Does the company own your face? (June 21)

The bad news: California’s bid to protect the personal IP rights of voice actors, performers, authors, and others (AB 412) died in committee earlier this month.

The good news: A proposal to protect the neural rights of all workers looks to be on its way to final passage in Sacramento.

California’s AB 1883 would prohibit an employer from using a workplace surveillance tool that uses AI to, among other things, collect neural data or identify a worker’s emotional state.

“Neural data” is any information generated by measuring the activity of a person’s central or peripheral nervous system. Yes. That’s how invasive and creepy this technology has become. If lawmakers approve the bill this weekend—they’re scheduled to adjourn on Monday night—it still would need to gain Gov. Gavin Newsom’s signature to become law.

Pete Hegseth just issued a terrifying demand (Feb. 26)

This story reaches back to February, but it illustrates how crushingly slow our court system can be.

If you’ll recall: Anthropic CEO Dario Amodei met with Defense Secretary Pete Hegseth regarding the Pentagon’s use of the company’s powerful AI model, Claude. Amodei wanted to set limits. Like, say, not spying on American civilians and not using Claude to kill at the President’s whim. Hegseth promised retribution if Anthropic refused to give him and his boss carte blanche.

Six months and one Iran war later, a federal judge ruled (just Thursday night) that Hegseth and the Pentagon acted illegally by punishing Anthropic for its criticism of the Department of Defense’s views on AI use.

The judge wrote that the government’s actions “were based on a desire to make a public example out of Anthropic for its ‘arrogance’ in criticizing the government, not based on any articulable basis to believe that Anthropic would actually sabotage its model.”

AI shaming has a moment (Aug. 5)

The battle over AI writing and AI shaming reached the editorial page of The Wall Street Journal this week. Surprise, surprise: The Journal backs the machine.

Will Oremus, writing in The Atlantic, recounts how billionaire investor Stanley Druckenmiller published an editorial in The Journal on Monday criticizing Treasury Secretary Scott Bessent over Bessent’s attempt to manipulate the bond market. Readers smelled AI. Pangram, the AI detection tool, scored Druckenmiller’s words as 100% robot.

Druckenmiller, when pressed, confessed. Well…it wasn’t so much a confession as an indignant Yeah, so what if I did?

“Of course I used AI,” he said on Tuesday. “I’m not embarrassed by it. I write everything using AI now for the same reason I use a calculator when I do math problems.”

The Journal, being The Journal, sided with Bossy McMoneypants.

Paul Gigot, the paper’s editorial page editor, defended the practice. “AI is a fact of modern life,” he said, then followed up with a column on the subject, mocking the “media pharisees” who shame those who let the machines do the scribbling. Gigot went on:

“Many contributors, especially prominent politicians or CEOs, employ speechwriters or hire a Beltway writing shop. How different is that from using AI to streamline an essay?”

My prayer to the blessed god of karma this weekend: May the inbox of Paul Gigot flood like the tide of Noah with the slop-written submissions of a million autonomous agents.

The Chicago Law AI Manifesto (July 18)

A few weeks ago we noted the University of Chicago Law School’s deeply considered policy regarding the proper and improper use of AI in the education of attorneys. Since that midsummer report was issued, a number of other universities have released their own reports and policies. To wit:

MIT’s AI report contains fewer specifics and more general guidance:

“Rather than simply ‘AI-proof’ current methods of assessment, instructors need to revisit what they really want students to know and devise assessments that foster, or even include, the kind of productive struggle that builds durable understanding and mastery.

We urge instructors to consider forms of assessment that are less vulnerable to AI, and more valuable for learning, such as oral exams, semester portfolios, and out-of-class assignments paired with in-class conversations.”

UC Berkeley Law’s newly announced policy, by contrast, gets into the nitty gritty:

“This policy…provides students with the opportunity to develop the skills they need to conceptualize, outline, draft, revise, and edit their work by forbidding the use of AI for these purposes in connection with work submitted for credit. It also forbids using AI to translate work for credit. And it prohibits AI use for any purpose in any exam situation.”

Berkeley Law now forbids:

• Asking an AI tool to brainstorm a paper topic or thesis

• Asking an AI tool to propose an organizational structure for a paper

• Using AI to compose a paragraph summarizing a legal rule for use in a paper

• Asking an AI tool to identify repetitive passages in a paper that should be cut

• Asking an AI tool to polish a paper by correcting grammatical mistakes

• Using AI to generate an exam outline

• Asking AI to translate a paper originally written in another language into English

I’m still waiting to hear from my alma mater, the University of Washington. When last I checked the campus seemed to be divided between UW admin leaders pushing the use of Purple, a new in-house generative AI model, and the rabble (faculty and students) resisting under the vocal leadership of Emily Bender. There’s an empty policy space big as Red Square waiting for someone or some group to step in and fill.

Earlier this year New York Times wine critic Eric Asimov predicted “A.I. Is Coming for the Sommeliers.” Restaurant chat boards are peppered with stories of patrons challenging a well-considered recommendation with “Well, ChatGPT says you’re wrong…” Every blowhard with an iPhone now believes he holds the entirety of world wine knowledge in his palm.

I’ve seen a number of terms used to describe this dynamic. The one I find most apt is machine supremacy. It carries an appropriate edge of menace and entrapment. Like white supremacy, it’s not just an assumption but a sinister and all-enveloping worldview.

When did we start assuming a computer knows better than a human expert? And why?

Rationality as the highest value

For insight I turned to Benjamin Recht, a UC Berkeley professor of computer science whose new book, The Irrational Decision: How We Gave Computers the Power to Choose for Us, explores the origins and errors of machine supremacy.

“It too often feels like every choice in our daily experience, no matter how momentous, gets reduced to a risk analysis of chances and costs,” Recht writes. “Where does this unshakable pressure to constantly optimize come from?”

The answer: computers and war.

Complex calculation machines

World War II, Recht notes, involved an unprecedented number of civilian scientists, engineers, mathematicians, and economists. Mathematicians developed digital devices to carry out complex calculations at faster-than-human speed. Those early computers helped predict the paths of rockets, break codes, and filter radar signals.

“Mathematicians were tantalized by the possibility that if humans could make a decision,” Recht writes, “then that decision could not only be made by machines, but could possibly be better made by machines.”

Because those decisions were arrived at through mathematics, a science (not an art), they were dubbed “rational” and “scientific.” The gendering of language is notable here. Think of the opposite terms. Irrational. Emotional. You don’t have to be Betty Friedan to smell a rat here. Numbers aren’t squishy, the math boys implied. Numbers aren’t swayed by emotional appeals. Numbers are the hard men of the world, and the hard men of the world deal in numbers.

Using the power of a computer to optimize human decisions

The computer has always been a numbers-based machine. And so its early development and use followed the path of least resistance—through the numerically-based fields. Finance. Economics. Aerospace.

Calculation speed wasn’t enough for some of the early pioneers in computer development, though. They were obsessed with what might be thought of as crossing over: Using numbers to calculate optimal decision-making in non-numerical situations.

Enter game theory and the RAND Corporation. John von Neumann, the mathematics genius who worked on the Manhattan Project and laid the foundation for quantum physics, among other hobbies, also created what we know today as game theory. This is essentially the idea that in any given situation each individual seeks to maximize their own gain while minimizing loss.

In 1948, von Neumann and others founded the RAND Corporation, a government-supported think tank focused on military research and development. Per Recht: “It seemed reasonable that military strategy might be formalizable as a zero-sum game between warring factions.”

A cultural trope takes hold

RAND became synonymous with brainiacs using computers to out-think the Soviets. The cultural trope of a white-coated—and white-skinned, and white-haired—scientist plugging numbers into a wall-size computer, which then spit out a paper-receipt answer, seeped into America’s brain.

For kids raised in the 1970s, this indoctrination came every Saturday morning. In the Hall of Justice, the Super Friends relied on their Justice League computer to solve the most vexing crime puzzles.

Different machines served different purposes. The general computer, with its whirling reel-to-reel tapes, spit out answers as below:

The Trouble Alert was a global surveillance system that anticipated our own Flock-driven spy state:

The most uncanny futurecast, though, was the G.E.E.C., Goodfellow’s Effort Eliminating Computer. Its inventor, the kindly white (-haired, -skinned, -coated) Prof. Goodfellow, predicted the G.E.E.C. would free humans from all physical labor and responsibility. Students would be taught by computers. Actors would be computer-generated.

Business! Transportation! Theater! Athletics! All taken care of for us by the G.E.E.C.:

I’m not claiming Super Friends as a dominant cultural force. It’s but a single illustration of the many ways we’ve been taught not to question the unimpeachable intellectual power of the computer.

Wrong answer

Here’s the funny thing about RAND and its game theory: It never worked.

“Game theory,” writes Benjamin Recht, “didn’t describe how people actually acted at all.” (Emphasis his.)

Humans don’t play games optimally. We never have perfect information. We’re swayed by a moment’s optimism, or pessimism, or anger, or hunger. Sometimes we’re drunk and reckless. Other times we play with scared money. Today the American president is motivated by vanity, greed, and retribution. God knows what kind of physical and mental condition he’s in. You’re going to game theory that?!

Game theory and statistical machine learning, Recht writes, are “remarkably powerful in the appropriate contexts but limited as a universal basis of decision making.”

Less rational, more sophisticated

It turns out that humans aren’t more rational decision-makers than computers but we are more sophisticated decision-makers. Writes Recht:

“The most talented human experts thrive on ineffable intuition that often resembles extrasensory perception. Though they can try to mentor trainees in their ways, their best practices are often inarticulable and must be learned through experience. Human expert knowledge is precisely the kind that can’t be cleanly written down as a set of rules.”

Or, I would add, contained in a set of data.

I’m reminded of the senior partners scene in Margin Call, J.C. Chandor’s exquisite film about the start of the 2008 credit crash. With a Lehman Brothers-like firm teetering on a pile of bad debt, the board chairman—Jeremy Irons at his best—helicopters in for an emergency 2 a.m. meeting.

“Do you care to know why I am in this chair with you all? I mean why I earn the big bucks?” he asks the gathered vice-presidents. “I am here for one reason and one reason alone. I am here to guess what the music might do a week, a month, a year from now. That’s it. Nothing more.”

Computers, AI, algorithms—they’re amazing at finding patterns and making best-guess predictions based on past events rendered as data. But they don’t have the lived experience and the feel for the room that Jeremy Irons possesses.

You can find analogues in other fields. Dave Roberts, the brilliant manager of the Los Angeles Dodgers, made a number of moves in last year’s World Series that went against the analytics. Roberts knew that baseball analytics are based millions of data points in a long 162-game season. A tight seven-game series, played in cold November weather under intense pressure, is a different beast entirely. After the series win Roberts said he was aware of what the analytics said he should do. “But I know my guys,” he said. His eyes and ears picked up clubhouse chatter. He read body language. He knew who was feeling loose, who was wound too tight, which arm was lively and which was ragged. “I know my guys.”

And so it is with the sommeliers. ChatGPT can tell any grape fancier how a chablis pairs with a steak, or that Wine Enthusiast scored this bottle an 89. But the sommelier knows the local vintners. She sampled ten different bottles two days ago. The chef’s preparation of that rib-eye wants a fuller-bodied red.

Her expertise isn’t as rational as the chatbot’s. It’s far more sophisticated.

For more from author Ben Recht…

Subscribe to his Substack arg min, interesting stuff, worth checking out.

This is a dynamic I’m well familiar with. I wrote about marijuana legalization and drug policy for a decade. Upon informing a stranger of my job, they’d get that look. And ask that question. Do you…partake?

Read more

It was March.

This week another dustup dusted up in ye olde groves of literature. Will Oremus, the Atlantic staff writer who’s done groundbre…

Read more

“Is this bad?” he asked.

Yes. When a powerful new AI model escapes the bounds of its developer’s sandbox —the entire purpose of which is to keep it in—and immediately goes Viking berserker,…

Read more

If you’re an AI Humanist reader you probably don’t like it.

I get it. If you’d asked me three years ago about digital age-gating I would have told you to ge…

Read more

Finally, last week, we had a breakthrough. The University of Chicago La…

Read more

Last Tuesday, Meta released Muse Image, a new AI product that allowed users to generate imagers using other people’s Instagram pictures without their knowledge or consent.

By Wednesday the backlash had begun. A tech justice non-profit official told the BBC it was a recipe for disaster: “It is hard to see why Mark Zuckerberg thinks facilitating yet more …

Read more

Corporate executives, desperate to appear on-trend and bloodlessly efficient, are forcing their companies to adopt AI systems that aren’t ready for prime time.

Div…

Read more

I’ve been thinking about that a lot lately, and not just because of deepfakes.

You may assume an autonomous adult owns an inherent intellectual property right in their own…

Read more

In the Humanist household, today marks the official onset of summer, as my dearly beloved partner bid a final adieu to her students at 3pm and promptly collapsed on the classroom floor. As one does.

So. The season turns. Let’s catch up on some of the stories we’ve been following over the past nine months.

Read more

Note: This is a feature available to paid subscribers of The AI Humanist. Become a supporting subscriber to unlock the full article.

If you haven’t taken steps to protect your privacy, your data is being scraped by AI bots and used to train the world’s largest AI models.

AI bots, also known as crawlers, are pieces of software that explore the internet, scan websites, and gather data. That data is stolen and used, without consent, to train the world’s largest and most lucrative AI models. The global networking company Cloudflare estimated that by mid-2025 bots accounted for 30% of all web traffic.

You alone can’t stop all this bot-scurrying. But you should take steps to lock your own digital front door.

A good portion of all AI training data comes from within an app itself. In other words: If you’re using Microsoft’s MS 365 suite (Word, PowerPoint, Excel, Outlook), Microsoft is using your data to train its AI models. Unless you take steps to opt out.

Nearly all of the popular apps allow users to opt out of this involuntary data scrape. (Except for Meta. Of course.)

I’ll keep updating and adding to this list as more opt-out levers become known. The apps addressed here include ADOBE : AMAZON ALEXA : CHATGPT : FACEBOOK : GOOGLE : GOOGLE GEMINI : GROK : INSTAGRAM : LINKEDIN : MICROSOFT 365 : MICROSOFT COPILOT : PERPLEXITY : PINTEREST : SNAPCHAT : SQUARESPACE : TIKTOK : WORDPRESS: X

Adobe

Adobe’s terms of service grant the company the right to use your artwork to train their AI models. This means any content you create on their platforms (Illustrator, Photoshop, et al) may be used to train generative AI models that could devalue or replace your work. (Hat-tip to Nashville-based artist Ginny St. Lawrence, who posted this information on her blog.)

Do this to disable Adobe’s ability to scrape your work and use it to train AI:

Go to the Adobe Account Privacy Settings page. Locate the “Content analysis” setting and turn it off.

Amazon Alexa

We all know how Alexa devices work: They listen, listen, listen all day long and then perk up when they hear “Hey, Alexa” in your voice. This is, depending on your outlook, incredibly clever and convenient, or an example of how dystopian surveillance capitalism gets its hooks into the home.

Amazon has a lengthy page—Amazon, Echo Devices, and Your Privacy—addressing all the privacy concerns around Alexa.

Q: Is Alexa recording all my conversations?

A: No. By default, Alexa-enabled devices are designed to detect only your chosen wake word. The device detects the wake word by identifying acoustic patterns that match the wake word. No audio is stored or sent to the cloud unless the device detects the wake word.

Are those nice words? Yes. Do I trust them? No. Tech companies are constantly caught breaking privacy rules, using data that’s not theirs, and generally acting in bad faith when it profits them. The EU’s GDPR privacy laws are the best in the world and they have a tracker that updates violations hourly.

You’re going to have to decide for yourself. I have an Alexa device that I try to unplug when I’m not using it to listen to music. When it’s on, I assume it’s hearing and recording every sound in the room.

The only privacy toggle for Alexa is the cord that plugs into your wall socket.

ChatGPT

Here’s how to stop ChatGPT from using your data and inputs from training its own models, while still continuing to use the chatbot.

Web controls (as a logged in user):

To disable model training, navigate to your profile icon on the bottom-left of the page, select Settings > Data Controls, and disable “Improve the model for everyone.” While this is disabled, new conversations won’t be used to train OpenAI’s models.

Web controls (as a logged out user):

To disable model training, navigate to the ? icon on the bottom-right of the page, select
Settings > Data Controls, and disable “Improve the model for everyone.”

iOS app:

Tap the three dots on the top right corner of the screen > Settings > Data Controls > toggle off “Improve the model for everyone.”

Android app:

Open the menu through the three horizontal lines in the top left corner of your screen, select Settings > Data Controls, and toggle off “Improve the model for everyone.”

Facebook

Read more

That’s no longer the case. Things have changed.

Read more

Last week Pope Leo XIV released a book-length consideration of artificial intelligence and its relationship to humanity.

His 42,300-word encyclical, Magnifica Humanitas (Magnificent Humanity), is nothing short of astonishing. It may be the best book on AI published this year.

Read more

Greetings to all of you long-suffering Substack subscribers to The AI Humanist. Thank you for your patience.

For the past nine months I’ve based 95% of the newsletter on Beehiiv, a Substack competitor.

Beehiiv has outstanding graphic capabilities and excellent te…

Read more

The New York Times broke a story this morning that, even as I read it over coffee, felt kind of inevitable.

Reporter Benjamin Mullin revealed:

“The author of a nonfiction book about the effects of artificial intelligence on truth acknowledge…

Read more

The 24-hour churn of CNBC has bred a generation of…

Read more

Har har har.

Fast-forward 40 years and here we are: Vizio, Sony, and other smart TV manufacturers are getting out of the television hardware b…

Read more

Yesterday The Bookseller, a publishing trade publication, revealed that book editors are now routinely feeding entire manuscripts into ChatGPT and other AI bots in order to create instant plot summaries and market …

Read more

For more than two years we’ve been told that the AI takeover was unstoppable. Inevitable. Talking with tech hypesters felt like confronting the hivemind of the Borg: Resistance is futile.

Over the past four weeks, though, the weather has changed. The public taste for AI, excited by those first encounters w…

Read more