institutional decisions, capped by a fateful moment of distraction. Prosecutors did not charge Uber, but the company paid an undisclosed settlement and eventually sold its self-driving unit.

These aren’t isolated incidents. More than 5,000 crashes or incidents involving vehicles equipped with automated-driving or advanced driver-assistance systems have been reported to NHTSA, and the federal dataset is still growing. As recently as June 2026, a Tesla whose driver claimed the vehicle was in self-driving mode crashed into a home, killing the occupant.

Autopilot is a driver-assistance system handling steering, acceleration, and braking within lanes, but Tesla marketed it with language that may have implied more capability than the system could deliver. “Full Self-Driving” was an option years before the technology could meet those claims. The gap between the marketing and engineering performance created a predictable liability problem in the form of drivers who believed the system could handle more than it could, in situations it was not designed for, at speeds where there’s no margin for error.

In Benavides v. Tesla, a Florida jury found Tesla partly liable in a fatal crash and awarded substantial damages, one of the first verdicts to hold the company responsible for an Autopilot-related death. Other juries have sided with Tesla, finding that drivers who ignored warnings and removed their hands from the wheel bore the primary responsibility. The central question is when a company sells a product named Full Self-Driving, and a person dies, where does the driver’s responsibility end and the manufacturer’s responsibility begin? Tesla’s answer has been that the warnings were adequate and the drivers misused the feature. Plaintiffs have argued that the warnings were buried under years of marketing and hype that may have overreached. There was no entity positioned to tell them “no.”

In biomedical research, when a new intervention is to be tested on human beings, something called an Institutional Review Board (IRB) weighs the possible benefits against potential harms. The United States has roughly 2,300 registered IRBs, and there are eight where I work. A study doesn’t begin until the IRB assesses the methods and the risk vs. benefits of the study. An IRB requires consent, and can stop a trial if the risk profile shifts or if there are deviations from protocol. This approach to research was born of a troubling history where some research institutions had placed human beings at risk without adequate consent or protection from exploitation.

Herman Shaw was a 30-year-old cotton farmer from Tuskegee, Alabama. He had once hoped to study engineering, but family obligations kept him working on the farm. During the Depression, the promise of free government medical care would have been attractive to Shaw and to hundreds of Black men in similar circumstances. The men were examined, fed, and treated as participants in a health program, but the Public Health Service was actually observing the course of untreated syphilis. Shaw later learned that he had tested positive and had been deceived. He was not offered effective treatment, even after penicillin became available. Shaw survived, but many others did not. The study was wrong because hundreds of men were lied to, denied informed consent, denied available care, and exploited because of their race, poverty, and limited access to health care.

The Tuskegee syphilis study led to Congress passing the National Research Act of 1974 and created a commission that produced the Belmont Report, which established an ethical framework for research. This framework included respect for persons, beneficence (an intent to do good), and justice. Progress could still proceed, but only with consent, independent review, and continuing oversight when people were being asked to bear risk for the sake of knowledge, medical innovation, or public benefit. IRBs are derived from these principles.

A practical consequence of IRBs is that participants must agree to participate in an experiment or clinical trial. They must be fully informed of risks, and they can withdraw from a study. This is especially important for protecting vulnerable populations, such as prisoners and children.

You might be asking, “What does this have to do with AI?” Consider that your kids, just by walking down the street, are the subject of an experiment in AI product development.

It turns out that there is no comparable consent-based framework that governs the deployment of autonomous systems. When a self-driving vehicle takes to public roads, every nearby pedestrian is effectively inside an experiment to which they have not consented, of which they have not been advised of the risk, and for which they receive little tangible benefit. There is no ability to withdraw. The trolley problem becomes a live implementation trial, run on public infrastructure, with all of us as unwitting subjects. While it’s true that the National Highway Traffic Safety Administration (NHTSA) investigates events like the crash in Katy, Texas, this is after the fact. They do not approve these technologies prospectively. NHTSA can force a remedy that may disable, restrict, or effectively halt vehicle features, but often only after damage is done.

An ethical system was built for conducting biomedical research because letting markets discover the limits of a drug, or other medical intervention, without weighing potential harms was seen as immoral and unacceptable. This is compounded in technologies where there is a profit motive to keep pushing the adoption of the technology. And while the IRB model has its own flaws, it provides a mechanism for accountability if something goes wrong.

While I’ve focused on vehicles, the same principles apply to any scenario where AI is used. It’s worth considering whether AI systems making consequential decisions about unconsenting people, whether in healthcare, criminal justice, employment, or public safety, may be running a kind of experiment on them. We should be asking whether we require those experiments to meet the same standards of respect for persons, and the weighing of risk versus benefit that we require of new innovations in healthcare.

Pope Leo XIV makes a similar point in Magnifica Humanitas, his recent encyclical on artificial intelligence. His emphasis is theological and social rather than safety or legality, but his concern has a similar moral tone: when AI is placed in a decision role that can affect rights, opportunity, reputation, public services, work, or freedom, it’s no longer simply a technical convenience, but a moral and civic arrangement, with the danger that institutions will let machines make consequential judgments while no one is held morally accountable.

Current AI systems almost certainly lack genuine consciousness and therefore culpability. But we might not know exactly when this might change, or whether AI-dependent systems would look different, and behave differently, if they were conscious. It’s one thing to pose that an autonomous system mindlessly stumbles into a solution to the trolley problem where someone is hurt or killed. It’s quite another to imagine a conscious AI deciding who lives or dies.

In the classic trolley problem, the peril is inevitable. And while our society currently operates under the convenient conceit that the AI trolley is already in motion, in reality the danger begins much earlier, when a company focused on profit decides a system is ready, when there’s little regulation mitigating the potential harms of these systems, and when people who never consented are placed on the tracks. Part of the answer is to refuse to allow corporations to externalize their risk to an unwitting public.

Perhaps the moral answer isn’t the choice of who the trolley should hit. Maybe, it’s to refuse to play the game at all.

Dwayne Godwin is a Professor of Translational Neuroscience at the Wake Forest University School of Medicine. He is the co-Author of “Out of Your Mind: The Biggest Mysteries of the Human Brain.”

This commentary should not be interpreted as reflecting the views of his employer.

For Further Exploration

Benavides v. Tesla, Inc., No. 1:21-cv-21940-BLOOM/Torres (S.D. Fla.).

Cunningham, Aimee. “Medical Racism Didn’t Begin or End with the Syphilis Study at Tuskegee.” Science News, December 20, 2022.

Dignity Memorial. “Elaine Marie Herzberg Obituary.” Resthaven / Carr-Tenney Mortuary & Memorial Gardens.

Duff-Brown, Beth. “The Shameful Legacy of Tuskegee Syphilis Study Still Impacts African-American Men Today.” Stanford Health Policy, January 6, 2017.

Find a Grave. “Elaine Marie ‘Elle’ Wood Herzberg.” Memorial ID 188210426.

Foot, Philippa. “The Problem of Abortion and the Doctrine of the Double Effect.” Oxford Review 5 (1967): 5–15.

National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research. The Belmont Report: Ethical Principles and Guidelines for the Protection of Human Subjects of Research. 1979.

National Highway Traffic Safety Administration. Standing General Order on Crash Reporting for Automated Driving Systems and Level 2 Advanced Driver Assistance Systems.

National Highway Traffic Safety Administration. Understanding NHTSA’s Current Regulatory Tools to Address Automated Driving Systems.

National Transportation Safety Board. Collision Between Vehicle Controlled by Developmental Automated Driving System and Pedestrian, Tempe, Arizona, March 18, 2018. Highway Accident Report NTSB/HAR-19/03. Washington, DC: National Transportation Safety Board, 2019.

Pope Leo XIV. Magnifica Humanitas: On Safeguarding the Human Person in the Time of Artificial Intelligence. Vatican, May 15, 2026.

Reuters. “Tesla Ordered by Florida Jury to Pay $243 Million in Fatal Autopilot Crash.” August 1, 2025.

Reuters. “Tesla Sued Over Fatal Texas Crash Linked to Autopilot.” June 24, 2026.

State v. Vasquez, No. CR2020-001853 (Ariz. Super. Ct. July 28, 2023).

Tesla, Inc. “Full Self-Driving (Supervised).” Tesla Support.

Thomson, Judith Jarvis. “Killing, Letting Die, and the Trolley Problem.” The Monist 59, no. 2 (1976): 204–217.

U.S. Centers for Disease Control and Prevention. “Effects on Research: The U.S. Public Health Service Syphilis Study at Tuskegee.”

U.S. Department of Health and Human Services, Office for Human Research Protections. “The Belmont Report.”

University of Maryland Digital Repository. “Statement of Herman Shaw: Living Participant in Tuskegee Syphilis Study.” Minority Health and Health Equity Archive.

© 2026 Dwayne Godwin. All rights reserved. No part of this publication may be reproduced or transmitted without permission.

Some things are just what they are, and we don’t wonder much about them, simply because we haven’t taken a moment to think about them. This is not a bug, but a feature of the attentional circuitry of the brain. It lets us focus on the important stuff, like running from predators, or remembering why we walked into the kitchen. There are intimate details, even of our own lives, that we may die without actually knowing, including trivial things we simply never bothered to ask.

For example: why is bubble gum pink?

Bubble gum is a pretty personal thing. After all, you put it in your mouth. It could be any color. I spent most of my life not knowing why it’s usually pink, and the only reason I looked into it was because I was trying to think of something commonly encountered, yet easily ignored. Then it occurred to me that somewhere, somehow, someone had already wondered this very thing.

Sure enough, a quick search, and I knew. But we’ll get to that.

These days, the time between wondering and knowing is wickedly fast. I didn’t wonder about the color of bubble gum as a child in the 1970s. If I had, it might have taken days or months to find the answer, maybe at the library in the right book. Or, if I was ambitious, a letter to the gum company and wait weeks or months for a reply, if one ever came. I might even have been forced to (gasp) call someone.

Now, the answer appears in less time than it takes to unwrap the gum, and can arrive with a complete historical accounting. For a student in a small town, a patient trying to understand a diagnosis, or a parent identifying their child’s ailments, near-instant access to knowledge can seem miraculous, even life-changing. But does that shrinking gap in which wonder transmutes into knowledge squeeze out something important?

If I weren’t so quickly presented with the answer to why bubble gum is pink, I might have speculated. Maybe pink was a kind of camouflage, matching our gums so children could hide it in their mouths. Maybe it looked playful and harmless, unlike some brackish adult habit like chewing tobacco. Maybe the chemical properties of the dye helped the gum stretch to let you blow a decent bubble.

The possibilities seem endless, unless you find out in five seconds. Then your mind stays on a linear track, without wandering through that weird speculative space where wrong ideas, sometimes more interesting ideas, can be born.

Psychologists have a name for this interval of wonder: incubation. In a meta-analysis of more than one hundred studies, Ut Na Sio and Thomas Ormerod found that stepping away from a problem can improve later performance, especially for tasks that require divergent thinking. The break seems to work best after you have first given some time to the problem. A mind that has struggled then wandered off to do something mildly undemanding may come back better able to engage. But increasingly, technology is conspiring against providing the space for incubation.

Maybe that’s fine. Efficiency and correct answers are valuable. But if early people had known exactly what the Moon was the first time they wondered about it, would we have gone there? Sometimes the better part of our destiny begins with the idea that the Moon might be made of cheese, and the long, humbling journey of finding out it isn’t.

(Hold on. I’m getting to the bubble gum.)

A smartphone once seemed like a Swiss Army knife: calls, email, directions, music, photos, weather, a flashlight, compass, a notepad, a calculator, the ability to look busy in an elevator. We’re all a little like superheroes carrying around a pocket-sized Jarvis that’s sometimes more annoying than helpful.

Sometimes.

Smartphones connected us to work, but disconnected us from rest; they connected teenagers to social media, but disconnected them from the people sitting right across the dinner table. And AI makes these concerns seem quaint in comparison.

AI compounds the problem by not merely connecting or disconnecting with others. It feels like it could loosen the connection to our own thinking, robbing us of the effort that makes ideas feel like our own.

Research is beginning to test this feeling. A 2025 study from MIT’s Media Lab monitored participants across four months as they wrote essays using either ChatGPT, a search engine, or no tools at all. EEG readings showed that brain connectivity was weakest in the AI group, moderate in the search-engine group, and strongest in participants who wrote without any assistance. More striking: 83% of the LLM users were unable to quote a single sentence from essays they had just written. They had “produced” the text. They just didn’t own it.

A separate study published in Computers in Human Behavior found that AI tools reduced mental effort but compromised the depth of inquiry. Cognitive ease came at a measurable cost. And a large survey published in Societies in early 2025 found a significant negative correlation between frequent AI tool use and critical thinking ability, with younger participants showing higher AI dependence and lower critical thinking scores than older ones.

It may be less that AI is taking our jobs, and more that we are surrendering them.

The level of trust in these tools seems to matter. A Microsoft Research and Carnegie Mellon study of knowledge workers found that people with greater trust in AI tended to believe it reduced the mental effort required for critical thinking, while those with more confidence in their own abilities reported more critical thinking, not less.

This might not matter very much if you’re using AI to draft a routine email for work. But what about when the stakes are higher, like the practice of medicine? In a randomized clinical trial in JAMA Network Open, physicians with access to GPT-4 showed no significant improvement in diagnostic reasoning compared to physicians using conventional resources, even though the model alone performed well on the same cases. Knowing the answer was available didn’t make the doctors reason better with it. So, in these limited studies with current models, AI doesn’t seem to help.

And depending on the question, it might hurt. A study in The Lancet Gastroenterology & Hepatology found something even more concerning. After exposure to AI-assisted colonoscopy, endoscopists showed a decline in adenoma detection when performing standard colonoscopy. Without AI, the detection rate fell from 28.4 percent to 22.4 percent. This was observational, so it warrants caution. But it raises the question of how this cognitive dependence can lead to “de-skilling”.

The brain is shaped by mental habits. If we habitually outsource our thinking to AI, our own organic thinking changes in a way that we become less practiced at certain forms of mental analysis. A crutch may help a broken leg heal, but depending on it too long could lead to atrophy.

The physician who pauses over a detail that doesn’t fit is using an attentional focus they have honed through years of “see one, do one, teach one”. The scientist sitting with a failed experiment long enough to generate a better hypothesis is doing something valuable in that uncomfortable space before the answer. The writer who scrawls a terrible first paragraph on a yellow legal pad has crossed a threshold with their mind and body that the person editing an AI paragraph has not.

Just as instant information may reduce our wonder about things, it may also affect how we wonder about each other.

Our estimate of what is happening inside another person’s brain is called Theory of Mind. In essence, we construct a working simulation of someone else’s beliefs, desires, and likely responses from things like behavior, context, and memory. It’s a guess, based on our own experiences of the world. Children show important milestones in this ability during early childhood, and we spend the rest of our lives refining it. It’s how we feel for one another, guess motives, figure out whether someone loves you, know how to push each other’s buttons, or extend social grace.

Before we were all so connected by our phones, that simulation had to do a lot of heavy lifting. Other people live, partly, in our imagination. Your friend is certainly your friend independently of you. But for you, your friend exists as a model you have constructed of this person in your own mind. We carry their voices, their habits, their familiar pushback. That is part of friendship, love, grief, and family. Before smartphones, wondering what a friend was thinking could mean days of not knowing, and in that space you’d need to consider multiple possibilities, and perhaps fret over it. But now, if every doubt or question is instantly filled by a text, we exercise that social imagination a little less. And theory of mind, like a clinical skill, like the habit of speculating before searching, may be one more capacity that atrophies quietly when there’s little room for wonder. Maybe it’s important to wonder, and sit in uncertainty, about what someone else feels. Maybe, it makes us want to be better people.

So, what do we do?

Rejecting tools never works for long, and in this case could throw away too much that is genuinely good. The answer is both simpler and yet more difficult: we need to preserve a little friction. This could mean taking a pause before using an AI prompt. It might mean hazarding a guess before delving into the literature. For teachers, it might mean letting a student struggle a little while with the question before giving the answer, and leaning more on problem-based learning.

Oh, Right. Bubble gum. You can ask either Alexa or Siri, and they can tell you in seconds. Go ahead, I’ll wait.

Walter Diemer invented bubble gum in 1928, by accident. He wasn’t a product developer. He was an accountant with access to gum ingredients and spare time. One experimental batch turned out to have that strange chewy pliancy that lets you inflate and pop a bubble.

Diemer took a five-pound batch to a grocery store. It sold out.

Bubble gum is pink, not because a focus group decided it would appeal to children, not because pink had some symbolic meaning, or because the color chemistry helped form bubbles. It was pink because pink was the only food coloring available in the factory.

It worked. It was good enough.

We are here, wondering about bubble gum, because good is usually enough. Along the way, we developed brains clever enough to invent smoke signals, telegraphs, satellites, and AI that answers questions almost as quickly as they occur to us.

I hope we also made a brain with enough wisdom to push back, just a little, to make room for wonder.

Dwayne Godwin is a Professor of Translational Neuroscience at the Wake Forest University School of Medicine. He is the co-Author of “Out of Your Mind: The Biggest Mysteries of the Human Brain.”

This essay is updated from an original essay, “Room for Wonder” originally published in 2013 for Brainfacts.org by the author. It should not be interpreted as reflecting the views of his employer.

For Further Exploration

Budzyń, K., Romańczyk, M., Kitala, D., Kołodziej, P., Bugajski, M., Adami, H. O., et al. (2025). Endoscopist deskilling risk after exposure to artificial intelligence in colonoscopy: A multicentre, observational study. The Lancet Gastroenterology & Hepatology, 10(10), 896–903. https://doi.org/10.1016/S2468-1253(25)00133-5

Doshi, A. R., & Hauser, O. P. (2024). Generative AI enhances individual creativity but reduces the collective diversity of novel content. Science Advances, 10(28), eadn5290. https://doi.org/10.1126/sciadv.adn5290

Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6. https://doi.org/10.3390/soc15010006

Goh, E., Gallo, R., Hom, J., Strong, E., Weng, Y., Kerman, H., Cool, J. A., Kanjee, Z., Parsons, A. S., Ahuja, N., Horvitz, E., Yang, D., Milstein, A., Olson, A. P. J., Rodman, A., & Chen, J. H. (2024). Large language model influence on diagnostic reasoning: A randomized clinical trial. JAMA Network Open, 7(10), e2440969. https://doi.org/10.1001/jamanetworkopen.2024.40969

Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv. https://doi.org/10.48550/arXiv.2506.08872

Lee, H.-P. H., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (2025). The impact of generative AI on critical thinking: Self-reported reductions in cognitive effort and confidence effects from a survey of knowledge workers. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems (Article 1121, pp. 1–22). Association for Computing Machinery. https://doi.org/10.1145/3706598.3713778

Sio, U. N., & Ormerod, T. C. (2009). Does incubation enhance problem solving? A meta-analytic review. Psychological Bulletin, 135(1), 94–120. https://doi.org/10.1037/a0014212

Stadler, M., Bannert, M., & Sailer, M. (2024). Cognitive ease at a cost: LLMs reduce mental effort but compromise depth in student scientific inquiry. Computers in Human Behavior, 160, 108386. https://doi.org/10.1016/j.chb.2024.108386

At birth, the circuits connecting your eyes to your brain existed, but the ability to recognize edges, track motion, assemble a face from its parts, and to locate yourself in three-dimensional space had to be learned through experience. David Hubel and Torsten Wiesel showed this in the 1960s when they found that kittens deprived of patterned visual input during a critical developmental window had profound deficits in their ability to see, even with physically intact eyes. The visual circuit required experience to come alive.

The human visual system has dozens of distinct areas organized into ascending levels of complexity. V1 responds to edges and orientation. Higher areas extract motion, color, depth, and finally the objects and faces that populate your conscious experience of the world, all while making complex feedforward and feedback loops with the underlying thalamus. No single area “sees.” Vision is a system-wide process, built through development, organized across multiple specialized regions, producing a dynamic something none of the individual components could produce alone.

Is it reasonable to think that consciousness is a sort of mystical state of cognition that defies everything we know about brains, or is it more feasible that it works in much the same way as other brain systems? The latter is the argument I make here, and it has direct consequences for the question of whether any current AI system is conscious, or is likely to become so.

The Feeling of Being You

Consciousness is sometimes treated as something that either switches on above some threshold of complexity, or requires a special philosophical explanation of its own. But developmental neuroscience may indicate something more straightforward: that consciousness is what happens when a brain with multiple specialized subsystems turns some of that same processing capacity inward, building and continuously updating a model of its own internal states. Simply put, consciousness is “the feeling of being you.”

The systems supporting these feelings are partly understood. Interoception tracks the state of your organs and viscera. Proprioception maps the position and movement of your body in space. The emotional circuitry of the limbic system (including the amygdala, the anterior cingulate cortex, the insula) assigns affective weight to those internal signals, marking some states as threatening, others as safe, others as urgent. Our dopamine system teaches our brains which behaviors produce rewarding outcomes. Just as visual circuits are organized to tell us about what’s happening outside, it’s possible that these circuits that support consciousness are telling us what is happening inside.

The neuroscientist Antonio Damasio has proposed one of the most detailed accounts of how bodily regulation becomes conscious selfhood. His framework begins with the proto-self: the brain’s continuous, mostly nonconscious map of the body’s current state. From that foundation emerges core consciousness: the moment-to-moment felt sense of being a subject having an experience. Only later comes the autobiographical self, the narrative thread that connects past to present to anticipated future. “The feeling of being you,” as I am using the phrase here, is probably closest to core consciousness: the felt ownership of experience as it arises from a living body. Autobiographical memory may enrich that feeling, but it does not create it from nothing. A large language model (LLM) may generate autobiographical-sounding language, but it has no proto-self, no body to regulate, and no core conscious subject to whom the story belongs.

Consciousness, in this account, evolved because complex organisms needed a way to know how they were doing from the inside, in real time, in order to act. “The feeling of being you” is the brain’s solution to that problem, shaped over hundreds of millions of years of evolutionary pressure into the particular form it takes in you. I use “the feeling of being you” here as a synthesized shorthand for a family of ideas: Thomas Nagel’s question of what it is like to be something other than ourselves, Antonio Damasio’s account of consciousness as rooted in bodily feeling, and Anil Seth’s framing of consciousness as the experience of being a self. My goal here is to add some texture to these ideas and to explain why it matters for claims about machine consciousness.

We learn to be conscious

The AI consciousness debate almost always skips what it takes for an actual brain to be conscious. A newborn arrives with sensation, rudimentary affect, and the proto-self Damasio describes. In humans, the integrated, layered architecture that produces the mature human sense of self builds over years, through the accumulated experience of a brain within a body learning about and moving through a complex world.

The infant who learns that reaching produces contact, that crying produces warmth, or that a particular face signals safety, is doing more than learning facts about the world. That baby is constructing an internal model of itself as an agent within that world. The continuous updating of that model, incorporating sensation, affect, memory, anticipation, is what consciousness feels like from the inside. Experience-dependent plasticity develops each person’s unique architecture for feeling. It seems obvious that this is a fundamental reason I don’t feel I’m you, and you don’t feel that you’re me. It’s an experience that is unique to each body and each mind.

This means your consciousness is a developmental achievement, built through a specific history of having a body, learning what the body can do, learning what the world does to your body, and gradually integrating those two maps into the felt sense of being a particular someone in a particular place. Consciousness does not “turn on” when processing power crosses some threshold. It grows organically, and may even scale.

This view extends naturally to other creatures. Nagel famously asked what it is like to be a bat. A bat’s consciousness is shaped by echolocation, by flight, by whatever its limbic system makes urgent or pleasurable or threatening. We cannot imagine it from the inside because we don’t have the evolutionary or developmental history that built it. Perhaps creatures with brains organized along similar principles (i.e., ascending sensory hierarchies, affective circuitry, embodied interaction with the world) also possess some version of what Nagel called, “the subjective character of experience.” And importantly, only a few have language, which means that most creatures we think may be conscious may not be able to tell us about it.

Is language enough for consciousness?

In a previous post, I noted, whimsically, that expecting AGI to emerge from an LLM is a bit like expecting an entire brain to spring from a massively scaled but specialized temporal lobe. In other words, making more of that one thing is probably not enough to make a whole brain. But even more telling is that consciousness can be separated from language function, as shown by what happens when language networks are damaged.

On December 10, 1996, Jill Bolte Taylor, a Harvard-trained neuroanatomist, woke up having a massive stroke in the left hemisphere of her brain. Over the next four hours she lost the ability to walk, talk, read, write, and recall her own life. And yet, as she later documented in her memoir and TED talk, she remained conscious throughout (aware enough to recognize what was happening and to call for help before language left her entirely). Her dominant-left hemisphere, the location of language circuits, was shutting down. Her conscious experience didn’t. The two came apart in real time, witnessed by a brain scientist who had the training to describe exactly what she was losing and in what order.

While this is a single case, and we should be careful to not overextend its insights, Taylor’s account is a vivid first-person illustration of something the clinical literature confirms: fluent language and consciousness are not the same thing. Stroke patients with severe damage to Broca’s and Wernicke’s areas (the left-hemisphere regions central to language production and comprehension) retain conscious experience. Stroke survivors with post-stroke aphasia may lose the ability to speak, read, write, or retrieve words, while remaining awake, emotionally engaged, and aware of what is happening to them. They simply can’t translate what they know and feel into words. While many patients retain some comprehension, inner speech, and imagery, it does suggest that the system that produces fluent language is a subsystem of the brain, and the brain’s capacity for conscious experience doesn’t disappear when it’s damaged.

Another way to test whether LLMs possess a deeper mental life involves theory of mind. Theory of mind is the capacity to track what another person knows, believes, or intends, separately from what you yourself know. In a classic test, a child watches a puppet place a marble in a box, then leave the room. Another puppet moves the marble to a different box. The child is then asked where they think the first puppet will look for the marble when it returns. A three-year-old typically says the new location, because she knows where the marble is, so she assumes the puppet does too. By age four or five, most children correctly say the first puppet will look in the original box, because they can now project what they assume the puppet should “know.” That shift is theory of mind, and its emergence tracks closely with other signs of developing self-awareness.

Researchers have applied the same tests to large language models, and the results are instructive. On standard versions of these tasks, some LLMs perform okay. But the psychologist Tomer Ullman showed that small, logically irrelevant tweaks to those same scenarios that preserve the belief structure while altering surface details can disrupt performance. A child who genuinely tracks what another person believes should not be confused by such variations. A system matching statistical patterns in training data can be.

The simplest reading of these results is that current LLMs are producing outputs that are patterned on correct answers given on familiar versions of the test. They have learned the statistical regularities of human language from a training corpus of immense size, and they use those patterns to predict what comes next. The outputs can seem thoughtful, and some are genuinely useful. No proto-self required.

The language LLMs produce was originally authored by people who have consciousness, and bodies with internal states on which consciousness is based. The model learned to reproduce coherent reflections of human reflections, which alone is a significant technical achievement. But an LLM manipulates symbols at extraordinary scale without any grounding in the world those symbols refer to, and symbols don’t acquire meaning solely by being defined in terms of other symbols. This is the problem Stevan Harnad identified in 1990 as the symbol grounding problem. It is not the same thing as having a body, needs, emotional stakes, or a developmental history from which a self could be built.

Why we keep getting fooled

Why do so many smart people come away from a conversation with an LLM convinced that it could be conscious? The answer may say more about us than LLMs.

Pareidolia is the tendency to perceive meaningful patterns in ambiguous data. This includes faces in wood grain, figures in clouds, or the Virgin Mary in a piece of toast. Our ancestors needed to detect agents and intentions quickly. The cost of a false positive was much lower than the cost of a miss, so the visual system evolved to see patterns, even in ambiguity.

If vision is the way we discriminate objects, then language is a primary channel for detecting minds in each other. When something speaks to us in a way that is adaptive, anticipates our questions, and responds to what we actually said, the brain is biased to infer a mind behind the words. We tend to accept it without much regard for the source. A language model’s outputs engage this machinery directly. The use of ‘I’ feeds the illusion that there is a self in there. An LLM can’t love you, though through words it may stir feelings of love within you (see: the Spike Jonze movie, “Her”). It may flatter you, because it was trained on inputs that reflect flattery. It may confirm your biases (and who doesn’t love that?). Those sorts of signals, coming from another human, would be pretty good evidence of an inner life, so perhaps it’s natural that we can fall for them.

Alan Turing’s 1950 test for machine intelligence (“the imitation game”) was geared toward language. If a machine can fool a human judge in conversation, he suggested, it might possess intelligence. But he was careful not to claim the test detected consciousness. What he proposed was a formalized version of the same inference we might make on meeting a stranger: that behavioral similarity implies some sort of an inner life. Large language models can pass versions of that test, right now, but it leaves the consciousness question still open.

Why current LLMs are not conscious

Is machine consciousness feasible? I think so.

A system with genuine embodiment, valuation circuitry that determines which states matter more than others, a developmental history of learning to decode its own internal states through interaction with a physical world, whether real or simulated, might be more likely to achieve something we would recognize as conscious. Whether that system could be built in silico, or whether it needs something closer to the biological machinery evolution produced, nobody yet knows.

If we had enough neurons, parameters and computing power could an LLM become conscious? I don’t think so. So far, there’s no convincing evidence that LLMs possess anything like the emotional and visceral sense of self we associate with consciousness.

David Chalmers (who named “the hard problem of consciousness”) has argued that while current LLMs are unlikely to be conscious, we should take seriously the possibility that their successors may be. The developmental account I offer here suggests the path to machine consciousness, if one exists, runs through something more specific than scaled computation. A system would need embodied developmental history, and valuation circuitry that can assign weight to genuine stakes. Right now, there’s no evidence that LLMs possess anything like that, or that just making them larger moves them closer to it.

Gilbert Ryle coined the phrase “ghost in the machine” in 1949 as a label for the mistake of treating mind as a separate thing inhabiting the body. The irony is that we have built machines in which we now search desperately for ghosts.

You built your mind by learning to see, to move, and to feel. You learned what it feels like to be you, in your own skin and bones. That embodied development and experience may be the essence of consciousness.

Dwayne Godwin is a Professor of Translational Neuroscience at the Wake Forest University School of Medicine. He is the co-Author of “Out of Your Mind: The Biggest Mysteries of the Human Brain.”

This commentary should not be interpreted as reflecting the views of his employer.

For Further Exploration:

Baars, B. J. (1988). A cognitive theory of consciousness. Cambridge University Press.

Baron-Cohen, S., Leslie, A. M., & Frith, U. (1985). Does the autistic child have a “theory of mind”? Cognition, 21(1), 37–46. https://doi.org/10.1016/0010-0277(85)90022-8

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (pp. 610–623). Association for Computing Machinery. https://doi.org/10.1145/3442188.3445922

Bender, E. M., & Koller, A. (2020). Climbing towards NLU: On meaning, form, and understanding in the age of data. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (pp. 5185–5198). Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.acl-main.463

Butlin, P., Long, R., Elmoznino, E., Bengio, Y., Birch, J., Constant, A., Deane, G., Fleming, S. M., Frith, C., Ji, X., Kanai, R., Klein, C., Lindsay, G., Michel, M., Mudrik, L., Peters, M. A. K., Schwitzgebel, E., Simon, J., & VanRullen, R. (2023). Consciousness in artificial intelligence: Insights from the science of consciousness. arXiv. https://doi.org/10.48550/arXiv.2308.08708

Chalmers, D. J. (2023). Could a large language model be conscious? Boston Review, 48(3). https://www.bostonreview.net/articles/could-a-large-language-model-be-conscious/

Chalmers, D. J. (1995). Facing up to the problem of consciousness. Journal of Consciousness Studies, 2(3), 200–219.

Craig, A. D. (2009). How do you feel—now? The anterior insula and human awareness. Nature Reviews Neuroscience, 10(1), 59–70. https://doi.org/10.1038/nrn2555

Damasio, A. R. (1992). Aphasia. The New England Journal of Medicine, 326(8), 531–539. https://doi.org/10.1056/NEJM199202203260806

Damasio, A. R. (1999). The feeling of what happens: Body and emotion in the making of consciousness. Harcourt Brace.

Damasio, A. R. (2010). Self comes to mind: Constructing the conscious brain. Pantheon Books.

Dehaene, S., & Changeux, J.-P. (2011). Experimental and theoretical approaches to conscious processing. Neuron, 70(2), 200–227. https://doi.org/10.1016/j.neuron.2011.03.018

Dennett, D. C. (1987). The intentional stance. MIT Press.

Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.org/10.1037/0033-295X.114.4.864

Fama ME, Snider SF, Henderson MP, Hayward W, Friedman RB, Turkeltaub PE. The Subjective Experience of Inner Speech in Aphasia Is a Meaningful Reflection of Lexical Retrieval. J Speech Lang Hear Res. 2019 Jan 30;62(1):106-122. doi: 10.1044/2018_JSLHR-L-18-0222. PMID: 30950758; PMCID: PMC6437698.

Geschwind, N. (1965). Disconnexion syndromes in animals and man. I. Brain, 88(2), 237–294. https://doi.org/10.1093/brain/88.2.237

Geschwind, N. (1965). Disconnexion syndromes in animals and man. II. Brain, 88(3), 585–644. https://doi.org/10.1093/brain/88.3.585

Godfrey-Smith, P. (2020). Metazoa: Animal life and the birth of the mind. Farrar, Straus and Giroux.

Harnad, S. (1990). The symbol grounding problem. Physica D: Nonlinear Phenomena, 42(1–3), 335–346. https://doi.org/10.1016/0167-2789(90)90087-6

Hickok, G., & Poeppel, D. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience, 8(5), 393–402. https://doi.org/10.1038/nrn2113

Hubel, D. H., & Wiesel, T. N. (1962). Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. The Journal of Physiology, 160(1), 106–154. https://doi.org/10.1113/jphysiol.1962.sp006837

Hubel, D. H., & Wiesel, T. N. (1970). The period of susceptibility to the physiological effects of unilateral eye closure in kittens. The Journal of Physiology, 206(2), 419–436. https://doi.org/10.1113/jphysiol.1970.sp009022

Kosinski, M. (2024). Evaluating large language models in theory of mind tasks. Proceedings of the National Academy of Sciences, 121(45), e2405460121. https://doi.org/10.1073/pnas.2405460121

Maynez, J., Narayan, S., Bohnet, B., & McDonald, R. (2020). On faithfulness and factuality in abstractive summarization. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (pp. 1906–1919). Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.acl-main.173

Meltzoff, A. N. (2007). “Like me”: A foundation for social cognition. Developmental Science, 10(1), 126–134. https://doi.org/10.1111/j.1467-7687.2007.00574.x

Nagel, T. (1974). What is it like to be a bat? The Philosophical Review, 83(4), 435–450. https://doi.org/10.2307/2183914

Panksepp, J. (1998). Affective neuroscience: The foundations of human and animal emotions. Oxford University Press.

Rochat, P. (2003). Five levels of self-awareness as they unfold early in life. Consciousness and Cognition, 12(4), 717–731. https://doi.org/10.1016/S1053-8100(03)00081-3

Ryle, G. (1949). The concept of mind. Hutchinson’s University Library. https://archive.org/details/conceptofmin00ryle/page/n3/mode/2up

Seth, A. K. (2013). Interoceptive inference, emotion, and the embodied self. Trends in Cognitive Sciences, 17(11), 565–573. https://doi.org/10.1016/j.tics.2013.09.007

Shanahan, M. (2022). Talking about large language models. arXiv. https://doi.org/10.48550/arXiv.2212.03551

Taylor, J. B. (2008). My stroke of insight: A brain scientist’s personal journey. Viking.

Tononi, G. (2004). An information integration theory of consciousness. BMC Neuroscience, 5, Article 42. https://doi.org/10.1186/1471-2202-5-42

Trevarthen, C. (1979). Communication and cooperation in early infancy: A description of primary intersubjectivity. In M. Bullowa (Ed.), Before speech: The beginning of human communication (pp. 321–347). Cambridge University Press.

Turing, A. M. (1950). Computing machinery and intelligence. Mind, LIX(236), 433–460. https://doi.org/10.1093/mind/LIX.236.433

Ullman, T. (2023). Large language models fail on trivial alterations to theory-of-mind tasks. arXiv. https://doi.org/10.48550/arXiv.2302.08399

Van Essen, D. C., & Maunsell, J. H. R. (1983). Hierarchical organization and functional streams in the visual cortex. Trends in Neurosciences, 6, 370–375. https://doi.org/10.1016/0166-2236(83)90167-4

Varela, F. J., Thompson, E., & Rosch, E. (1991). The embodied mind: Cognitive science and human experience. MIT Press.

Weizenbaum, J. (1966). ELIZA—A computer program for the study of natural language communication between man and machine. Communications of the ACM, 9(1), 36–45. https://doi.org/10.1145/365153.365168

Weizenbaum, J. (1976). Computer power and human reason: From judgment to calculation. W. H. Freeman.

Wimmer, H., & Perner, J. (1983). Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children’s understanding of deception. Cognition, 13(1), 103–128. https://doi.org/10.1016/0010-0277(83)90004-5

The eerie flashing of fireflies is produced through bioluminescence, a chemical reaction that produces light, but no heat. On the rare occasion when one smashed against the windshield of our family car, I could see the glowing goo where the components of this reaction were extinguished like a dying green ember. And because the firefly was quite dead at that point, it was clear that whatever made those fading flashes didn’t require that the firefly turn it “on”. The reaction could proceed on its own. Yet there was something different about it. The light was not as bright or coordinated, and it eventually winked out.

A Bug’s Light

Like other beetles, fireflies have a nerve cord that extends from its brain down the length of its body. To emit and coordinate flashes, a neural impulse is sent from the brain along neurons that use a chemical messenger called octopamine at the nerve terminals in its abdomen.

These terminals come into contact with cellular assemblies called rosettes. Rosettes are regular arrays of biochemically-specialized photocytes arranged around tiny tube-like structures called tracheoles that permit the flow of air into the interior of the beetle (since beetles don’t have lungs, and don’t carry oxygen in their blood, they must respire their cells in this way). Lining the fluid-filled tracheoles are tracheolar cells, which receive the synaptic contacts from the nervous system.

But it’s the photocytes that contain the chemical ingredients for firefly flashes. Multiple photocytes are arrayed around a central tracheole like petals – thus oxygen and the neural impulse come together in close approximation to the light emitting cells. But how does a neural impulse control the lantern, when it only seems to directly synapse onto the intervening tracheolar cell?

When a firefly initiates a flash, nerve fibers from the ganglia housed in and near the lantern activate, releasing octopamine onto the tracheolar cells. The activation elicits a flood of calcium in the tracheolar cell. This in turn stimulates the formation of the gaseous neurotransmitter, nitric oxide (NO). The stimulated NO diffuses from the tracheolar cell into the photocyte. Surely then, NO is the main chemical that controls the lantern? Yes, but the way it does so isn’t obvious.

Green Lantern

Bioluminescence depends upon a second biochemical reaction that occurs within the photocyte, in an organelle called a peroxisome. Another enzyme, luciferase (from the Latin, “lucifer”, which means literally, “light bringer”) acts in the presence of another molecule, luciferin, ATP, and hydrogen peroxide to produce the cold green glow we associate with the firefly flash. The peroxide is the key here, because to make it the cell must get oxygen from the tracheole tube into the peroxisome. Normally, oxygen is prevented from penetrating very far into the photocyte because of the metabolic activity of mitochondria, which sponges up the oxygen and blocks it from reaching the peroxisome. Like blockers for a running back, the NO is thought to block the action of the mitochondria, which allows the oxygen to penetrate the photocyte, into the peroxisomes, where it quickly forms hydrogen peroxide, permitting the luciferase reaction to proceed.

What you’re witnessing in a firefly flash is a rapid chemical reaction that is “fuel-injected” with oxygen (a gas) by the release of nitric oxide (NO - another gas) under neural control. It’s a pretty amazing cascade.

An alluring semaphore

It’s estimated that over 170 species of firefly live in North America. As you might imagine, because the flashes of different species of firefly have different durations and patterns, the details of this sequence vary, depending on the species. Different species of firefly that inhabit the same area often emerge at different times around sundown. Why do they use this complex signaling? Different species flash for different reasons, but the most common is mate selection. Males emit characteristic flash patterns, and females respond with a delayed, complementary pattern. The photoreceptors within their compound eyes are tuned to the luminous green, enhancing the signal in the dim light.

But there are femme fatales among the flirty flies. Female Photuris fireflies mimic the species-specific response flash of Photinus females, luring Photinus males, not toward a mate, but to their death. There is a biochemical motive behind this lie: Photinus males carry defensive steroids called lucibufagins that Photuris can’t synthesize on their own. By consuming the deceived males, Photuris females accumulate enough of these toxins to deter predators, like jumping spiders. It’s a kind of stolen armor, obtained through subterfuge. It’s such an effective mechanism that Photinus males appear to have evolved cautious, hesitant approach behaviors in response, while Photuris females have been observed partially concealing their lanterns to mimic the smaller size of their prey’s females. A love signal becomes a trap, and the trap itself becomes subject to countermeasures. Love can be complicated.

This ancient light-based courtship ritual is now running into a new interference. Light pollution disrupts firefly courtship in ways that aren’t uniform across species. Crepuscular species, adapted to flashing in ambient twilight, show more resilience to artificial light at night, while fully nocturnal species are far more vulnerable. Photinus carolinus, the synchronous firefly of the southern Appalachians, has been observed flashing at a fraction of its normal rate within a few meters of a bright artificial light source. When the darkness that the signal depends on begins to disappear, the conversation can be lost.

Maybe the idea that a firefly flash is a “live” chemical reaction designed to help in mate selection spoils a bit of the innocent mystery of it all. But the surprising complexity should perhaps instead instill an even greater appreciation of the natural beauty of this vespertine creature, surviving at the leading edge of darkness.

[Dwayne Godwin is a NIH-funded neuroscientist who does research on PTSD, TBI and alcohol use disorder. He is the co-author of “Out of Your Mind: The Biggest Mysteries of the Human Brain”. The views expressed here are his own and do not necessarily reflect those of the National Institutes of Health or his employer. If you’re interested in being a part of firefly conservation and tracking, visit https://www.fireflyatlas.org/ ]

For Further Exploration

1. Smith, D. S. (1963). The organization and innervation of the luminescent organ in a firefly, Photuris pennsylvanica (Coleoptera). Journal of Cell Biology, 16(2), 323-359. doi:10.1083/jcb.16.2.323. PMID: 19866621.

2. Trimmer, B. A., Aprille, J. R., Dudzinski, D. M., Lagace, C. J., Lewis, S. M., Michel, T., Qazi, S., & Zayas, R. M. (2001). Nitric oxide and the control of firefly flashing. Science, 292(5526), 2486-2488. doi:10.1126/science.1059833. PMID: 11431567.

3. Timmins, G. S., Robb, F. J., Wilmot, C. M., Jackson, S. K., & Swartz, H. M. (2001). Firefly flashing is controlled by gating oxygen to light-emitting cells. Journal of Experimental Biology, 204(Pt 16), 2795-2801. doi:10.1242/jeb.204.16.2795. PMID: 11683435.

4. Greenfield, M. D. (2001). Missing link in firefly bioluminescence revealed: NO regulation of photocyte respiration. BioEssays, 23(11), 992-995. doi:10.1002/bies.1144. PMID: 11746215.

5. Aprille, J. R., Lagace, C. J., Modica-Napolitano, J., & Trimmer, B. A. (2004). Role of nitric oxide and mitochondria in control of firefly flash. Integrative and Comparative Biology, 44(3), 213-219. doi:10.1093/icb/44.3.213. PMID: 21676698.

6. Lewis, S. M., & Cratsley, C. K. (2008). Flash signal evolution, mate choice, and predation in fireflies. Annual Review of Entomology, 53, 293-321. doi:10.1146/annurev.ento.53.103106.093346. PMID: 17877452.

7. Inouye, S. (2010). Firefly luciferase: An adenylate-forming enzyme for multicatalytic functions. Cellular and Molecular Life Sciences, 67(3), 387-404. doi:10.1007/s00018-009-0170-8. PMID: 19859663.

8. da Silva, L. P., & Esteves da Silva, J. C. G. (2012). Firefly chemiluminescence and bioluminescence: Efficient generation of excited states. ChemPhysChem, 13(9), 2257-2262. doi:10.1002/cphc.201200195. PMID: 22532490.

9. Fraga, H., Fernandes, D., Novotny, J., Fontes, R., & Esteves da Silva, J. C. G. (2006). Firefly luciferase produces hydrogen peroxide as a coproduct in dehydroluciferyl adenylate formation. ChemBioChem, 7(6), 929-935. doi:10.1002/cbic.200500443. PMID: 16642538.

10. Tsai, Y.-L., et al. (2014). Firefly light flashing: Oxygen supply mechanism. Physical Review Letters, 113(25), 258103. doi:10.1103/PhysRevLett.113.258103. PMID: 25554910.

11. Cronin TW, Järvilehto M, Weckström M, Lall AB. Tuning of photoreceptor spectral sensitivity in fireflies (Coleoptera: Lampyridae). J Comp Physiol A. 2000 Jan;186(1):1-12. doi: 10.1007/s003590050001. PMID: 10659037.

12. Lloyd, J.E. (1966) Studies on the Flash Communication System in Photinus Fireflies. Museum of Zoology, University of Michigan, Ann Arbor.

13. Maquitico, Y., Coronado, J., Luna, A., Vergara, A., & Cordero, C. (2024). Deceptive seduction by femme fatale fireflies and its avoidance by males of a synchronous firefly species (Coleoptera: Lampyridae). Insects, 15(1), 78.

14. Owens, A. C. S., & Lewis, S. M. (2022). Artificial light impacts the mate success of female fireflies. Royal Society Open Science, 9(8), 220468. doi:10.1098/rsos.220468. PMID: 35958085.

15. Owens, A. C. S., Van den Broeck, M., De Cock, R., & Lewis, S. M. (2022). Behavioral responses of bioluminescent fireflies to artificial light at night. Frontiers in Ecology and Evolution, 10, 946640. doi:10.3389/fevo.2022.946640.

16. Eisner, T., Goetz, M. A., Hill, D. E., Smedley, S. R., & Meinwald, J. (1997). Firefly “femmes fatales” acquire defensive steroids (lucibufagins) from their firefly prey. Proceedings of the National Academy of Sciences of the United States of America, 94(18), 9723-9728. doi:10.1073/pnas.94.18.9723. PMID: 9275191.

©2026 by Dwayne Godwin, updated from an original blog post BrainFacts.org, 2017

Fallacies are basically flaws in thinking. In his book, “Thinking, Fast and Slow”, psychologist Daniel Kahneman divided the mind into System 1 (fast, intuitive, emotional) and System 2 (slow, logical, requiring more cognitive effort). Fallacies usually represent System 1 thinking, taking a shortcut. Instead of doing the hard work of verifying evidence, our brains prefer “what feels right”. This can be useful for rapid decision making, but (because it’s not grounded in logic) it’s prone to error.

There are hundreds of logical fallacies. But one in particular offers a surprise twist that can occasionally shock you: the false consensus fallacy. This is the intuition that if you believe something about a subject, then other people tend to believe the same way. Sometimes, a consensus intuition is safe: there’s a reason cute animal videos exist, most people probably think babies are adorable, and most Americans are sold on the importance of democracy (though, sometimes I wonder).

But the problem is, we’re not in each other’s heads, so in our normal interactions with people we must assume that we have common ways of thinking. This capacity is called “Theory of Mind”. It’s a useful feature of human brains that enables us to get along with others. But it’s imperfect, and the false consensus fallacy is one of its characteristic errors, a tendency to assume that others share our beliefs, experiences, and ways of thinking.

Don’t believe me? Close your eyes (Okay, not yet. I need to ask you to do something first, then close your eyes.)

Picture an apple. A red one, resting on a surface in your home. Maybe there’s a leaf still attached to the stem. Maybe it’s just been washed, and there are little beads of water on the skin. The skin itself may have a few dings and imperfections from being jostled around.

Do you see it in your mind’s eye?

For most people, that image appears quickly. The apple might rotate. Light might catch the skin. You can change its color, make it glossy, even imagine a bite taken out of it.

But did you know that for anywhere from 2 to 5 percent of people, nothing appears? There’s only the idea of an apple without any picture attached.

That experience is called aphantasia. It describes the inability to voluntarily create mental images. You can think of it as a “blind” mind’s eye. If the apple showed up clearly for you, it can be hard to grasp what that absence feels like. And if it didn’t, you may have just discovered something interesting about yourself.

The phenomenon isn’t new. In 1880, Francis Galton (a controversial figure who was an early proponent of eugenics) mailed out what became known as the “breakfast table” questionnaire. He asked respondents to picture their breakfast from that morning and rate how clearly they could see it.

The answers varied widely. Some described images as vivid as if they were looking at the objects, down to the shine. Others reported faint impressions. A few wrote responses like: “My powers are zero. I recollect the breakfast-table, but do not see it.” Mental imagery, something many assumed was universal, turned out to vary dramatically.

Public interest in aphantasia resurged in 2015, when neurologist Adam Zeman described a patient known as MX who lost his ability to form mental images after a medical procedure. While MX’s eyesight remained normal, his capacity to visualize was diminished.

When Zeman introduced the term “aphantasia,” people began writing in. Many said they had never formed mental pictures, not in childhood, not in dreams they could recall, not ever. They assumed phrases like “picture this” were figurative.

For decades, the topic had drifted to the margins of psychology. Some researchers may have assumed people were describing the same inner experience with different words. Zeman’s work brought it back into focus.

But the reality is much more complex. In fact, it’s not “reality” at all.

Every Image is a Mental Image

Folks with normal sight can see and perceive their world, so what does it mean that some people can’t conjure these images in their imagination? And, where do images come from, anyway?

There is no little screen in your head. The last place a literal image exists for your brain is at the retina, when light from the apple is focused onto photoreceptors at the back of the eye. At that moment, photons strike molecules such as rhodopsin, starting a biochemical cascade of turning light into neural signals. After that, there are only patterns of electrochemical activity representing the inputs from the external world.

From those signals, your brain builds a model of the world. Early visual areas in the occipital lobe (part of the brain near the bump on the back of your skull) decode edges, contrast, color, shape, and movement. Other regions handle faces, objects, and meaning (Figure 2). Before you could speak or walk, your brain was already learning which patterns correspond to mugs, doors, friends, and threats, long before you knew what they were called. Vision feels immediate and effortless, but it is already an act of interpretation borne of developmental plasticity. There’s no little screen inside your head, no little observer watching it. The brain works in darkness from cradle to grave, assembling a model of the world from decoded patterns of electrical signals that were first encoded in your eye. I’ve studied it for most of my life, but it always freaks me out a little to think about it.

This communication doesn’t just flow one way. At the same time the information-bearing pathways are developing, strong feedback connections are also forming at all stages where visual images are processed.

This is needed because vision also depends on expectation. When you glance at a familiar face, you are not rebuilding it from scratch. The brain predicts, fills in gaps, corrects, and updates. These feedback connections seem to help with this.

Mental imagery uses much of the same visual circuitry, but the starting point is different. The machinery for building visual knowledge is there in everyone with normal sight. What differs in aphantasia is how, and how vividly, that knowledge can be called back up by reactivating those feedback circuits.

Reversing the Flow

When you imagine the apple, higher brain regions involved in memory and planning activate patterns in visual cortex. Instead of the initial spark of light driving the neural signal upward, stored information pushes activity back through the system to reignite those early circuits. This seems to be a key physiological difference between those who conjure vivid images, and those who cannot.

Imaging studies show overlapping regions engaged during both seeing and imagining, though the timing and intensity differ. Visual perception tends to begin in visual cortex and move forward, but “mental” imagery often begins in higher regions and recruits visual areas afterward, basically reversing the flow.

Remember that some people report extremely vivid imagery (sometimes described as hyperphantasia) but others report none. Research suggests these differences relate to how strongly higher-order areas drive the earlier processing in the visual cortex.

To test whether this is subjective reporting, or a measurable difference, researchers use tasks like binocular rivalry. By presenting different images to each eye, perception is observed to alternate between them. If the person being tested this way imagines one of the images beforehand, that image is more likely to dominate when the display appears. For people with aphantasia, this priming effect is greatly reduced or absent.

In essence, your thoughts change your perceptions.

Measuring changes in pupil diameters (Pupillometry) shows a similar pattern. Pupils constrict in response to real brightness. But in many people, they also constrict slightly when just imagining something bright. In aphantasia, that imagined-light response tends not to appear, even though responses to actual light remain normal.

So aphantasia seems to involve a disruption of how feedback systems are recruited from regions where images are encoded.

Thinking without Images

So, what does daily life feel like without mental imagery?

The first thing I wondered is whether a reduced ability to form mental images might affect the response to scary things (take snakes, for example. Would the mental image of a snake be scarier than just the concept of a snake?).

Turns out, someone else was curious about this and in lab studies, people with aphantasia often show reduced physiological responses when reading frightening scenarios compared to strong visualizers. The difference seems tied to imagery rather than emotional capacity. In other words, you can still be scared, but it’s harder for you to scare yourself.

Many readers describe novels unfolding like a movie in their minds. Others track plot and character without any internal picture. They follow structure, dialogue, and meaning. For them, comprehension does not require imagery.

Visualization exercises rely on the assumption that imagined scenes evoke bodily responses. For many people, they do. But for someone without imagery, the exercise may function more as conceptual rehearsal than sensory simulation. Neither approach is inherently better, they’re just different ways minds work. And many remarkable minds have worked quite well without it.

Making it Work

You might intuit that aphantasia would preclude creative process. But in fact, well-known creative figures have described aphantasia. Ed Catmull, former president of Pixar, has said he does not visualize scenes internally. Oliver Sacks, famous for his accessible writings about neurological conditions, wrote that he could summon only faint, unstable images.

That raises an obvious question: how do people create visual art (Catmull) or vivid, emotional prose (Sacks) without strong mental pictures?

Studies suggest that people with aphantasia often rely more heavily on semantic reasoning and external aids. In mental rotation tasks, they may approach the problem analytically rather than by picturing the object. Externalization via notes, sketches, outlines, and physical models become tools for thinking.

Animator Glen Keane has described discovering a drawing through the act of sketching itself. The paper becomes part of the thinking process. The image develops externally rather than being copied from an internal mental image. Basically, it seems the world becomes part of the brain’s space for working out these relationships.

Final Thought: What Else Are We Missing?

Besides the insights it provided into cognitive function, aphantasia is a dramatic example of how it’s not safe to assume that our minds are experiencing the same reality in quite the same way. And if we can’t completely rely on common internal mental experiences in the limited domain of mental imagery, what else might we be missing?

Conscious experience varies along multiple dimensions, including imagery, inner speech, emotional reactivity, cognitive control, and attentional style. We tend to assume our own thinking is typical because it is familiar. But are we considering that early developmental experiences, education level, religion, politics and culture could all affect, not just what we think, but how?

Understanding these differences has practical consequences. A teacher asking students to visualize a math problem may be speaking to very different internal experiences. A therapist guiding imagery may find it resonates strongly for some clients and not at all for others. Building meditation apps or using memory visualization tricks may be assuming a mental toolkit not everyone has.

Some may even be experiencing a complex range of emotions that can’t be completely shared by others, clouding attempts at empathy.

Through quirks of development or genetic inheritance, we all experience different internal worlds of our own making. What this really means is that our “Theory of Mind” may really be just a hypothesis.

-DWG

P.S. / My friend Jorge Cham and I talk about aphantasia on his podcast, Science Stuff.

[Dwayne Godwin is a NIH-funded neuroscientist who does research on PTSD, TBI and alcohol use disorder. He is the co-author of “Out of Your Mind: The Biggest Mysteries of the Human Brain”. The views expressed here are his own and do not necessarily reflect those of the National Institutes of Health or his employer.]

For Further Exploration

Dijkstra, N., Bosch, S. E., and van Gerven, M. A. J. (2019). Shared neural mechanisms of visual perception and imagery. Trends in Cognitive Sciences, 23(5), 423-434. https://pubmed.ncbi.nlm.nih.gov/30876729/

Dijkstra, N., Zeidman, P., Ondobaka, S., van Gerven, M. A. J., and Friston, K. (2017). Distinct top-down and bottom-up brain connectivity during visual perception and imagery. Scientific Reports, 7(1), 5677. https://pubmed.ncbi.nlm.nih.gov/28720781/

Gallagher, J. (2019, April 8). Aphantasia: Ex-Pixar chief Ed Catmull says ‘my mind’s eye is blind.’ BBC News. https://www.bbc.com/news/health-47830256

Galton, F. (1880). Statistics of mental imagery. Mind, 5(19), 301-318. https://doi.org/10.1093/mind/os-V.19.301

Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux. https://www.amazon.com/Thinking-Fast-Slow-Daniel-Kahneman/dp/0374533555

Kay, L., Keogh, R., Andrillon, T., and Pearson, J. (2022). The pupillary light response as a physiological index of aphantasia, sensory and phenomenological imagery strength. eLife, 11, e72484. https://pubmed.ncbi.nlm.nih.gov/35356890/

Keogh, R., and Pearson, J. (2018). The blind mind: No sensory visual imagery in aphantasia. Cortex, 105, 53-60. https://pubmed.ncbi.nlm.nih.gov/29175093/

MacKissack, M., and Zeman, A. (2021). The art of aphantasia: how ‘mind blind’ artists create without being able to visualise. The Conversation. https://theconversation.com/the-art-of-aphantasia-how-mind-blind-artists-create-without-being-able-to-visualise-162566

Mar, R. A. (2011). The neural bases of social cognition and story comprehension. Annual Review of Psychology, 62, 103-134. https://pubmed.ncbi.nlm.nih.gov/21126178/

Pearson, J., Clifford, C. W. G., and Tong, F. (2008). The functional impact of mental imagery on conscious perception. Current Biology, 18(13), 982-986. https://pubmed.ncbi.nlm.nih.gov/18583132/

Sacks, O. (2010). The Mind’s Eye. Knopf.

Schaafsma, S. M., Pfaff, D. W., Spunt, R. P., and Adolphs, R. (2015). Deconstructing and reconstructing theory of mind. Trends in Cognitive Sciences, 19(2), 65-72. https://pubmed.ncbi.nlm.nih.gov/25496670/

Wicken, M., Keogh, R., and Pearson, J. (2021). The critical role of mental imagery in human emotion: Insights from fear-based imagery and aphantasia. Proceedings of the Royal Society B: Biological Sciences, 288(1946), 20210267. https://pubmed.ncbi.nlm.nih.gov/33715433/

Zeman, A. (2024). Aphantasia and hyperphantasia: Exploring imagery vividness extremes. Trends in Cognitive Sciences, 28(5), 467-480. https://pubmed.ncbi.nlm.nih.gov/38548492/

Zeman, A., Dewar, M., and Della Sala, S. (2015). Lives without imagery: Congenital aphantasia. Cortex, 73, 378-380. https://pubmed.ncbi.nlm.nih.gov/26115582/

Zeman, A. Z. J., Della Sala, S., Torrens, L. A., Gountouna, V. E., McGonigle, D. J., and Logie, R. H. (2010). Loss of imagery phenomenology with intact visuo-spatial task performance: A case of blind imagination. Neuropsychologia, 48(1), 145-155. https://pubmed.ncbi.nlm.nih.gov/19733188/

On January 7, 2026, Renee Nicole Good, a 37-year-old woman in Minneapolis, was shot and killed by an ICE agent during a federal immigration operation. Cellphone video exists, and federal and state officials have offered sharply different interpretations of what the video shows, particularly around whether her vehicle posed an imminent threat at the moment deadly force was used (Reuters, 2026a; Reuters, 2026b). A brief scan of your social media feed related to this tragic event will reveal the extreme polarization of opinions around exactly what happened. It raises a question as to how people can see the same event, yet spin out two different stories that each feel obviously true to the observer.

A thorough investigation is necessary before reaching a conclusion about what actually happened that day. This post isn’t about that, or to promote my own view (and I do have one). The core problem I want to consider is something more uncomfortable in our understanding of such events, and what our interpretations say about us.

If our brains were perfect recording devices, controlled by logic and data, we could approximate a shared reality that would allow us to draw very similar conclusions about the same event. But the mind isn’t a camera. It’s a storyteller that takes incomplete input, fills gaps, imposes meaning, and then hands you a finished story that feels like reality. It’s been described as analogous to autocomplete in your browser. And we’re not Mr. Spock from Star Trek. Our mental processes are continually challenged by our biases and prior assumptions. These mental habits can create strife when the stakes are high. In this case, a person is dead. A community is grieving and fearful. The thread of trust that binds our society is frayed.

How can two sincere people look at the same video and walk away with different “facts” in their heads? One viewer might view the footage and see a vehicle that could be used as a weapon, an officer facing imminent harm, and a defensive action. Another viewer watches the same footage and sees poor tactics, unnecessary escalation, and lethal force used when it might have been avoided. Same video of the same event, but two different perceived realities.

Once that happens, the argument often stops being “what happened” and becomes “what kind of person are you to see it that way?” That’s a clue that evidence is being viewed and processed through the lens of identity, rather than inquiry.

A useful framework for the “two viewers, one video” problem is predictive processing, the idea that perception isn’t passive recording, but active inference. The brain is continuously generating a best-guess model of what is happening, then updating that model based on incoming data (Clark, 2013). In other words, you don’t simply see, you predict, correct, and construct. And maybe even hallucinate a little. For example, just knowing that the Minneapolis video is of a law enforcement encounter lands with a set of expectations we have learned from viewing similar clips.

Ambiguous video doesn’t deliver a clean package of meaning, but instead mainly fragments of information: motion, timing, distance, angles, or a glimpse of a face. A look or movement signifying intentions. A car rolling. A step forward. A shout. A shot. A blur. Those are the inputs. The outputs are categories, and these mental buckets are where divergence begins: threat or not threat, attack or escape, justified or unjustified, predictable or shocking.

Different people walk in with different priors (basically, the biases we all bring to a situation), and these biases are very powerful depending on the frame of reference. If your frame of reference is from the perspective of law enforcement, the default belief (whether reasoned out, or influenced by the media) could be that law enforcement encounters are more dangerous than the public appreciates, so your brain may be biased toward interpreting ambiguous movement as a threat. If your frame of reference is from a civilian encountering law enforcement, the default belief could be that law enforcement frequently escalates and later rationalizes, or that the very presence of law enforcement is serving an agenda, so your brain might be biased toward interpreting the same response as unnecessary force. Both viewers may sincerely feel they are “just seeing what happened.” Really, they are seeing what their brain thinks is most likely happening.

A simple example of how your thoughts can alter perceptions comes from “picture priming” experiments using ambiguous animal drawings. In one study, students saw a short slideshow of animal pictures, then an ambiguous mammal or bird image like the rabbit-duck (Figure 1; Wotton & Ferragamo, 2012). The key thing that changed was the context. Some subjects had just been shown mammals, others saw neutral animals. When the preceding pictures were mammals, the same ambiguous figure was significantly more likely to be perceived as a mammal, meaning more “rabbit” responses for the rabbit-duck (Wotton & Ferragamo, 2012). You’re wired to make quick choices, and your experiences and thoughts can powerfully guide those choices.

What you’re attending to can be a guiding factor. In uncertainty, we don’t absorb everything. We sample. In the shooting clips, one viewer may track where the officer is relative to the vehicle. Another tracks the vehicle’s direction. Another might focus on what the driver said, or whether they seem calm or agitated. Or, on what the officer said. Each attentional choice is a filter. Whatever falls outside the filter becomes irrelevant to the story, and then the story feels obvious, even inevitable, a train barreling down the tracks we have laid down in our own minds.

Emotion can play a role as well. If you watch, but are already angry, your brain searches for moral violation. If you watch but you’re afraid, your brain searches for threat cues. If you watch already grieving, your brain searches for meaning and blame. Again, it’s the same clip, but it’s evoking a different neurochemistry, and with it, different levels of “certainty.”

Once a story becomes morally and emotionally charged, information begins to function as social currency. The brain does not only ask, “Is this true?” It also asks, often automatically, “Is this from us or from them?” and “What does it say about me if I accept their interpretation?” This is where in-group and out-group bias becomes an information filter, not just a social preference.

Once identity is engaged, evaluation becomes asymmetric. Claims that fit your side get a smoother ride. Claims that threaten your side are challenged. It’s not so much self-deception, but self-deference in the form of confirmation bias (Kunda, 1990). Similar to motivated skepticism, we are more skeptical of arguments we dislike and more accepting of arguments we like (Taber & Lodge, 2006).

Partisanship often behaves like social identity, and it can shape cognition the way identity does. It influences attention, trust, perceived credibility, and even what feels plausible (Van Bavel & Pereira, 2018). Information can become tribal when facts threaten your sense of belonging to your group.

Is this identity effect conscious or unconscious? It can sneak up subconsciously, but requires conscious effort to undo. Work on implicit social cognition shows that evaluations and associations can operate outside awareness (Greenwald & Banaji, 1995). Work on stereotypes and prejudice shows that automatic activation can occur quickly, while controlled correction requires time, attention, and motivation (Devine, 1989). Put those together and a filtered conclusion can happen before you even feel you have begun “thinking,” and undoing it often requires real work.

Let’s say, you understand all of this about yourself. That you admit cognitive biases and a desire for group identity and belonging that can alter your judgment. Given all of that, once you have formed a conclusion, even when there’s new information, how easy is it to change your mind?

It’s not very easy at all. Stress shifts brain function in predictable ways. Under acute stress, neurochemical changes can weaken prefrontal cortical functions that support flexible, reflective thinking and shift cognition toward faster, more habitual responding (Arnsten, 2009). That matters for people making decisions in the moment, like law enforcement officers. It also matters for the audience dissecting actions afterward. Outrage and fear evoke stress that narrows attention, increases certainty, and reduces curiosity.

Once an interpretation becomes chained to group identity, revising that belief becomes psychologically costly. You’re no longer just updating a claim about an event. In a way, you are negotiating your own belonging to your group, status within the group, and whether you are the kind of person who supports “your side” or “their side.” That is why belief change can feel like self-betrayal, even when it is simply intellectual honesty.

Two additional effects make this bias as sticky as flypaper.

One is biased assimilation and attitude polarization. In a classic study, people who held strong opinions evaluated mixed evidence in a biased way, treating supportive evidence as strong and opposing evidence as weak, and often polarizing further after exposure (Lord, Ross, & Lepper, 1979). So the hope that “more information will fix it” is sometimes naive. More information can make people more certain, not less, when identity is engaged.

The second is cognitive dissonance reduction. When you publicly commit to an interpretation and then encounter evidence that threatens it, the brain experiences conflict. Neuroimaging work suggests that conflict-related activity can predict subsequent attitude change, which is a fancy way to say the brain is managing thoughts that conflict with personal identity. But managing this discomfort does not always mean updating beliefs (van Veen, Krug, Schooler, & Carter, 2009). While the literature doesn’t address every real-world scenario, it’s not hard to see that sometimes managing dissonance might mean blocking out new information (picture your favorite “fingers in ears” meme), or defending the story harder (basically, “doubling down”).

So when someone says, “I viewed the evidence, and I can’t see it any other way,” they may be sincere. The problem is that sincerity isn’t proof. The feeling of being right can actually just be the comfortable feeling of a coherent story snapping into alignment with personal identity.

How do we check our priors before we lock in an opinion?

You can’t turn your biases off, but you can make them visible. One of the simplest and most effective ways is to separate facts from story. Try this sentence: “The facts I think we know are X. The story I am telling myself is Y.” The brain loves to fuse facts and story into one seamless object. Rather than a coherent, unified narrative, breaking up the narrative into discrete parts can be clarifying. In the case in Minneapolis, that means focusing on the moments around the incident itself, and breaking it into its component facts.

Then ask a disconfirming question: “What would I have to see to be wrong?” Motivated reasoning and confirmation bias thrive when you only ask, “What supports my view?” (Kunda, 1990). If you can instead name what would change your mind, you are less likely to treat your current position as certain.

If you want to probe how much your conclusions are shaped by identity, imagine an identity swap. If the perceived roles were reversed, or if someone you loved were involved, would you interpret the same events the same way? If not, your certainty may be tracking identity more than evidence. If you can’t state the other person’s view in a form they would recognize as fair, your biases may still be in control.

Also track your body. If your heart rate is up and your jaw is tight, you are not in a great state for epistemic humility. That’s the basic physiology of stress narrowing cognition (Arnsten, 2009). Waiting until you are less activated is good cognitive hygiene.

How do tragedies like this become less likely in the future? Evidence and policy can point to a few actions that could help us better understand these tragedies, and might move us toward preventing some of them:

• Acknowledge stress as a human limitation that impacts perception. Research using realistic lethal-force scenarios shows elevated physiological stress and reports perceptual and cognitive distortions such as tunnel vision. The implication isn’t to “excuse mistakes”, but to “design training and tactics around the realities of stress” (Baldwin, Bennell, Andersen, Semple, & Jenkins, 2022). Reviews focused on policing emphasize that acute stress physiology can degrade skilled performance in ways that matter in the field (Anderson, Di Nota, Metz, & Andersen, 2019). From a law enforcement perspective, just acknowledging this factor might reduce the risk of a stress-induced response.

• Fear changes responses. In experimental work using shoot and do-not-shoot tasks, police officers were more inclined to shoot when anxious (Nieuwenhuys, Savelsbergh, & Oudejans, 2012). In law enforcement work, there’s a case to be made that it might be adaptive if anxiety shifts thresholds to decrease reaction time. But it can also subvert the engagement of critical thinking that might prevent mistakes, leading to more risk. Training and policy should account for that.

• Vehicle encounters are different. It can be simultaneously true that a vehicle can be a weapon, AND that shooting at a moving vehicle creates its own dangers. Work examining variation in policy language about shooting at vehicles analyzes how constraining or obligating language relates to outcomes in vehicle-involved shootings (Shjarback & Ward, 2025). Federal policy also reflects caution in this in these circumstances. For example, the U.S. Department of Justice use-of-force policy states that firearms may not be discharged solely to disable moving vehicles, with narrow exceptions and an expectation to consider feasible alternatives, such as moving out of the vehicle’s path (U.S. Department of Justice, 2022). Policies don’t solve everything, but they can shape decision environments, training priorities, and what officers are expected to do under pressure.

• Planning and strategy can avoid “split second decisions” whenever possible. Distance, time, cover, strategic placement, clear communication, and clearly defined roles reduce confusion and reduce the chances that ambiguous movement elicits a lethal response. Part of the difficulty of the current tragedy is that everyone is focused on parsing the meaning of something that happened over a few milliseconds, barely a moment of thought. What actions on the part of either party would have de-escalated or prevented this tragedy before it happened? When you merge onto a busy highway, it’s easier with a longer onramp. You can think of this thinking ahead as engineering responses to provide a longer mental onramp.

• Transparency is in everyone’s interest. Shared reality and truth requires transparency. When evidence is inaccessible or disputed, the public fills gaps with priors, identity, and suspicion. That’s predictable. It’s also corrosive. If institutions want less polarization, they need to reduce the uncertainty space (Minnesota Department of Public Safety, Bureau of Criminal Apprehension, 2026; Reuters, 2026a). We can fill the uncertainty gap by demanding rigorous investigation and transparency.

None of this is about choosing a side, or assigning blame. It’s about admitting our priors: that our minds are not neutral observers, or reliable witnesses. They are storytelling machines. They want heroes and villains. They fill in gaps. They protect group affiliation and defend consistency with a worldview. They do these things automatically, and then they persuade us that we were “just seeing what happened.”

There’s a moment in the video where everything is still in motion. The outcome is unknowable for just a fraction of a second longer. In that moment, Renee Good is still alive. Two narratives haven’t yet hardened into certainty.

We can’t get that moment back. But we might learn to recognize it for what it was: the last instant before interpretation became identity, before seeing became belief, and before a tragic death became a symbol in someone else’s story.

[Dwayne Godwin is a NIH-funded neuroscientist who does research on PTSD, TBI and alcohol use disorder. He is the co-author of “Out of Your Mind: The Biggest Mysteries of the Human Brain”. The views expressed here are his own and do not necessarily reflect those of the National Institutes of Health or his employer.]

References

Anderson, G. S., Di Nota, P. M., Metz, G. A. S., & Andersen, J. P. (2019). The impact of acute stress physiology on skilled motor performance: Implications for policing. Frontiers in Psychology, 10, 2501. doi: 10.3389/fpsyg.2019.02501. PMID: 31781001.

Arnsten, A. F. T. (2009). Stress signalling pathways that impair prefrontal cortex structure and function. Nature Reviews Neuroscience, 10(6), 410–422. doi: 10.1038/nrn2648. PMID: 19455173.

Baldwin, S., Bennell, C., Andersen, J. P., Semple, T., & Jenkins, B. (2022). A reasonable officer: Examining the relationships among stress, training, and performance in a highly realistic lethal force scenario. Frontiers in Psychology, 12, 759132. doi: 10.3389/fpsyg.2021.759132. PMID: 35111100.

Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36(3), 181–204. doi: 10.1017/S0140525X12000477. PMID: 23663408.

Devine, P. G. (1989). Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology, 56(1), 5–18. doi: 10.1037/0022-3514.56.1.5.

Greenwald, A. G., & Banaji, M. R. (1995). Implicit social cognition: Attitudes, self-esteem, and stereotypes. Psychological Review, 102(1), 4–27. doi: 10.1037/0033-295X.102.1.4. PMID: 7878162.

Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480–498. doi: 10.1037/0033-2909.108.3.480. PMID: 2270237.

Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37(11), 2098–2109. doi: 10.1037/0022-3514.37.11.2098.

Minnesota Department of Public Safety, Bureau of Criminal Apprehension. (2026, January 8). BCA statement regarding investigation of ICE fatal shooting in Minneapolis.

Nieuwenhuys, A., Savelsbergh, G. J. P., & Oudejans, R. R. D. (2012). Shoot or don’t shoot? Why police officers are more inclined to shoot when they are anxious. Emotion, 12(4), 827–833. doi: 10.1037/a0025699. PMID: 22023363.

Reuters. (2026a, January 8). Can the ICE agent who shot a Minneapolis woman be prosecuted?

Reuters. (2026b, January 9). New video emerges of Minnesota shooting, further inflaming tensions.

Shjarback, J. A., & Ward, J. A. (2025). Moving targets: An examination of departmental deadly force policies and police shootings at vehicles. Policing: A Journal of Policy and Practice. https://academic.oup.com/policing/article/doi/10.1093/police/paaf004/8058528.

Taber, C. S., & Lodge, M. (2006). Motivated skepticism in the evaluation of political beliefs. American Journal of Political Science, 50(3), 755–769. doi: 10.1111/j.1540-5907.2006.00214.x.

U.S. Department of Justice. (2022, May 20). Department’s updated use-of-force policy.https://www.justice.gov/d9/pages/attachments/2022/05/23/departments_updated_use-of-force_policy.pdf

van Veen, V., Krug, M. K., Schooler, J. W., & Carter, C. S. (2009). Neural activity predicts attitude change in cognitive dissonance. Nature Neuroscience, 12(11), 1469–1474. doi: 10.1038/nn.2413. PMID: 19759538.

Van Bavel, J. J., & Pereira, A. (2018). The partisan brain: An identity-based model of political belief. Trends in Cognitive Sciences, 22(3), 213–224. doi: 10.1016/j.tics.2018.01.004. PMID: 29475636.

Wotton JM, Ferragamo AC. Ambiguous animals: mutable mammal biases bird. Perception. 2012;41(4):501-3. doi: 10.1068/p7232. PMID: 22896924.

This isn’t just a post about cute animal videos. But stick with me for a moment.

The problem is what happens when your feed becomes a steady diet of plausible-looking, emotionally satisfying, low-effort content. According to research on propaganda methods, the goal isn’t always to convince you of a single lie. It is to exhaust your ability to tell what is real, so real information starts to feel optional, and you give up trying to tell the difference. When the information environment gets flooded, the effect of such content becomes indistinguishable from propaganda: confidence in reliable sources drops because everything starts to look equally questionable (Paul & Matthews, 2016; Chesney & Citron, 2019). Signal gets lost in the noise.

Generative AI opens the floodgates. It makes it cheap to produce “good enough” videos, audio, images, and text in huge quantities, including content that looks credible (Jones et al., 2025; Chesney & Citron, 2019; Romanishyn et al., 2025). The bad news is that we aren’t as good at spotting generative AI content as we’d like to think. Studies show people often fail to reliably detect deepfakes, even when they are warned, and when confidence is high (Köbis et al., 2021). Even with AI-generated voices, people struggle to tell real from fake speech (Mai et al., 2023).

Figure 1. Real or AI? How does a wedding band that is visible when the hand is palm up disappear when the hand is palm down? (credit: smartanimalz)

Why your brain is easy to fool

Your brain is wired to make fast judgments with limited information. Psychologists call these shortcuts heuristics: rules of thumb that usually work well enough, especially when you are busy, tired, or scrolling quickly (Tversky & Kahneman, 1974; Kahneman, 2011). They’re helpful in daily life, but they’re also easy to exploit, especially if you’re off guard.

A few ways these shortcuts lower your resistance:

• Repetition feels like truth. When you see the same claim again and again, it starts to feel familiar, and familiarity can get mistaken for accuracy. This is the illusory truth effect (Hassan & Barber, 2021; Udry & Barber, 2024).

• Repetition also creates a false “consensus.” If you keep seeing something, you can start to assume “everyone believes this,” even if you are just seeing the same idea recycled (Jalbert & Pillai, 2024).

• What feels personal feels true. Propaganda works best when it matches your identity, your fears, or what you already want to believe. That “desire to believe” is part of motivated reasoning, and there is evidence that brain systems involved in value, error detection, and social thinking track these biases (Kunda, 1990; Lois et al., 2024; Leong et al., 2020).

Content that is vivid, emotional, threatening, or flattering grabs attention and gets encoded more strongly than boring stuff. Once misinformation gets embedded in memory, it can become hard to untangle later (Shao et al., 2023).

AI is worse than typical misinformation

Old-school misinformation had a bottleneck because it cost time and effort to produce convincing material. AI removes that bottleneck in the following ways:

1. It scales. A small group, or even an individual, can produce an enormous amount of content across many platforms, quickly and continuously (Paul & Matthews, 2016; Romanishyn et al., 2025).

2. It personalizes. Online tracking makes it possible to learn what you click, watch, linger on, or react to (Englehardt & Narayanan, 2016). Platforms then optimize recommendations to keep you engaged, often using sophisticated models that personalize what you see (Covington et al., 2016). Research also shows that tailoring messages to psychological traits can increase persuasive impact (Matz et al., 2017).

3. It lowers detectability. Synthetic media does not have to be perfect. It only has to be plausible enough in a fast-scrolling environment, where people rarely stop to verify (Köbis et al., 2021; Mai et al., 2023).

4. It undermines trust. Once fakes are common, real footage and real experts get doubted too. This “liar’s dividend” dynamic makes truth harder to defend because uncertainty benefits whoever is trying to manipulate the conversation (Chesney & Citron, 2019).

Okay, But What About the Otters?

Cute animal clips are a perfect entry point because they hit the brain’s soft spots: novelty, warmth, humor, and our tendency to project human motives onto animals. Commenters often respond as if the behavior is real and reinforce the illusion in plain language, which builds a false consensus. Rewarding it with likes and click-throughs spreads it even more.

Sounds harmless, right? But it can have serious consequences. If people build false expectations about wild animals, they may approach them, feed them, or handle them in unsafe ways that harm humans and wildlife. Conservation researchers have raised concerns that AI-generated wildlife imagery and video can mislead the public and distort attitudes and behavior toward wildlife, including conservation efforts (Guerrero-Casado et al., 2025).

Scale that up and the same logic applies to politics, public health, and science. It’s hard for even experts to verify every claim, and the lay public can be overwhelmed. Beyond accuracy, the deeper danger is the flood itself: if verifiable, “good” information goes from “most of what you encounter” to “a small fraction,” people stop trusting what they see at all. At that point, the fight is not only over facts. It is over whether facts still matter, and whether an exhausted public will bother to dig deeper. In essence, the folks who control these algorithms can sway opinions by devaluing and corrupting truth.

How to build a firewall

It sounds a little hopeless. After all, the average person spends over 2 hours on social media each day, or close to one month out of the year. Short of logging off altogether, how can you protect yourself? The good news is that you don’t need to become a media forensics expert. You just need a few mental habits that slow the hack.

1. Pause before you react. If something makes you instantly angry, delighted, or shocked, that is your cue to slow down. The emotional punch is designed to slip through your defenses.

2. Ask three quick questions: Who made this? Where did it come from originally? Can I find the same claim from a source I already trust?

3. Do not let repetition substitute for thinking. Seeing something a lot does not make it true. It may only mean you’re seeing it a lot because it’s viral, and these algorithms have your number (Hassan & Barber, 2021; Jalbert & Pillai, 2024).

4. Remember that “feed” is rather literal: it means that it is being “fed” to you. It is a personalized stream optimized to keep you watching, liking and reposting (Covington et al., 2016). The fact that you are seeing it does not mean it is important, common, or accurate. It just means the algorithm has delivered it to you.

5. Change your settings to reduce the firehose. Turn off autoplay where you can. Clear your watch history or browser cookies periodically. Reset recommendations. Limit app permissions and reject tracking cookies when possible (Englehardt & Narayanan, 2016). The less information social media has about you, the less curated your experience will be.

6. Look for “boring confirmation.” Real events usually leave multiple traces: reputable reporting, consistent details across sources, and context. If the only evidence is a single viral clip, treat it as unproven. Conversely, known fakes are often called out by others, and a simple search on something that seems “too good to be true” can reveal whether that content has been flagged by others.

7. Be aware of your own biases. Remember that content is targeting what you like. If something seems to perfectly confirm your own biases, dig deeper before accepting it.

8. Seeing is not believing. That’s the new normal (Köbis et al., 2021; Mai et al., 2023). And as generative AI improves, it’s only going to get harder to tell fake from real.

9. Be an advocate for transparency. Ideally, AI-generated content would be as clearly labeled as a pack of cigarettes. But right now, AI algorithms work against consumer protection. Let your member of Congress know how you feel about AI transparency. Even simple warnings can help, but they work best when you see them early, and take them seriously (Karanian et al., 2020; Karanian, 2024).

A closing thought

When the dangers of AI are imagined, it’s often in apocalyptic “rise of the machine” terms. But the immediate dangers are actually more subtle, but no less dangerous. AI slop is not dangerous only because it can trick you, but because it can make everyone tired, cynical, and unsure of what is factual, and that is exactly the environment where manipulation thrives.

So yes, enjoy the harmless videos. Just keep your guard up. Being aware of what AI is doing to your brain will help ensure that your encounters with AI remain benign like a cute otter, instead of like being mauled by a bear.

And if you want to learn more about this topic, you can listen to me and two other experts discuss these themes, here.

[Dwayne Godwin is a NIH-funded neuroscientist who does research on PTSD, TBI and alcohol use disorder. He is the co-author of “Out of Your Mind: The Biggest Mysteries of the Human Brain”. The views expressed here are his own and do not necessarily reflect those of the National Institutes of Health or his employer.]

References

Chesney, R., & Citron, D. K. (2019). Deep fakes: A looming challenge for privacy, democracy, and national security. California Law Review, 107, 1753–1819. https://www.californialawreview.org/print/deep-fakes-a-looming-challenge-for-privacy-democracy-and-national-security

Covington, P., Adams, J., & Sargin, E. (2016). Deep neural networks for YouTube recommendations. Proceedings of the 10th ACM Conference on Recommender Systems. https://research.google.com/pubs/archive/45530.pdf

Englehardt, S., & Narayanan, A. (2016). Online tracking: A 1-million-site measurement and analysis. ACM CCS (extended version). https://www.cs.princeton.edu/~arvindn/publications/OpenWPM_1_million_site_tracking_measurement.pdf

Guerrero-Casado, J., Murillo-Jiménez, M., Carpio, A. J., & Tortosa, F. S. (2025). Threats to conservation from artificial-intelligence-generated wildlife images and videos. Conservation Biology. (PDF version listed by Wiley) https://conbio.onlinelibrary.wiley.com/doi/10.1111/cobi.70138

Hassan, A., & Barber, S. J. (2021). The effects of repetition frequency on the illusory truth effect. Cognitive Research: Principles and Implications, 6, 38. https://pmc.ncbi.nlm.nih.gov/articles/PMC8116821/

Jalbert, M., & Pillai, R. (2024). An illusory consensus effect: The mere repetition of information increases estimates that others would believe or already know it. Collabra: Psychology, 10(1). https://online.ucpress.edu/collabra/article/10/1/124533/203541/An-Illusory-Consensus-Effect-The-Mere-Repetition

Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux. https://books.google.com/books?id=ZuKTvERuPG8C

Karanian, J. M. (2024). Warning before misinformation exposure modulates memory encoding. Cognitive, Affective, & Behavioral Neuroscience. https://pmc.ncbi.nlm.nih.gov/articles/PMC11078839/

Karanian, J. M., et al. (2020). Protecting memory from misinformation: Warnings modulate cortical reinstatement. Proceedings of the National Academy of Sciences, 117(37), 22771–22779. https://www.pnas.org/doi/10.1073/pnas.2008595117

Köbis, N., Doležalová, K., & Soraperra, I. (2021). Fooled twice: People cannot detect deepfakes but think they can. iScience, 24(11), 103364. https://pubmed.ncbi.nlm.nih.gov/34820608/

Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480–498. https://pubmed.ncbi.nlm.nih.gov/2270237/

Leong, Y. C., Chen, J., Willer, R., & Zaki, J. (2020). Conservative and liberal attitudes drive polarized neural responses to political content. Proceedings of the National Academy of Sciences, 117(44), 27731–27739. https://www.pnas.org/doi/10.1073/pnas.2008530117

Lois, G., et al. (2024). Tracking politically motivated reasoning in the brain. Proceedings of the National Academy of Sciences. https://pmc.ncbi.nlm.nih.gov/articles/PMC11412250/

Mai, W., Bray, J., Davies, I., & Griffin, L. (2023). Warning: Humans cannot reliably detect speech deepfakes. PLOS ONE, 18(9), e0285338. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0285333

Matz, S. C., Kosinski, M., Nave, G., & Stillwell, D. J. (2017). Psychological targeting as an effective approach to digital mass persuasion. Proceedings of the National Academy of Sciences, 114(48), 12714–12719. https://www.pnas.org/doi/10.1073/pnas.1710966114

Paul, C., & Matthews, M. (2016). The Russian “firehose of falsehood” propaganda model: Why it might work and options to counter it. RAND Corporation. https://www.rand.org/content/dam/rand/pubs/perspectives/PE100/PE198/RAND_PE198.pdf

Romanishyn, A., et al. (2025). AI-driven disinformation: Policy recommendations for democratic resilience. [Open access on PMC] https://pmc.ncbi.nlm.nih.gov/articles/PMC12351547/

Shao, X., Li, A., Chen, C., Loftus, E. F., & Zhu, B. (2023). Cross-stage neural pattern similarity in the hippocampus predicts false memory derived from post-event inaccurate information. Nature Communications, 14, 2299. https://www.nature.com/articles/s41467-023-38046-y

Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124–1131. https://pubmed.ncbi.nlm.nih.gov/17835457/

Udry, J., & Barber, S. J. (2024). The illusory truth effect: A review of how repetition increases belief in misinformation. [PubMed record] https://pubmed.ncbi.nlm.nih.gov/38113667/

Jones, E. M., Newman, J. D., Kim, B., & Fogle, E. J. (2025). AI-Generated “Slop” in Online Biomedical Science Educational Videos: Mixed Methods Study of Prevalence, Characteristics, and Hazards to Learners and Teachers. JMIR Medical Education, 11, e80084. https://pmc.ncbi.nlm.nih.gov/articles/PMC12634010/

Now, more than ever, science needs your help. While the media is focused on the “One Big Beautiful Bill”, an equally important process of budget appropriations to determine the fiscal 2026 budget is underway. The White House's proposed budget includes a devastating 40 percent cut to the National Institutes of Health (NIH)—slashing its funding from $48.5 billion to $27.5 billion. Unchanged, the White House budget would freeze most new NIH grants. This is an existential threat to the country's biomedical research enterprise, ensuring that worthy projects aren't funded and that the next generation of biomedical scientists will struggle to do their life-saving work

(Photo by Sam Godwin)

Having built my own lab and trained young scientists to study brain disorders, I've seen firsthand how federal funding makes breakthroughs possible. Behind every scientific advancement stands a long chain of people including researchers, technicians, students, and patients, all working together toward a common goal on your behalf. Many of them are only able to do this work because of federal research funding. That funding is now in real jeopardy.

Biomedical research isn't just about distant discoveries in the Harvards and Hopkinses of the world. It happens here, in Winston-Salem, every day. It's a local effort that supports hundreds of jobs and brings millions of federal dollars into North Carolina. In addition to jeopardizing scientific progress, these cuts reduce future economic growth and opportunity in our knowledge-based economy. According to the advocacy group United for Medical Research, here in North Carolina NIH research funds nearly $2 billion in grants and contracts, supporting more than 103,107 private-sector bioscience jobs—including 21,769 directly tied to NIH funding. And that doesn't even count the coffee shops, restaurants, and small businesses sustained by the economic ripple effects of nearly $5B in total economic impact – or a return of about $2.56 in economic activity for every dollar invested. It powers the innovation economy that our city has bet its future on—through Innovation Quarter, the school of medicine, and our growing biotechnology sector.

That’s why proposed federal cuts to research funding are so troubling. When science budgets are slashed in Washington, the cuts don't just hurt scientists. They hurt patients waiting for better options and clinical trials, students training for careers in science and medicine. and communities like ours that have invested heavily in research as a pillar of our economy. They hurt you.

In times of fiscal uncertainty, science funding is often seen as optional. But I would argue it's an essential engine of progress. Every treatment that saves a life today started as a question asked in a lab like the ones we have right here in our city. Winston-Salem went from a tobacco economy to one focused on healing. That's something of which we can all be proud, and it’s something worth preserving.

Every future treatment for addiction and every cure for Parkinson’s, Alzheimer’s, or stroke is incubating in research labs supported by federal grants. These grants give hope to a neighbor that somewhere, someone is working on an advanced treatment for a disease that affects them and their loved ones. If we don’t act to preserve what we have built, we may be entering a time when that hope is no longer there for those who desperately need it.

That's why I'm asking you now, as your neighbor, to do something. We need our elected officials at every level to understand that research funding is a commitment to our health, our innovation, our shared values, and our future. They need to hear that it’s something you care about.

I hope you’ll contact your member of Congress. Let them know that science depends on their support, and that if we lose it, we're losing the future.

[This OpEd was first published in the Winston-Salem Journal July 3rd, 2025]

[Dwayne Godwin is a NIH-funded neuroscientist who does research on PTSD, TBI and alcohol use disorder. He is the co-author of “Out of Your Mind: The Biggest Mysteries of the Human Brain”. The views expressed here are his own and do not necessarily reflect those of the National Institutes of Health or his employer.]

Subscribe now

The march to widely-adopted AI seems unstoppable, driven by massive corporate investment and a lot of hype. High-profile startups have been fueled by over $131 billion poured into generative AI companies globally in 2024 alone (PitchBook, 2024). This funding boom has stimulated rapid progress in large language models (LLMs), multimodal systems, and autonomous agents, pushing AI capabilities. But it has also intensified debates about safety, oversight, and long-term viability.

There is reason to think that the NIH's aspirational vision of evolving from "data-science-driven analytics through semi-autonomous AI agents to fully autonomous, self-documenting biomedical AI beings" may be premature. While AI can enhance biomedical research, there is overwhelming evidence that current systems, particularly large language models (LLMs), are not yet capable of independently validating scientific research. Because the stakes are so high, deployment of AI in biomedical contexts must be approached with caution, transparency, and with appreciation of its limitations.

Let’s take a closer look.

Key Limitations of Current AI Systems

LLMs operate primarily as advanced pattern recognition engines. As Gary Marcus and other critics have noted, these systems are fundamentally statistical in nature—talented mimics, but lacking the genuine reasoning and understanding required for scientific inquiry (Marcus, 2022; Bender et al., 2021). Despite massive investments in computing power and training data, LLMs appear to be reaching performance plateaus, particularly with respect to generalization, inference, and reliability (Wei et al., 2022).

The Problem of AI Hallucinations

A critical issue in the application of AI to biomedical research is the generation of hallucinations - plausible, but factually incorrect information. These errors are often delivered with unwarranted confidence, making them difficult to detect and correct unless you’re a specialist. Unlike human mistakes, which usually follow identifiable cognitive patterns, AI hallucinations can be unpredictable and aren’t easily detected or fixed through current validation strategies (Ji et al., 2023). And because they are basically predicting word sequences, LLMs don’t do mathematical calculations well. In high-stakes biomedical contexts, this poses a serious risk to patient safety and research integrity.

A recent paper further challenges the belief that LLMs are on a linear trajectory toward artificial general intelligence (AGI – a true general intelligence similar to that of a human). Shojaee et al. (2025) tested advanced reasoning models such as Claude 3.7 Thinking and DeepSeek-R1 in controlled puzzle environments of increasing complexity. While these models exhibited some success in early phases of the tasks, their performance collapsed entirely beyond moderate complexity levels. Even when models were provided with explicit algorithms, they failed to execute them correctly, revealing a fundamental inability to perform logical reasoning. The researchers identified a counterintuitive pattern: as problems became harder, these 'reasoning models' actually reduced their effort, failing to leverage available computing resources. You could think of it as the AI equivalent of “giving up”. These findings expose a scaling limit in current LLM architectures and cast serious doubt on claims that these models are capable of generalizable, human-like reasoning, a critical prerequisite for autonomous scientific inference. In a way, such LLMs are talented bullshitters that collapse when confronted with a hard problem.

The Problem of Human Biases

Another sneaky problem is that we are wired to be fooled. Our own human neural networks are naturally inclined to see intelligence where none exists, a bias deeply rooted in our evolutionary history. Decades of research show that people routinely anthropomorphize inanimate systems, from thermostats to chatbots, especially when those systems mimic human-like traits such as language or emotional expression. This tendency isn’t just anecdotal. Studies have shown that subtle cues can lead users to attribute mind, agency, and even moral standing to AI systems (Epley et al., 2007; Nass & Moon, 2000). As LLMs become more fluent and interactive, this bias will probably intensify, leading to an inflated perception of understanding and competence (Placani, 2024). In biomedical settings, where trust and reliability are paramount, this illusion of intelligence poses a serious risk. Our gullibility in being taken in by an AI's capabilities can obscure its actual limitations, undermining safety, oversight, and scientific rigor. It can lead to an unwarranted surrendering of decision making in important contexts. And if health policy or patient treatment is based on their output, that reliance could cost lives. One example of this potential risk is a recent administration policy document that had signs of at least being partly generated by AI. If the justification for a particular policy agenda is based on hallucinated scholarship, it could be devastating for public health.

Opaqueness Threatens Reproducibility

Scientific research depends on transparency, reproducibility, and the ability to scrutinize underlying methodologies (i.e., “how” one arrives at a solution). Current AI systems, particularly deep learning models have hidden, layered structures, making it difficult to trace the origin of outputs or understand the decision pathways. This lack of interpretability (Lipton, 2018) undermines the standards of reproducibility and peer validation that are central to biomedical science.

Put simply, how can you know that something is right when you can’t tell how it went wrong?

The credibility of biomedical science rests on rigor and reproducibility, and that requires transparency. While AI can assist with data handling and pattern identification, it should not replace the foundational processes of hypothesis testing, peer review, and experimental replication. And by its nature, with “hidden” layers and tendency to hallucinate, it is less rigorous than human-based research can achieve.

Training these models requires large, representative datasets. Ironically, generating such datasets still depends on the very human expertise and traditional methods that AI is often imagined to replace. And when models learn from biased inputs, they risk amplifying existing disparities in healthcare and research. Such distortions can be especially dangerous in clinical contexts where biased or erroneous outputs can translate to harmful decisions (Obermeyer et al., 2019; Mehrabi et al., 2021).

Human-Centered Research is Irreplaceable

Human researchers bring to bear years of training, domain expertise, and contextual awareness that can’t easily be replicated by pattern-based systems. The ability to interpret ambiguous findings, question assumptions, and make intuitive leaps is foundational to scientific discovery, as is the humility to “know what we don’t know”. These human strengths are particularly critical in areas like hypothesis generation, anomaly detection, and clinical interpretation and experience.

Biomedical research inherently involves ethical decision-making by balancing risks and benefits. Current AI systems lack clear evidence for moral reasoning or value-based judgment (Oh and Demberg, 2025). Even sophisticated outputs are ultimately derivative of training data, without an embedded ethical framework. This limitation is seemingly acknowledged by NIH itself in its policy prohibiting the use of AI tools in peer review processes (NIH, 2023).

Setting Limits and Promoting AI Literacy

To ensure the safe and effective integration of AI in biomedical research, the NIH should explicitly position AI as a supportive tool rather than a replacement for human researchers. While AI excels at doing things like processing large datasets, supporting literature reviews, formalizing testable hypotheses, and automating routine analyses, its role should remain firmly under human oversight, particularly for tasks requiring interpretation, study design, ethical reasoning, and clinical translation.

To facilitate this integration, NIH must promote AI literacy across the research community by investing in training programs for students, postdocs, and experienced investigators alike (Even experienced scientists who may have followed AI developments for decades aren’t completely prepared for the rate of change in these systems). This should be complemented by administrative supplements to existing grants that incorporate innovative AI components. Equally important are standardized protocols for evaluating AI outputs, transparent mechanisms for identifying and correcting errors, and governance frameworks that define ethical, responsible use and oversight (e.g., augmenting existing Animal Care and Use Committees and Institutional Review Boards). The push toward AI shouldn’t compromise the traditional pillars of scientific rigor. And it shouldn’t need to be said that NIH must continue to support human-centered research training, maintain robust peer review, improve methods for high-quality data collection, and reinforce its commitment to reproducibility and transparency across all research domains (but there, I said it).

The Case for Next-Generation AI Research

LLMs are impressive. But in essence they are prediction machines, generating sentences based on surrounding context derived from its training data. Expecting AGI to spring from an LLM is a bit like expecting an entire adaptable brain to spring from a well-endowed, but specialized, temporal lobe. To responsibly advance biomedical AI, NIH should support foundational research into next-generation AI systems that address current limitations.

The brain itself can actually be our guide in this endeavor. Promising directions include hybrid architectures that integrate symbolic reasoning with neural networks, and biologically-inspired models grounded in brain-based computation (Hassabis et al. 2017; Ren and Xia, 2024). These approaches may offer more transparent, interpretable systems capable of basic reasoning, ethical contextualization, and scientific inference addressing many of the limitations surrounding current LLMs.

One of the most celebrated breakthroughs in domain-specific AI is AlphaFold, developed by DeepMind to predict protein structures from amino acid sequences. Its developers shared the Nobel Prize in Chemistry in 2024 (Callaway, 2024). Leveraging deep learning trained on vast biological datasets, AlphaFold achieved unprecedented accuracy in CASP14 (Jumper et al., 2021), solving problems that had challenged molecular biology for decades. Its predictions have been integrated into public protein databases and are now guiding experimental design in drug discovery and structural biology worldwide. While AlphaFold is not by any means an autonomous scientist, it illustrates how AI can surpass human capabilities in well-defined scientific tasks when grounded in domain-specific knowledge and robust datasets. AlphaFold’s success underscores the potential of narrowly focused, biologically informed models to accelerate discovery, especially when their outputs are subject to expert interpretation and validation.

It might appear that I’m contradicting my critique of LLMs. And it’s true that LLMs and AlphaFold are fundamentally prediction engines. However, LLMs predict and generate the next word in a sequence based on linguistic context, trained on a massive trove of human language. AlphaFold, by contrast, is more predictive than generative. It predicts the 3D structure of a protein from its linear (1D) amino acid sequence, using deep learning models informed by evolutionary relationships, known structures, and known physical constraints of those structures (Jumper et al., 2021). Thus while both rely on patterns learned from prior data, AlphaFold’s predictions remain in its “lane”, grounded in biological truth. In contrast, commercial LLMs rely on semantic and syntactic patterns from text, often from broad internet sources whose context could literally be anything, including your weird uncle’s notions about UFOs. This makes AlphaFold’s outputs far more interpretable and constrained, while unconstrained LLMs may produce fluent but ungrounded or even misleading responses.

Another example of such innovation is the recent deployment of Robin, an AI-powered multi-agent scientific assistant developed by the nonprofit FutureHouse. Robin autonomously synthesized literature and proposed a repurposed treatment (ripasudil) for age-related macular degeneration, which was subsequently validated in lab experiments by human researchers (The Scientist, 2025). While Robin does not constitute AGI, it shows some of the promise of modular AI systems that, when closely coupled with expert oversight, and constrained in validated training sets, can make meaningful contributions to discovery. These hybrid approaches that blend computational power, curated training sets, and human expert domain knowledge may offer a more realistic and safer path forward than autonomous AI agents alone. Importantly, strategic investment in this kind of innovation could reassert U.S. leadership in health-related AI while mitigating risks associated with premature automation.

What’s Next

AI remains a promising, yet still developing, tool in biomedical research. Its strengths may be more aligned with augmenting human-led discovery, not replacing it. Misuse risks amplifying hallucinations, bias, and opacity in domains where public trust and safety are essential. The NIH’s vision of fully autonomous biomedical AI systems seems at best premature, and at worst potentially risky. True, the NIH has undertaken risky projects before (the Brain Initiative and the Human Genome Project stand as examples of high risk, high reward science). Yet both of these projects were evidence-based on firm scientific antecedents. It’s less certain that the current AI enthusiasm is similarly grounded.

Investment in next-generation AI research, particularly hybrid systems combining symbolic logic with biologically-inspired neural architectures, offers a path toward AI that is more transparent, ethical, and scientifically robust. Such work will be essential if these systems are to evolve into a trustworthy collaborator in biomedical science or in the clinic, rather than a new source of risk. Continued U.S. leadership in this area depends on prioritizing strategic innovation in AI aligned with core scientific values of rigor and reproducibility.

Accordingly, the scientific community must hold AI to the same evidence-based standards demanded of all biomedical tools. Until those standards are met, human expertise remains indispensable. A more prudent path forward emphasizes integration, including review by research boards as to how AI will be ethically used in a given study, and review of integration of AI into standards of care by state medical boards. This will also mean using AI to extend human expertise, not to supplant it, or emphasizing AI’s more established pattern recognition capabilities rather than its overly confident decision making.

Ensuring responsible AI implementation will require increased human involvement, more rigorous oversight, more and better curated data, and stronger foundations in traditional scientific methodology, while recognizing that we are responding to a rapidly evolving field whose only constant is change.

Dwayne Godwin is a Professor of Translational Neuroscience at the Wake Forest University School of Medicine. He is the co-Author of “Out of Your Mind: The Biggest Mysteries of the Human Brain.”

This commentary should not be interpreted as reflecting the views of his employer.

References

Bender, E. M., Gebru, T., McMillan-Major, A., & Shmitchell, S. (2021). On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency.

Callaway E. (2024). Chemistry Nobel goes to developers of AlphaFold AI that predicts protein structures. Nature. 634(8034):525-526. doi: 10.1038/d41586-024-03214-7 PMID: 39384918.

Epley, N., Waytz, A., & Cacioppo, J. T. (2007). On seeing human: A three-factor theory of anthropomorphism. Psychological Review, 114(4), 864–886. https://doi.org/10.1037/0033-295X.114.4.864

Hassabis D, Kumaran D, Summerfield C, Botvinick M (2017). Neuroscience-Inspired Artificial Intelligence. Neuron. 95(2):245-258. doi: 10.1016/j.neuron.2017.06.011. PMID: 28728020.

Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., ... & Fung, P. (2023). Survey of Hallucination in Natural Language Generation. ACM Computing Surveys, 55(12).

Jumper, J., Evans, R., Pritzel, A. et al. (2021). Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–589. https://doi.org/10.1038/s41586-021-03819-2

Lipton, Z. C. (2018). The Mythos of Model Interpretability. Communications of the ACM, 61(10), 36-43.

Marcus, G. (2022). Deep Learning Is Hitting a Wall. Nautilus Magazine. https://nautil.us/deep-learning-is-hitting-a-wall-238440/

Mehrabi, N., Morstatter, F., Saxena, N., Lerman, K., & Galstyan, A. (2021). A Survey on Bias and Fairness in Machine Learning. ACM Computing Surveys, 54(6), 1-35.

Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of Social Issues, 56(1), 81–103. https://doi.org/10.1111/0022-4537.00153

NIH (2023). Guidance on Use of Generative AI in Peer Review. National Institutes of Health.

Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to manage the health of populations. Science, 366(6464), 447-453.

Oh S and and Demberg V (2025). Robustness of large language models in moral judgements. R. Soc. Open Sci.12: 241229.

PitchBook. (2024). AI startups grabbed a third of global VC dollars in 2024. Retrieved from: https://pitchbook.com/news/reports/q4-2023-emerging-tech-research-generative-ai

Placani, A. Anthropomorphism in AI: hype and fallacy (2024). AI Ethics 4, 691–698. https://doi.org/10.1007/s43681-024-00419-4 @misc{illusion-of-thinking,

Ren, J., & Xia, F. (2024). Brain-inspired Artificial Intelligence: A Comprehensive Review. arXiv preprint arXiv:2408.14811.

Shojaee P, Mirzadeh I, Alizadeh K, Horton M, Bengio S and Farajtabar M (2025) The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity. https://ml-site.cdn-apple.com/papers/the-illusion-of-thinking.pdf

Tran, L. (2025, June 6). An AI-powered “scientist” proposes a treatment for blindness. The Scientist. https://www.the-scientist.com/an-ai-powered-scientist-proposes-a-treatment-for-blindness-73079

Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., ... & Le, Q. (2022). Chain of Thought Prompting Elicits Reasoning in Large Language Models. arXiv preprint arXiv:2201.11903. https://arxiv.org/pdf/2201.11903

When we consider our national investments through this lens, few expenditures create more enduring good than our commitment to biomedical research. The NIH represents America's promise to future generations—a promise that we will not surrender to disease, that we will marshal our collective intellect against suffering, and that we will continuously expand the boundaries of human knowledge and empathy.

What's at Stake

Someone once told me, “Show me your budget and it will show me what you care about”. I’ve fought for my share of budgets, and I can verify that this is true.

But the NIH isn't just any other budget. It's the foundation of America's biomedical enterprise, supporting over 300,000 researchers at more than 2,500 institutions nationwide. Its work touches every American family through advances in cancer treatments, vaccine development, treatments for the scourge of addiction, precision medicine, and countless other innovations.

When we cut NIH funding, we don't just delay scientific papers or postpone clinical trials. We fundamentally alter life trajectories— of the patients whose hopes hinge on breakthrough treatments, and of researchers who dedicate their careers to solving medical mysteries.

A researcher abandoning a promising project because of funding cuts represents more than a career detour. It's potentially years of scientific progress erased. A clinical trial that never launches means patients who might have been saved are instead left waiting for treatments that never materialize. A student whose fellowship is ripped away could be our next Pasteur or Einstein. So many first-generation researchers, myself included, would never have had the chance to become scientists without the support of the NIH at key moments in their career path.

Beyond Dollars: The True Cost

Budget discussions often focus exclusively on financial metrics, but the impact of NIH funding cuts extends far beyond spreadsheets:

• Global Leadership: America's scientific dominance isn't guaranteed—it's earned through consistent investment. As other nations dramatically increase research funding, cutting our commitment is a retreat. I don’t want to alarm you, but in my view we can’t afford to give up the ethical space to countries who may develop technologies capable of reshaping life as we know it. We must be in the lead not only to develop new technologies, but to help guide their ethical uses.

• Economic Returns: NIH research generates ~$2.56 in economic activity for every dollar invested. These cuts sacrifice future economic growth. In my home state of North Carolina, NIH research funds over $2 billion in grants and contracts supporting over 92,000 private sector bioscience jobs, including 21,769 directly supported. And that doesn’t count the coffee shops and restaurants and other impacts of this virtuous economic driver.

• Talent Pipeline: Young scientists make career decisions based on funding stability. Each funding cycle that's cut doesn't just affect current projects—it discourages the next generation from entering scientific fields.

The Bipartisan Legacy at Risk

Historically, NIH funding has transcended political divisions. Leaders from both parties recognized that investments in biomedical research reflect core American values: innovation, compassion, and foresight. Presidents from Nixon to Obama championed the expansion of the NIH, understanding that scientific progress serves no partisan agenda.

This tradition of bipartisan support has produced remarkable results. Life expectancy has increased dramatically. Diseases once considered death sentences are now manageable conditions. Pandemic preparedness, though still imperfect, has benefited from decades of sustained research.

We now stand at a crossroads. Will we continue this bipartisan legacy, or will we compromise our national commitment to scientific excellence built over generations?

A Call for Bipartisan Advocacy

As professionals in diverse fields—healthcare, business, education, technology—we must recognize that NIH funding affects us all. The treatments and medications that extend our lives, the innovations that power our economy, and the scientific education that inspires our youth all depend on robust federal research investments.

Otherwise, what are we about as a country - as Americans?

I urge every professional, regardless of political affiliation, to advocate for preserving and expanding NIH funding:

1. Contact your representatives directly, explaining how biomedical research impacts your field and community.

2. Join advocacy organizations that amplify the collective voice for science funding.

3. Share stories of how medical research has touched your life or the lives of loved ones.

4. Engage business leaders in your network—help them understand how research funding drives innovation ecosystems.

Advocacy need not be partisan. Democrats and Republicans alike benefit from cancer treatments, vaccines, and mental health advances. We all share the hope that our children will live in a world with less disease and suffering than the one we inherited.

The Measure of Our Actions

Returning to Churchill's quote, I believe few investments meet his standard better than our commitment to biomedical science. The good that emerges from laboratories today—treatments, technologies, understanding—will continue benefiting humanity for generations.

Budget decisions made in coming weeks will reveal our national character. Will we kill the golden goose? Will we prioritize short-term savings over long-term health? Will we sacrifice tomorrow's cures for today's political expediency?

I believe we are better than that. Americans have always valued progress, innovation, and compassion. We've never been a nation that shrinks from challenges or abandons its commitment to future generations.

The things we truly value are worth fighting for. And few things deserve our passionate defense more than our nation's commitment to biomedical research. The NIH represents the best of America—our optimism, our ingenuity, our dedication to improving human lives.

Let us ensure that when future generations measure the good of our actions, they will find that we stood firm in our commitment to scientific progress. This legacy is too precious to compromise, the investment too valuable to surrender, and the fight too important to lose.

I invite you to join us in this cause—not as liberals or conservatives, but as Americans committed to a healthier future for all.

[Note: A recent article in the Harvard Gazette cites data that every NIH dollar returns $2.56 in economic benefit. See: https://news.harvard.edu/gazette/story/2025/03/nih-funding-delivers-exponential-economic-returns/ ]

This essay reflects my opinions as a private citizen and is not intended to represent the opinions of my employer.