Misconception of the Month

“Every problem has a solution. We just need the right data and the right algorithm—and we can fix it.”

This is the solutionist creed, and it’s deeply embedded in how we think about progress. To put it another way, problems exist to be solved, and technology is how we solve them. And, if a problem persists, we simply haven’t found the right solution yet.

But some problems are not the kind of thing that gets solved and treating them as if they were doesn’t just fail. It often makes things worse.

Against Solutionism

Recasting every social and political question as an engineering problem with a technical fix has a name: solutionism, a term Evgeny Morozov coined in his 2013 book To Save Everything, Click Here. Whatever you might think of his execution elsewhere, even more than a decade later, his description of Silicon Valley’s habits hit home. Morozov was writing about the internet, but the argument fits AI even better. AI is a more general-purpose solvent than anything he had in 2013, and the temptation to pour it on everything has only grown.

It’s fair to say that this solutionist mindset has produced real wonders. Problems that once seemed intractable, disease and scarcity among them, have yielded to technical ingenuity, and the successes have trained us to expect more of the same.

But, at the same time, it’s important to recognize that this mindset has limits, and those limits matter for how we make sense of the current moment.

Problems and Predicaments

In a 1973 paper on planning theory, Horst Rittel and Melvin Webber distinguished “tame” problems from “wicked” ones. Wicked problems, they argued, have no definitive formulation and no clear stopping point or definitive solution. Rather, a proposed solution can only be judged better or worse, never simply true or false. They used poverty, crime, and educational reform as their examples as they described an approach that has subsequently informed fifty years of planning and design theory.

I want to push past this important distinction. A wicked problem is still, formally, a problem, something that in principle admits of a better or worse answer, however hard that answer is to locate or settle. Yet some of what follows reflects something different entirely. Call it a predicament — a feature of human existence that admits no convergent answer — where the honest options are response and accommodation, not solution, however partial.

Death is such a predicament, and so is conflict. So is the tension between individual freedom and collective welfare, the gap between what we want and what we can have, and the difficulty of truly understanding people who are different from us.

These aren’t failures of technology or of our ability to find solutions to complexity. Rather, they are conditions of being human. Treating them as problems to be solved produces a characteristic distortion: false hope, followed by a persistent bafflement and frustration about why we can’t solve the problem or why the problem won’t stay solved once we think we’ve achieved a solution.

Is This Just an Excuse?

A skeptical reader is right to ask whether “predicament” is an excuse or a permission slip: a way of declaring a problem outside the reach of solutions precisely when we’ve stopped trying to solve it, or when solving it would threaten institutional power. Segregation was called a predicament, a fact of human nature, by people who benefited from it. So was maternal mortality, by physicians uninterested in obstetric research and, not coincidentally, uninterested in the deaths of poor and Black mothers, until germ theory and antiseptic practice turned out to be a solution after all. The predicament label has a bad history of protecting the status quo.

So, here’s the test I’d propose. A predicament remains a predicament rather than becomes a problem after every party with a self-interest in declaring it unsolvable is removed from the room. Maternal mortality was a problem mislabeled as a predicament, and it collapsed once anyone with the right incentives looked for causes. Death has not budged despite two centuries of biomedicine aimed directly at it, and no incentive structure explains that away. One of these is solutionism’s failure to solve a real problem; the other is a genuine predicament. The distinction has to be empirical, not rhetorical, and it has to survive whether the people calling something a predicament had a reason to want it that way.

The humanities have always been concerned with predicaments, tested this way. Literature explores difficulties that never fully resolve, such as love and loss or the distance between our ideals and our conduct. History shows how every solution generates new problems. Philosophy sits with tensions that don’t go away: freedom and determinism, or justice and mercy.

This is realism about what kind of creatures we are and what kind of world we inhabit — essential equipment for facing a moment when solutionist thinking is reaching its limits.

The Solutionist Temptation in AI

AI, we’re told, will fix hiring bias by stripping human prejudice from the process. It will fix misinformation by detecting and filtering it. It will fix loneliness by supplying companionship on demand.

That’s the pitch, and it’s a tempting one: the technology is powerful enough, and improving fast enough, that solutionism has found in it its most persuasive instrument yet.

Each of these applications has something to it, for AI can help meaningfully with each of these challenges. But notice how the framing obscures what’s actually hard.

Take hiring bias. The hard part was never technical. It’s a question of what we value, and how historical injustice persists in present structures. An algorithm can be made to optimize for certain outcomes, but choosing the outcome worth optimizing is a human judgment involving contested values.

Misinformation runs into the same wall. Detecting false content was never that hard; the real problem is trust, the relationship between citizens and institutions, and the conditions under which people believe anything at all. You can filter content, but you cannot algorithmically restore the social conditions that make shared truth possible.

Loneliness follows the same pattern. What’s missing isn’t connection in general. It’s a particular kind of connection: being known and seen by someone who actually cares. An AI companion can simulate aspects of connection, but the simulation may in fact deepen loneliness by substituting for what we really need.

In each case, the solutionist framing takes a human predicament and treats it as a problem with a technical solution. While the technical intervention addresses one dimension of the issue, it often leaves the deeper difficulty untouched.

Where the Algorithm Wins

Wine tasting looks like exactly the kind of judgment call solutionism cheapens: a trained palate, built over decades, can recognize nuances no formula could catch. In 1990, Princeton economist Orley Ashenfelter shook up the wine-tasting world when he published a weather-based equation that predicted how a Bordeaux vintage would ultimately be judged, years before any critic could taste the finished wine. The reaction was immediate. The New York Times ran it on the front page under the headline “Wine Equation Bends Some Noses Out of Joint,” and Robert Parker, then the most powerful wine critic alive, called the formula “ludicrous and absurd.”

In reality, Ashenfelter’s formula turned out to be highly accurate. It flagged 1986, a vintage the tasting establishment had praised on release, as mediocre, and rated 1989 and 1990 as exceptional years before anyone could confirm it by drinking them; both verdicts held up.

As this example illustrates, sometimes the formula can out-perform trained human judgment. It isn’t the whole story, though. Critics were tasting something the formula never touched, a wine’s character and balance at the moment of drinking, not just its long-run value and trajectory. Ashenfelter later conceded that the model works at the level of a vintage as a whole and says little about any single bottle. Both kinds of judgment can be defended; they were just answering different questions.

The actual argument is narrower and distinguishes between questions that a formula can actually answer and questions that depend on things that no formula measures.

What the Humanities Teach

What the humanities cultivate is comfort with difficulty that simply won’t resolve: sitting with problems that don’t close, making peace with permanent imperfection, using the test described above rather than convenience to tell a genuine predicament from a problem in disguise.

I see this as maturity and acceptance rather than despair — the difference lies in what you do with a difficulty you can’t resolve.

Take Paradise Lost again. Milton’s poem is fundamentally about a predicament, the existence of evil and suffering in a world created by a good God. Theodicy, the attempt, in Milton’s words, “to justify the ways of God to men,” doesn’t make the difficulty disappear. The poem doesn’t solve the problem of evil; it explores it and offers ways of living with it.

What readers gain from Paradise Lost isn’t a definitive answer, but rather the capacity to hold difficulty without being destroyed by it, to find meaning in a world that still includes real suffering, even when the cosmic justice we might wish for is unavailable.

Donne’s poetry returns again and again to the predicaments of love and death, handled differently than Milton’s. “A Valediction: Forbidding Mourning” has no interest in solving the problem of the separation of two lovers; it gives the reader a way of understanding separation that makes it bearable. The Holy Sonnets go further, performing Donne’s spiritual struggles rather than resolving them, giving form to a faith that includes doubt, fear, and the felt absence of God.

Neither poet is failing to find a solution. Milton and Donne are succeeding at a different task: helping us live with what cannot be changed.

The Organizational Cost of Solutionism

In my years leading institutions, I’ve watched solutionism create real challenges.

The pattern is consistent. An organization faces a difficult situation, someone proposes a solution (a new policy, a new technology, a restructuring) that addresses the surface problem while leaving the underlying difficulty untouched. Often it creates new problems, and the organization is baffled: we solved that problem, why is it back?

During the pandemic, I saw this constantly. Universities searched for solutions to the disruption: new remote-learning technologies, new structures for decision-making under uncertainty. Many of these were necessary and helpful. But the deepest challenges weren’t amenable to solutions.

How do you maintain community when people can’t gather? How do you support students whose lives have been upended in wildly different ways depending on their circumstances? Decisions had to be made with incomplete information, among reasonable people who disagreed about values, and leaders had to act without the comfort of certainty.

These were predicaments, requiring a different kind of response, not a solution. The leaders who understood this distinction were more effective than those who kept searching for fixes: they could acknowledge difficulty without pretending it away and make decisions without claiming a certainty they didn’t have.

This capacity, the ability to work through predicaments rather than simply solve problems, is one of the things that humanities education develops. And it’s in short supply.

The Limits of Optimization

Minimizing costs, maximizing outputs is another way of describing optimization, and it’s the form solutionism that most often emerges once a goal is fixed in place.

Economists have a name for this: Goodhart’s Law, after Charles Goodhart, who observed over fifty years ago that any statistical regularity used as a policy target tends to collapse under the pressure of being targeted. The familiar version, that a measure that becomes a target stops being a good one, comes from anthropologist Marilyn Strathern’s 1997 restatement. Higher education runs directly into it.

I think about this often in higher education, where there are constant pressures to optimize: improve graduation rates, reduce time to degree, increase job placement, lower costs. Each is reasonable, but optimizing for any one outcome can undermine the others and undermine things that aren’t measured at all.

A university optimized for graduation rates might lower standards; one optimized for job placement might narrow the curriculum to vocational training; one optimized for cost might eliminate the small seminars and mentorship relationships that produce the deepest learning. The metrics improve while the thing they were supposed to measure, an actual education, erodes.

This is a general problem with optimization. It hits precisely the target it was set, while everything outside that target goes unnoticed, and the things that go unnoticed are often the things that matter most, are hardest to measure and slowest to develop.

The humanities teach resistance to premature optimization. They turn our attention to what gets omitted, to unintended consequences, to the difference between what we can measure and what we should value. Some things worth doing aren’t efficient, and some values exist in permanent tension. Wisdom lies in navigating these tensions.

Tragedy and Comedy

The humanities also teach two sensibilities, tragic and comic, terms the literary critic Northrop Frye used to organize the whole of literary criticism in 1957.

The tragic sensibility recognizes limits — that our reach exceeds our grasp, that good intentions can still produce bad outcomes. Some losses are permanent, and some choices are genuinely tragic: whatever we pick, something of value is lost.

Set against that is the comic sensibility, which trusts in resilience — that people adapt, that joy persists even alongside suffering. Meaning here comes less from solving problems than from living fully in spite of them: love, community, creativity, the ordinary pleasures that don’t wait for the world to be fixed.

Solutionism has neither sensibility: it doesn’t recognize limits, since it assumes every problem is solvable, and it can’t find meaning in persistence because it’s always chasing the next fix. It produces a kind of brittle optimism that shatters on contact with irreducible difficulty.

The humanities cultivate both sensibilities. Tragedy teaches us to see limits; comedy teaches us to live within them. Together, they produce a mature relationship with a world that includes both real problems and permanent predicaments.

Against Anti-Solutionism

I want to be clear that I’m arguing against solutionism here, not against solutions.

Some problems really are problems, with real solutions that we should find. The printing press solved manuscript scarcity. Vaccines solved certain diseases. The technical achievements of human civilization are real and worth celebrating. So, per the last section, is Ashenfelter’s wine equation.

The error is in treating every difficulty as this kind of problem: reaching for technical fixes when what’s needed is patient judgment or expecting resolution when what’s actually possible is only accommodation.

The humanities don’t teach us to stop solving problems. They teach us to recognize which difficulties are problems, which are wicked problems, and which are predicaments, and they equip us with capacities for all three: the analytical skills to address what’s solvable, and the wisdom to live with what can’t be fixed.

This is realism, not pessimism — the kind that makes hope possible because it doesn’t stake that hope on the impossible. Meaning here comes not from expecting every difficulty to be resolved but from the human capacity to live well despite difficulty that never fully goes away.

What We Need Now

The current moment is awash in solutionism. AI is offered as the answer to everything: inequality, loneliness, disease, death, the human condition itself. Techno-optimists promise that we’re on the verge of solving problems that have defined human existence since the beginning.

We need voices that can hold the real promise of technology while resisting the solutionist fantasy. This tool is genuinely powerful, and it can help with real problems. But some difficulties are not going away, and we need the capacity to live with them.

The humanities provide these voices, not because humanists are anti-technology (many of us use AI enthusiastically and gratefully) but because the humanities have always been concerned with the parts of human experience that don’t yield to technical intervention: love and death, the gap between what we are and what we might become.

These concerns aren’t antiquated. They’re more relevant now than ever, precisely because the solutionist temptation has never been stronger.

The Human Element is a monthly newsletter on humanities and durable skills in an age of artificial intelligence. Next month: “What Employers Actually Need (and Don’t Know How to Ask For)”—bridging the gap between what hiring managers say they want and what they actually need.

Misconception of the Month

“AI can handle the analysis now. We just need people to review the outputs and make sure nothing’s obviously wrong.”

This framing is everywhere—in job descriptions, in workflow redesigns, in the casual way organizations talk about integrating AI. The machine does the work; humans provide quality control. A quick review, a sanity check, and we’re done.

But this fundamentally misunderstands what happens between raw output and meaningful action. The “review” isn’t a minor final step. It is where the real work happens. And calling it review obscures how much judgment, interpretation, and contextual understanding it actually requires.

Between Output and Action

There is a layer of work that sits between information and action, between output and meaning, between what a system produces and what an organization actually does with it. I have come to think of this as the interpretation layer.

AI can generate analyses, summaries, recommendations, drafts, and predictions. What it cannot do on its own is determine what these outputs mean in context, whether they are appropriate for a particular situation, how they should be weighted against other considerations, or what should actually be done about them.

This interpretive work is where the value in any output is actually realized. And it requires capacities that need to be understood clearly before they can be defended.

The Illusion of the Final Step

The “review and approve” model of human-AI collaboration assumes that interpretation is simple—a matter of catching errors, flagging anomalies, giving a thumbs up or thumbs down. The casual version of this assumption is everywhere in business writing. The serious version, embedded in frameworks like the FDA’s rules for AI in medical devices or the EU AI Act’s high-risk category requirements, takes review more seriously—requires it to be meaningful and substantive. That the EU, in its 2026 Digital Omnibus, deferred those high-risk obligations to 2027 and 2028—because the standards needed to make oversight operable were not yet ready—only sharpens the point: substantive review is harder to build than to mandate. But even the serious version tends to conceive of review as a check on output rather than as the generative work of making output meaningful. That is the deeper misconception.

Consider what review actually requires in any consequential domain.

An AI generates a legal brief. To review it meaningfully, you need to understand the legal issues at a level that lets you assess whether the analysis is sound, whether the precedents are apt, whether the framing serves the client’s interests. You need to know what the brief does not say—what arguments it might have made, what risks it might be underestimating. You need judgment about how this brief will land with a particular judge, in a particular jurisdiction, at a particular moment in the case.

An AI produces a medical diagnosis. To interpret it, you need to integrate it with everything else you know about the patient—their history, their presentation, the things that don’t quite fit the pattern. You need to weigh it against your clinical experience, which includes cases where the algorithmic answer was technically correct but practically wrong. You need to communicate it to the patient in a way that is honest, humane, and actionable.

An AI drafts a strategic recommendation. To evaluate it, you need to understand the organizational context it is entering—the politics, the history, the capabilities and limitations of the people who would implement it. You need judgment about what is feasible, what is timely, what will actually move the institution forward versus what looks good on paper.

In each case, what we call review is the moment when human judgment transforms output into action. Without that judgment, the output is inert—or worse, dangerous.

What Interpretation Requires

The interpretation layer involves at least five distinct capacities.

Contextual understanding. AI outputs exist in contexts that determine their meaning. The same analysis might be brilliant in one situation and catastrophic in another. Interpretation requires understanding the context—the history, the stakeholders, the constraints, the goals—at a level that is often tacit, built from experience, and resistant to explicit articulation.

Judgment about relevance. Not everything an AI produces matters equally. Interpretation involves determining what is important and what is noise, what deserves attention and what can be set aside. This is not a mechanical filter; it requires understanding what you are trying to accomplish and how different pieces of information bear on it.

Integration with other knowledge. AI outputs are typically narrow—answers to specific queries, analyses of specific data. Interpretation involves integrating these outputs with everything else you know, including things about which the AI was not asked and could not know. The interpreter holds a broader picture and situates the AI’s contribution within it.

Translation across contexts. Raw outputs often need to be translated before they are useful—adapted for different audiences, reframed for different purposes, connected to different concerns. A technical analysis needs to become a board presentation. A data summary needs to become a patient conversation. This translation is interpretive work.

Judgment about action. Finally, interpretation involves deciding what to do. The AI can recommend, but someone must choose—and choosing involves weighing considerations that may not be captured in the analysis, accepting responsibility for outcomes, and committing to a course of action under uncertainty.

These capacities are central to how knowledge becomes useful in the world.

What Makes Interpretation Irreducible

A sophisticated reader will push back here. AI is getting better at the things I just described. Frontier models do process context. They weigh relevance against stated goals. They integrate across domains, translate across audiences, recommend actions. The five capacities I listed are a frontier the machines are crossing.

The objection holds, and it exposes the distinction that matters most. There is a difference between doing interpretation and being responsible for it, and the difference is not a difference of capability.

A model can produce the cognitive operations of interpretation. It can read the brief, weigh the precedents, generate the recommendation. What it cannot do is occupy the place from which the interpretation is undertaken—the place of someone who will bear the consequences of being wrong, who can be held to account by other humans, who has a stake in the outcome that is more than computational. Responsibility is not a cognitive capacity. It is a moral and relational position, and it is constitutively human: it requires someone who can be held to account by others.

This is why the interpretation layer remains human even as AI gets better at the cognitive work interpretation involves. The competent interpreter is not just someone who can produce the analysis; it is someone who can be answerable for it. When the judgment turns out to have been wrong, someone must be able to say, “I made that call.” A model cannot say this in any sense that organizational life, professional ethics, or legal liability recognizes. The “I” is missing from the model in exactly the way it must be present in the interpreter.

This is what is at stake in the deskilling problem. We are at risk of losing the trained human capacity to occupy responsible positions—to make calls one will own, to bear consequences one can articulate, to be the kind of professional whose judgment can be trusted because they have practiced being accountable for judgments.

The Renaissance Parallel

I find myself thinking about Renaissance courts—the environments where Donne and others operated, where knowledge and power intersected.

Courts were sites of interpretation. Information flowed in—diplomatic dispatches, intelligence reports, petitions, proposals. The task of courtiers, advisors, and secretaries was not merely to process this information but to interpret it: to understand what it meant in context, to judge its reliability, to connect it to other knowledge, to translate it for different audiences, to advise on action.

The good courtier—as Castiglione described in Il Cortegiano (1528), translated as The Book of the Courtier—needed judgment, discretion, and the ability to read situations. They needed to understand not just what was said but what was meant, not just what was proposed but what was possible. They served as an interpretation layer between raw information and meaningful action. Castiglione’s central concept, sprezzatura, captures the constraint under which they worked: the art of making difficult judgment appear effortless, of concealing the labor that produced it. The constraint persists today wherever interpreters of AI outputs work inside organizations.

Donne himself navigated these waters. Before he became Dean of St. Paul’s, he served as secretary to Sir Thomas Egerton, the Lord Keeper—a role that required exactly this interpretive capacity. He read documents, drafted responses, advised on sensitive matters. His literary training was not separate from this work; it was preparation for it. The same skills that let him read a poem with care let him read a political situation with care.

This role did not disappear with the Renaissance. It has been a constant feature of every complex organization. What is changing now is that AI is handling more of the information-processing work that used to fall to humans, which concentrates more attention on the interpretive work that remains.

The question is whether we have enough people with the capacities that interpretive work requires.

The Deskilling Risk

Here is a danger I think about often: as AI handles more of the analytical work, we may inadvertently deskill the interpretation layer.

If early career employees no longer do the foundational analytical work—because AI does it faster—how do they develop the judgment needed to interpret AI’s outputs? Judgment comes from experience, from doing the work yourself and learning from mistakes, from building tacit knowledge through practice. If that practice disappears, judgment may not develop.

I saw versions of this in my years as a university president. Students who relied too heavily on calculators sometimes lost the number sense needed to recognize when a calculation was obviously wrong. Researchers who relied too heavily on statistical software sometimes lost the intuition needed to interpret what the statistics actually meant.

Law firms are wrestling with the contemporary version of this question right now. Senior partners worry that associates who use AI to draft initial briefs and memos may never develop the legal judgment that comes from struggling with the material themselves—the slow accumulation of pattern recognition, the feel for how arguments land, the instinct for what a particular court will receive. The associates who take the shortcut produce passable work faster. The question their firms cannot yet answer is whether they will ever produce excellent work, and whether they will be able to interpret AI outputs at the senior level when their turn comes. The same conversation is happening in medical residency programs, in consulting firms, in newsrooms, and in engineering organizations.

The risk with AI is similar to the calculator and statistics cases but differs in kind, not merely in degree. A calculator offloads a bounded operation—arithmetic—and delegating it leaves the larger capacity, mathematical reasoning, largely intact. AI reaches higher up. The operations it now performs—framing a problem, weighing evidence, drafting an argument—are the very ones whose practice builds interpretive judgment in the first place. Offload the practice, and the judgment may never form. Where the work delegated is the work by which judgment is made, delegation shades into deskilling, and the two become difficult to tell apart. If we treat the interpretation layer as a simple review function—something junior peopleearly-career staff can do with minimal training—we may hollow out the very capacities that make interpretation valuable. The senior people who have those capacities will retire. The junior peopleearly-career colleagueswho should be developing them will have practiced reviewing outputs rather than building the judgment to interpret them.

This is not inevitable. But avoiding it requires intentionality about how we train people, how we design workflows, and how we think about the role of human judgment in AI-augmented work.

Teaching Interpretation

Interpretation is teachable, but not in the way technical skills are teachable. You do not learn interpretation by memorizing procedures. You learn it through practice with complex material, guided by people who have developed interpretive capacity themselves.

This is what humanities education has always done. A literature seminar is a laboratory for interpretive practice. You read a text, develop an interpretation, present it to others, encounter counterarguments, refine your reading. Over time, you develop the capacity to interpret—to move from surface to depth, to recognize what is significant, to construct meaning from complex material.

The skills transfer. A student who has learned to interpret Donne’s poetry—to move from confusion to comprehension, to recognize how the parts relate to the whole, to understand why interpretive choices matter—has practiced something they will use in every complex professional situation. They have learned that meaning is not simply extracted but constructed, that interpretation requires both rigor and creativity, that good readers see things poor readers miss.

The same is true of historical interpretation, philosophical interpretation, the interpretation of art and music and film. Each domain offers its own materials and methods, but all develop the underlying capacity: the ability to make meaning from complex material, to exercise judgment about significance, to translate understanding into action.

This is the interpretation layer. And it is what humanities education has always developed.

The Organizational Challenge

Organizations face a challenge they often do not recognize: they need to build and maintain interpretive capacity even as AI handles more analytical work.

This means rethinking how they train early career employees. If AI does the first-draft analysis, how do junior employees develop the judgment to evaluate it? The answer is not to withhold AI from junior peopleearly-career colleagues—that is inefficient and ultimately futile. The answer is to design workflows that build interpretive capacity intentionally: requiring junior peopleearly-career staff to predict what the AI will produce before seeing it, to critique AI outputs systematically, to defend their interpretations in discussion with more experienced colleagues.

It means rethinking how they hire. If the interpretation layer is where value is created, organizations need people with strong interpretive capacities. This means looking beyond technical credentials to assess judgment, contextual understanding, and the ability to translate across domains. It means recognizing that the humanities graduate who can interpret complex texts may be better prepared for this work than the technical specialist who has never practiced interpretation.

It means rethinking how they value interpretive work. In many organizations, analysis is visible and valued while interpretation is invisible and undervalued. The person who produces the report gets credit; the person who figures out what the report actually means and what to do about it is just “reviewing.” This needs to change. Interpretation should be recognized as skilled work, compensated accordingly, and developed as a core organizational capability.

The Human Remainder

There is a phrase I have come to appreciate: the human remainder. It refers to what remains irreducibly human after automation has handled what it can handle.

The interpretation layer is a large part of the human remainder. AI will continue to improve at analysis, at pattern recognition, at generating outputs. But the work of determining what those outputs mean, of being answerable for that determination, and of deciding what should actually be done—this remains human. AI will keep getting better at the cognitive operations; what it cannot take on is the answerability. Responsibility is the irreducible thing, and interpretation in any consequential setting requires a responsible interpreter.

If we understand that the interpretation layer is where human value is created, we can invest in it intentionally. We can educate people for it. We can design organizations around it. We can stop treating interpretation as a residual task and start treating it as the core competency it actually is.

The alternative is to let interpretation erode through neglect—to assume that review is simple, that judgment requires no development, that AI outputs can be trusted without deep contextual understanding. That path leads to organizations efficient at producing outputs and incompetent at using them well.

If you find yourself doing this work—translating outputs into meaningful action, bearing the judgment calls that turn analysis into decision—recognize what you are doing. Name it. Advocate for it. And if you are responsible for developing others, invest in building this capacity. It is not automatic, and it will not survive on its own.

The Human Element is a monthly newsletter on humanities and durable skills in an age of artificial intelligence. Next month: “Against Solutionism”—on the tendency to see every problem as a technical problem awaiting a technical solution, and what the humanities teach about problems that do not resolve cleanly.

History Doesn’t Repeat But It Does Instruct

Misconception of the Month

“This time is different. AI is unprecedented—nothing in history can help us understand what’s coming or how to navigate it.”

You hear versions of this constantly. The technology is so new, so transformative, so unlike anything that came before, that historical perspective is useless. We’re in uncharted territory. The past has nothing to teach us.

The dismissal of historical thinking is itself worth examining. But I want to start by granting more ground than humanists usually grant.

What’s Actually Unprecedented

The serious form of “this time is different” is not a vague claim that things feel new. It is a specific argument about category. Every prior general-purpose technology—the printing press, the steam engine, electricity, the computer—disrupted the distribution of human cognition or the augmentation of human labor. Gutenberg’s press did not read books; it reproduced them faster. The steam engine did not make decisions; it moved loads. Electricity did not think; it powered the things humans developed. AI is the first technology that does the cognitive work itself: it reads, summarizes, drafts, codes, decides, and increasingly acts.

This is not a difference of degree. It is a categorical break. Daron Acemoglu has argued in Power and Progress (2023, with Simon Johnson) that the shape of any technological transformation is determined by political and institutional choices, and that the choices currently being made around AI tilt toward displacement rather than augmentation. The substitution effects run deeper than in prior general-purpose technologies. The diffusion is faster. The implications for how knowledge work organizes itself are not yet legible.

I take this seriously. I take it seriously as a humanist.

And yet: the conclusion that history is therefore useless does not follow. In fact, it reverses. When a transformation is genuinely unprecedented—when we cannot find an exact precedent—the resources with which we are left are the second-order capacities historical thinking develops: the sense that the present is contingent, the comfort with complexity, the pattern recognition across imperfect analogies, the epistemic humility about our own certainty. These are more essential when no template exists, not less. The people who insist that history is useless are usually the same people who then reach reflexively for one bad historical analogy after another—the railroad, the Manhattan Project, the internet—without any framework for thinking about how analogies work or where they fail.

So the move I want to make is not the easy one. It is not “history rhymes, the printing press teaches us everything we need to know.” It is harder and more interesting: precisely because AI represents a genuine discontinuity, the kind of thinking history develops is what we most need.

The Printing Press, Used Properly

Take the analogy most often pressed into service—Gutenberg and his printing press—and use it the way historical thinking actually works. Not as a template, but as a structured comparison whose limits are part of the point.

Before Gutenberg, knowledge was scarce and controlled. Manuscripts were expensive, copying was laborious, and access was mediated by institutions—monasteries, universities, courts. The people who controlled texts controlled knowledge.

The printing press shattered this. Suddenly texts could be reproduced cheaply and distributed widely. Knowledge escaped institutional control. Authorities who had controlled information found their power challenged. The early print era was awash in propaganda, conspiracy theories, and outright fabrication: the Malleus Maleficarum (1487) propagated witch panic across Europe at a velocity that handwritten texts could never have achieved; printed forgeries of papal indulgences circulated faster than Rome could refute them; the early Reformation pamphlet wars turned Luther into something we would now recognize as a media phenomenon, his face mass-produced in Lucas Cranach’s woodcuts.

The disruption was profound and disorienting. Many predicted catastrophe; others predicted utopia. Both were partly right and largely wrong.

What actually happened was more complicated. The press did not simply democratize knowledge; it created new gatekeepers, new hierarchies, new forms of authority. It enabled the Reformation and the Scientific Revolution, but also the religious wars and new forms of propaganda. It took generations for society to develop the institutions, norms, and literacies needed to navigate the new information environment.

The analogy to AI is useful on the information-environment dimension: the disruption of gatekeepers, the explosion of content, the collapse of existing filters, the slow institutional adaptation. It is much less useful on the cognitive-substitution dimension, because the printing press did not substitute for cognition. This is what doing historical thinking properly looks like: knowing which dimensions of an analogy hold and which do not. Reaching for “this is like the printing press” without that discernment is not historical thinking. It is decoration.

What Historical Thinking Provides

Let me be more precise about what historical thinking actually develops.

A sense of contingency. The present feels inevitable—of course things turned out this way. Historical thinking reveals that at every juncture, things could have gone otherwise. The decisions that shaped our world were made by people who didn’t know how things would turn out, who were navigating uncertainty just as we are. This is liberating: the future is not determined, and the choices we make now matter.

Comfort with complexity. Historical situations are never simple. There are always multiple causes, multiple actors, multiple perspectives, unintended consequences, and long chains of causation. Studying history develops tolerance for this complexity—the capacity to hold multiple factors in mind without collapsing into false simplicity. This is essential for navigating a present that is equally complex.

Pattern recognition across contexts. History offers a vast repository of human experience—how societies have handled technological change, economic disruption, political crisis, cultural transformation. None of these situations match ours exactly, but patterns emerge. How do institutions adapt or fail to adapt? How do new technologies get absorbed into existing social structures? What distinguishes transformations that go well from those that go badly? Historical thinking develops the capacity to recognize these patterns and apply them, analogically, to new situations.

Epistemic humility. Every era has had its confident predictions about the future. Most have been wrong—not entirely wrong, but wrong enough to counsel humility. Historical thinking teaches us that we are probably wrong too, in ways we cannot yet see. This isn’t paralyzing; it’s freeing. We can act without certainty, make decisions without pretending to know what we cannot know, and remain open to revision as events unfold.

The Renaissance and Its Synthesizers

I return often to the Renaissance, not because it is my scholarly specialty but because it offers such a rich case study in navigating overlapping transformations. The printing press, as I mentioned. The recovery of classical texts that challenged medieval frameworks. The discovery of the New World, which shattered assumptions about geography and human diversity. The Reformation, which broke the religious unity of Europe. The beginnings of modern science, which would eventually transform how humans understood their place in the cosmos.

The people living through this did not know they were living in “the Renaissance.” They were navigating uncertainty, trying to make sense of a world where old frameworks were failing and new ones had not yet stabilized. Some clung to the old ways. Some embraced the new recklessly. The wisest found ways to synthesize, and we still read them because they did. Erasmus negotiated between scholastic rigor and humanist learning, neither rejecting inherited methods nor capitulating to them. Montaigne invented the essay as a form precisely because his moment required a way to think under uncertainty—his recurring Que sais-je? (“What do I know?”) is the epistemic humility I have been describing, given literary form. Bacon proposed a new empirical method to organize knowledge that no single mind could now contain. None of them had a template either.

Donne registered the disorientation directly. In “An Anatomy of the World” (1611), he watches the Copernican revolution and other ruptures dissolve the inherited order. The line most often quoted—“new philosophy calls all in doubt”—is followed by something more devastating: “’Tis all in pieces, all coherence gone; / All just supply, and all relation.” That is the experience of a framework collapsing in real time, written by someone living inside the collapse. What strikes me, reading Donne now, is not that he had answers. He plainly did not. What he had was the discipline to keep questioning honestly, to refuse the consolations of false certainty in either direction, and to write his way toward a synthesis that did not yet exist.

These are our questions too: how do we orient ourselves when the old frameworks fail; what endures when so much changes; how do we make choices with integrity when we cannot see where our choices lead. Historical thinking does not give us their answers. It gives us companionship in the questioning—and models of how to question well.

The Technologist’s Blind Spot

In my experience, the people most likely to dismiss historical perspective are technologists—those closest to the development of new technologies, most immersed in their novelty.

This is understandable. If your daily work involves creating things that have never existed before, it is natural to feel that the past has little to offer. And technologists are right that they understand certain aspects of AI better than historians do—the technical capabilities, the trajectory of development, the engineering challenges.

But technologists often have a blind spot about how technologies are absorbed into society. They see the technology clearly and assume that social effects follow straightforwardly from technical properties. They underestimate how much the impact of a technology depends on the social, economic, political, and cultural contexts into which it is introduced.

It is worth noting that this conclusion is not the exclusive property of humanists. Economist Paul Krugman, writing recently about AI and historical analogy, makes a version of the same argument from within a rigorous empirical tradition—and points specifically to the postwar productivity boom, driven not by radical new technologies but by organizational and human factors, as evidence that institutional and human responses to technology matter as much as the technology itself. History is not a humanist’s special pleading. It is an analytical resource that holds up across disciplines.

History teaches otherwise. The printing press did not have effects; it had effects in particular contexts. The same technology produced different outcomes in different places, depending on existing institutions, power structures, and cultural patterns. Understanding these contextual factors is where historical thinking excels.

The most thoughtful technologists I have encountered are those who have developed historical consciousness—who understand that they are not just building tools but intervening in complex social systems with long histories. They seek out historians, social scientists, and humanists as collaborators. They know that there is much they don’t know.

What Covid Taught—And What It Is Still Teaching

The pandemic offered a crash course in the value of historical perspective, though the lesson is more layered now than it appeared in 2021.

Early in the crisis, I noticed a divide in how people were processing events. Some treated Covid as entirely unprecedented—and therefore unnavigable. Others recognized patterns from previous pandemics, previous crises, previous moments when institutions faced novel challenges under uncertainty.

The second group was better equipped, though “better equipped” is doing more work than it once did. Six years on, the lessons of Covid are still being debated: about school closures, about expert authority, about the costs of policies that seemed prudent in the moment and look less so in retrospect. That argument is itself a historical process. The meaning of recent events is established slowly, contentiously, and never finally. People who expected the verdict on 2020 to be obvious by 2026 misunderstood how history works.

As a university president navigating that period, I drew constantly on historical perspective. I was not looking for a template. I was looking for orientation—ways to make sense of a situation that resisted easy sense-making, and ways to make decisions I would have to live with regardless of how the verdict turned. Historical thinking provided that orientation, not by giving answers but by providing frameworks for navigating uncertainty.

The same is true now with AI.

The Futurist Temptation

There is a genre of AI commentary that is essentially futurism—predictions about what AI will do, how society will change, when various milestones will be reached. Some of this is thoughtful; much of it is not. Almost all of it will turn out to be wrong in ways that will be obvious in retrospect.

History counsels skepticism toward futurism. Not because the future is unknowable—we can reason about it, prepare for it, shape it—but because confident predictions almost always miss what actually matters. The important developments are usually the ones no one predicted. The predicted transformations usually take longer than expected and unfold differently than anticipated.

This is not an argument for passivity. It is an argument for a different kind of preparation—not preparing for a specific predicted future, but for developing the capacities to navigate whatever future actually arrives. Adaptability over prediction. Resilience over optimization for a single scenario. Judgment over formulas.

These are the capacities historical thinking develops. Not the ability to predict what comes next, but the ability to respond wisely when what comes next arrives.

The Long View

One of history’s greatest gifts is perspective on timescales.

We tend to experience change as either glacially slow or catastrophically fast. Day to day, nothing seems to happen. Then suddenly everything seems to be happening at once. Both perceptions are distortions.

Historical thinking reveals that significant transformations usually unfold over decades, not years. The printing press did not transform Europe overnight; the process took generations. The Industrial Revolution was a century-long upheaval. The digital transformation has been unfolding for fifty years and is not finished.

This perspective matters now in a particular way. The pace of AI capability gains since late 2022 has been faster than the diffusion of most prior technologies, and it is tempting to project that pace forward indefinitely. But capability and absorption are different processes. Capabilities advance on a months-to-years cadence; the institutional, legal, educational, and cultural adaptations that will determine what those capabilities actually mean operate on a decades cadence. We are not in the final act; we are in the early chapters. There is time to shape what comes—time to develop the institutions, norms, and human capacities that will determine whether the transformation goes well or badly.

And there is a further wrinkle that history surfaces. A technology can be genuinely transformative without delivering its promised gains on the timeline or in the form its proponents predicted. The capability is real; the absorption is not automatic. What actually determines whether the transformation goes well—whether it is broadly beneficial, whether it compounds or collapses—is what human beings and institutions do with it. The technology sets the terms; it does not write the outcome.

But this time is not infinite, and it must be used wisely. History also teaches that choices made in the early stages of a transformation have outsized effects. Path dependencies set in. Opportunities close. The institutions and norms established early tend to persist.

So we face a paradox: the transformation is slower than it feels, but the time for shaping it is shorter than we might think. Historical consciousness helps us hold both truths simultaneously.

What History Is For

History does not repeat. But it does instruct.

It instructs us that transformative technologies are absorbed into existing social structures in ways that neither technologists nor their critics anticipate. It instructs us that capability and outcome are not the same thing—that a technology can be powerful and real and still produce results that confound the predictions of its champions and its skeptics alike. It instructs us that the process takes decades and involves conflict, adaptation, and the development of new literacies and institutions. It instructs us that choices made early matter disproportionately, but that the future remains open and contingent.

It instructs us that confident predictions—whether utopian or dystopian—are almost always wrong in their specifics, even when they capture something true about direction. It instructs us that the human capacities that matter most in periods of transformation are adaptability, judgment, and the ability to navigate uncertainty.

And it instructs us that those who know history have an advantage—not because they can predict the future, but because they have seen how futures unfold, and they carry that knowledge as a resource for navigating the future that is arriving now.

And it instructs us—this is the point with which I want to close—that historical thinking is not the private property of humanists defending their turf. It is an analytical discipline that economists, sociologists, and policy makers reach for when the questions get hard enough. The questions around AI are exactly that hard. The people arguing that history has nothing to teach us about this moment are not exhibiting rigor. They are exhibiting impatience.

This is why historical training is not a luxury. It is preparation for exactly the moment we are in—a moment when the past does not provide a template but does provide orientation, perspective, and the kind of wisdom that comes only from studying how humans have navigated transformation before.

We are not the first to face a world in upheaval. We will not be the last. But we can learn from those who came before us, and we can offer something to those who come after.

That is what history is for.

The Human Element is a monthly newsletter on humanities and durable skills in an age of artificial intelligence. Next month: “The Interpretation Layer”—on the work that happens between raw output and meaningful action, and why it remains irreducibly human.

“AI ethics is a technical problem. We need engineers who can audit algorithms for bias and build in safeguards. Philosophy majors need not apply.”

This view is understandable. When we discover that a hiring algorithm discriminates or a facial recognition system fails on certain populations, the fix seems technical—adjust the training data, refine the model, test for disparate impact. Engineers created the problem; engineers can solve it.

But this fundamentally misunderstands what ethical reasoning requires. The technical fixes are necessary, but they are not sufficient. And organizations are discovering, often painfully, that they face an ethics bottleneck they don’t know how to address.

The Ethics Bottleneck

Every significant AI deployment decision is, at its core, an ethical decision.

Should we use this algorithm to screen job applicants? That’s a question about fairness, opportunity, and what we owe to candidates. Should we deploy facial recognition in this context? That’s a question about privacy, consent, and the balance between security and freedom. Should we use AI to moderate content on our platform? That’s a question about speech, harm, responsibility, and who gets to decide what counts as each.

These are not technical questions with technical answers. They’re human questions that require human judgment—judgment informed by careful thinking about values, competing goods, unintended consequences, and what we owe to one another.

And here’s the problem: we have a shortage of people equipped to reason well about these questions. Not a shortage of people with opinions. A shortage of people with the training to think through ethical complexity systematically—to identify the values at stake, to reason about tradeoffs, to anticipate consequences, to articulate and defend a position while remaining open to counterarguments.

This is the ethics bottleneck. The technology is racing ahead. The capacity for ethical reasoning is not keeping pace.

What Covid Revealed

I served as a university president during the pandemic, and nothing in my career better illustrated the centrality of ethical reasoning to leadership.

Every decision we faced was an ethical decision dressed in operational clothing. Should we send students home or keep them on campus? That wasn’t just a public health question—it was a question about our obligations to students, to employees, to the surrounding community, about who bore what risks and who got to decide. Should we require vaccination? That involved competing values around individual autonomy, collective welfare, and institutional responsibility. Should we hold graduation in person, delay it, or move it online? Behind the logistics lay questions about what rituals mean, what we owed to students who had sacrificed so much, and how to weigh symbolic value against physical risk.

The technical inputs mattered—epidemiological data, public health guidance, legal constraints. But the data didn’t make the decisions. The decisions required judgment about values, and reasonable people with the same data reached different conclusions because they weighed the values differently.

What I noticed during those months was that some people were equipped for these conversations and others were not. The ones who were equipped could articulate the values at stake, reason about tradeoffs, hold multiple perspectives in mind, and remain thoughtful under pressure. They could disagree respectfully and change their minds when presented with good arguments. They understood that ethical decisions are not math problems with correct answers but exercises in practical wisdom.

Where had they developed these capacities? Not, for the most part, in technical training. They had developed them through education that took ethics seriously—through philosophy courses, through humanities seminars that required wrestling with competing interpretations, through exposure to frameworks for moral reasoning and practice applying them.

The Board View

Now, as a university board member, I see these same dynamics playing out in institutional governance.

Boards are increasingly being asked to weigh in on AI strategy, data governance, and technology ethics. How should the institution use AI in admissions? In advising? In research? What data should we collect on students, and how should we use it? What are our obligations around algorithmic transparency?

These questions land on boards composed largely of successful professionals—people with deep expertise in business, law, finance, medicine. Many of them are extraordinarily capable. But ethical reasoning is not a skill that business success automatically confers. And I watch board members struggle with questions that require precisely that skill.

The struggle isn’t a lack of intelligence or goodwill. It’s a lack of practice. Many board members have spent careers in environments where ethical considerations were either downstream of financial considerations or handled by compliance departments. They are not accustomed to treating ethics as a domain requiring rigorous thinking in its own right.

The institutions that navigate these questions well are the ones with leaders who have that capacity—who can facilitate conversations about values, who can help groups reason together about competing goods, who can articulate frameworks that make ethical tradeoffs legible. These leaders are in short supply.

What Employers Don’t Know How to Ask For

In my advisory work connecting higher education with employers, I’ve observed a pattern: organizations recognize they need people who can think through ethical complexity, but they have no idea how to hire for this capacity.

Job descriptions ask for “integrity” or “ethical judgment” as if these were fixed traits rather than developed capacities. Hiring processes assess technical skills rigorously and ethical reasoning not at all. The interview question “Tell me about an ethical dilemma you faced” typically evaluates whether the candidate gives an acceptable-sounding answer, not whether they can actually reason well about ethics.

And when I talk with employers about how graduates develop ethical reasoning—that it’s strengthened through practice, through engagement with ethical frameworks, through exactly the kind of education humanities programs provide—they often seem surprised. They’ve absorbed the message that philosophy and literature are impractical. It hasn’t occurred to them that the practical capacity they desperately need is developed in precisely those “impractical” disciplines.

This is a translation problem, but it also is an awareness problem. Employers need to understand that ethical reasoning is a skill, that skills are developed through training and practice, and that the humanities are where that training has traditionally occurred. Without this understanding, they’ll keep experiencing the ethics bottleneck without knowing how to address it.

The Bias Problem Is a Human Problem

Consider algorithmic bias, one of the most discussed AI ethics issues. A hiring algorithm trained on historical data learns to discriminate against women or minorities. A facial recognition system performs poorly on darker skinned faces. A healthcare algorithm systematically underestimates the needs of Black patients.

The technical diagnosis is clear: the training data reflected existing biases, and the model learned them. The technical fix is also clear: audit for bias, adjust the data, test for disparate impact.

But beneath the technical problem lies a human problem. Who decides what counts as bias? Who determines which disparities are acceptable and which are not? Who weighs the tradeoffs between different fairness criteria that sometimes conflict with each other? Who considers the interests of people affected by these systems, many of whom have no voice in their design?

These are not questions that algorithms can answer. They require human judgment informed by an understanding of justice, fairness, and competing conceptions of equality. They require the ability to think carefully about who is affected by our decisions and what we owe them. They require, in short, ethical reasoning.

The engineers building these systems are often acutely aware of this. They know they’re making value-laden choices and often feel unequipped to make them. They ask for guidance and find that their organizations don’t have people who can provide it. The ethics bottleneck is felt on the engineering floor as much as in the boardroom.

The Surveillance Question

Or consider surveillance—the deployment of AI systems that monitor, track, and predict human behavior.

The technical capabilities are extraordinary. AI can analyze video feeds, track movements, recognize faces, predict behavior patterns, identify anomalies. These capabilities have legitimate uses in security, safety, and efficiency. They also have profound implications for privacy, autonomy, and the kind of society in which we want to live.

Deciding how to deploy surveillance technologies requires reasoning about values that exist in tension. Security and privacy. Efficiency and autonomy. Institutional needs and individual rights. There are no formulas that resolve these tensions. They require judgment—informed judgment that understands what’s at stake and can reason carefully about tradeoffs.

This is philosophical work, whether we call it that or not. The question “How much surveillance is too much?” is not answered by more data. It’s answered by thinking carefully about human dignity, freedom, trust, and the kind of relationships we want between institutions and individuals.

Organizations deploying these technologies need people who can lead these conversations. Most don’t have them.

Milton’s Relevance

This might seem far removed from Renaissance poetry, but I’ve found unexpected connections.

Paradise Lost is, among other things, a twelve-book meditation on the conditions of ethical action. Milton’s God creates humans with free will, knowing they will fall, because a choice that is not genuinely free is not meaningful. The poem wrestles with what it means to choose, what responsibility requires, and how we act rightly in a world where our choices have consequences we cannot fully foresee.

Satan appears grandiose in the early books—his speeches are magnificent, his defiance stirring. Generations of readers have been seduced by him. But careful reading reveals what careless reading misses: Satan is diminished throughout the poem, his rhetoric increasingly hollow, his apparent freedom actually bondage to his own pride and resentment. He begins as an archangel and ends as a serpent, condemned to hiss. The poem teaches us to distrust compelling surfaces, to read more carefully than Satan’s rhetoric invites us to read.

This is ethical training of a particular kind. Not a set of rules to follow, but practice in moral attention—in noticing how self-deception operates, how rationalization works, how what presents itself as freedom can be its opposite. The Fall that is the subject of Milton’s epic itself is a case study in how reasonable-seeming choices lead to catastrophic consequences, how easily we convince ourselves that what we want to do is what we ought to do.

I don’t generally quote Milton in boardrooms. But the habits of mind his poem develops—the capacity to see past compelling surfaces, to question grandiose rhetoric, to attend carefully to how choices actually unfold—these have served me in every ethical conversation I’ve navigated as a leader. The poem is training in moral discernment, and moral discernment is what the ethics bottleneck demands.

Developing Ethical Reasoning

If ethical reasoning is a capacity that must be developed, how is it developed?

Not primarily through codes of conduct or compliance training, though these have their place. You don’t learn to reason well about ethics by memorizing rules any more than you learn to write well by memorizing grammar. The rules matter, but facility with them requires practice.

Ethical reasoning develops through engagement with hard cases—situations where values conflict, where reasonable people disagree, where the right answer is not obvious. It develops through exposure to different ethical frameworks and practice applying them. It develops through conversation with others who reason well about ethics and can model what good ethical thinking looks like. It develops through wrestling with powerful texts that embody such thinking—texts that don’t give easy answers but demonstrate how to grapple seriously with hard questions.

This is what humanities education provides. A philosophy course that works through trolley problems and their variants is training in ethical reasoning. A literature course that explores how novels present moral complexity is training in ethical reasoning. A history course that examines how past societies navigated ethical challenges is training in ethical reasoning.

The training doesn’t produce people who always agree. It produces people who can disagree well—who can articulate their positions, consider counterarguments, and engage in the kind of collective reasoning that complex decisions require.

The Practical Imperative

This is not an abstract concern. Organizations are facing ethical decisions about AI right now, often without the capacity to make them well.

They’re deploying hiring algorithms without adequately considering fairness implications. They’re implementing surveillance systems without thinking through privacy tradeoffs. They’re using AI for content moderation without frameworks for navigating free speech and harm. They’re making decisions that will affect millions of people, and they’re making them with ethical reasoning capacity that doesn’t match the stakes.

The consequences are already visible. Public backlash against biased algorithms. Regulatory pressure mounting in response to surveillance overreach. Trust eroding as people sense that institutions are making consequential decisions without adequate ethical reflection.

The organizations that thrive in this environment will be the ones that develop ethical reasoning capacity—that hire people who can think well about values, that create cultures where ethical considerations are integrated into decision-making rather than relegated to compliance, that treat ethics as a domain requiring investment and development like any other core competency.

Closing the Gap

So how do we close the ethics bottleneck?

For educators, it means recognizing that ethical reasoning is one of the most practical things we teach—and being more intentional about both developing it and helping students articulate it. Every discipline can contribute to this. The humanities have particular depth to offer, but ethical reasoning should be woven throughout all disciplines in higher education.

For employers, it means understanding that the capacity they need is developed somewhere, and that “somewhere” includes philosophy departments and literature seminars. It means redesigning hiring processes to assess ethical reasoning rather than checking boxes. It means investing in ongoing development of this capacity, not just one-time training.

For boards and senior leaders, it means building teams with genuine ethical reasoning capacity, not just technical expertise and business acumen. It means creating space for ethical reflection in strategy and governance, not treating it as an afterthought or a compliance function.

For individuals, it means recognizing that ethical reasoning is a skill that can be developed—and pursuing opportunities to develop it, whether through formal education, reading, or seeking out communities where serious ethical conversation happens.

The technology will keep advancing. The ethical questions will keep multiplying. The only way through is to develop the human capacity to reason well about them.

That capacity is not a luxury. It’s a bottleneck. And clearing it is among the most urgent practical challenges we face.

The Human Element is a monthly newsletter on humanities and durable skills in an age of artificial intelligence. Next month: “History Doesn’t Repeat, But It Does Instruct”—on how historical thinking provides essential perspective on technological transformation.

Misconception of the Month

“Nobody has time to read whole books anymore. We need to teach students to extract key points efficiently—that’s what the modern workplace demands.”

This view has gained currency not just because of AI but because of broader shifts in how we process information. We skim, scan, excerpt, summarize. We’ve developed elaborate techniques for getting the gist without the commitment. And now AI offers the ultimate extraction tool: feed it a book and receive the main points in seconds.

But something essential is lost when we treat reading as extraction. And what’s lost turns out to be precisely what makes reading valuable in the first place.

Reading in an Age of Summaries

I want to make an unfashionable argument: that slow, careful, cover-to-cover reading is not a luxury or an anachronism but a discipline that develops capacities available no other way. And that in an age when summaries are instantly available, the ability to read deeply becomes more valuable, not less.

This is not nostalgia. It is not a longing for a pre-digital past. It’s an observation about what actually happens to the mind when it engages in sustained reading—and what doesn’t happen when reading is replaced by extraction.

What Happens When We Read

Reading a difficult text slowly is a different cognitive act than extracting information from it.

When I taught Milton’s epic, Paradise Lost, I required students to read the entire poem—all twelve books, over ten thousand lines of blank verse. This was not popular at first. The poem is long. The syntax is demanding. Milton’s sentences unspool across multiple lines, delaying resolution, requiring the reader to hold grammatical structures in mind while moving forward. It’s work that takes focus.

But something happens in that work. The difficulty is not an obstacle to understanding; it’s the mechanism of understanding. Milton’s syntax enacts the experience of the Fall—the way we lose our footing, the way meaning emerges only through sustained attention, the way premature closure leads us astray. You cannot get this from a summary. The experience of reading is the meaning.

Students who persevered through Paradise Lost developed capacities they didn’t know they were developing. They became able to hold complexity in mind without rushing to resolution. They learned to tolerate ambiguity, to sit with difficulty, to trust that meaning would emerge if they stayed with the text. They practiced, in a concentrated form, the discipline of sustained attention that every complex endeavor requires.

This is not content that can be transferred. It’s a capacity that must be developed through practice. And the practice is the reading itself.

The Extraction Trap

The extraction mindset treats texts as containers of information. The goal is to get the information out as efficiently as possible. Once extracted, the container can be discarded.

This works tolerably well for certain kinds of texts—instructions, reports, straightforward arguments where the value really is in the propositional content. But it fails catastrophically for texts whose value lies elsewhere: in their form, their rhetoric, their way of modeling thinking, their capacity to change the reader.

Ask AI to summarize Paradise Lost and you’ll get something like: “Epic poem about the Fall of Man. Satan rebels against God, is cast into Hell, tempts Eve, humanity falls, but redemption is promised.” This is accurate. It is also useless for developing any of the capacities the poem develops in those who actually read it.

The same is true of Donne’s poetry. I could tell you that the poem “A Valediction: Forbidding Mourning” compares two lovers to a compass, with one as the fixed foot and one as the moving foot. You now have the information. But you haven’t experienced the conceit unfolding, the way the image earns its strangeness through the poem’s argument, the moment when the compass becomes not just a comparison but a revelation about what love can be. The experience of reading—the temporal unfolding, the resistance and then surrender to the image—is where the poem does its work.

We’ve become so focused on what texts contain—and the seeming inability of our current generation of students to read substantive texts—that we’ve forgotten what texts do and why they matter. And what the best texts do cannot be summarized.

Attention as a Discipline

There is a reason we use the word “discipline” for both academic fields and practices of self-regulation. The disciplines—history, literature, philosophy—are disciplines in both senses. They require and develop disciplined attention.

Sustained reading is increasingly countercultural. Our information environment is designed to fragment attention, to serve content in bite-sized pieces, to reward skimming and scrolling. The ability to focus on a single complex text for hours is no longer a default capacity; it’s an achievement, something that must be cultivated against the grain of our technological environment.

This is precisely why it’s valuable. In an economy of fragmented attention, the capacity for sustained focus becomes rare and therefore precious. The person who can read a complex report carefully, who can sit with a difficult problem without distraction, who can give deep attention to another person in conversation—this person has a capacity that most people are losing.

The humanities develop this capacity systematically, especially if we engage students in substantive reading. A student who has spent weeks with Paradise Lost, returning to it again and again, living inside its language, has practiced sustained attention in a way that transfers to every subsequent challenge. The content of Milton’s epic may or may not be directly relevant to their later work. The capacity for attention so rare in our current moment certainly will be.

Reading as Relationship

Here is something that sounds mystical but is actually quite practical: reading a great book carefully is a form of relationship.

You spend time with another mind. You learn its rhythms, its assumptions, its characteristic moves. You argue with it, are persuaded by it, resist it, return to it. Over time, you internalize something of how that mind works. It becomes part of your own cognitive repertoire.

I have spent so much time with Donne and Milton that their ways of thinking have become part of how I think. Donne’s habit of yoking unlikely things together—his metaphysical conceits—shaped how I see unexpected connections in my own work. Milton’s ability to hold the cosmic and the intimate in the same frame, to move between scales without losing coherence, influenced how I approach complex institutional problems. These aren’t things I learned from them in the sense of information transfer. They’re capacities I developed through long relationship with their minds.

This is what we mean when we talk about the humanities as formation, not just information. You don’t just learn about great thinkers; you learn to think with them, and in so doing, you become capable of thoughts you couldn’t have had before.

No summary provides this. No extraction captures it. It requires time, attention, and the willingness to let another mind work on yours.

The Paradox of Efficiency

The efficiency argument for extraction and summarization contains a hidden assumption: that the goal of reading is to acquire information, and that faster acquisition is therefore better.

But if the goal of reading is development—of attention, judgment, interpretive capacity, the ability to think with complexity—then efficiency is the wrong metric. You cannot efficiently develop these capacities any more than you can efficiently build physical strength. The time is part of the process. The difficulty is part of the point.

I think of it like the difference between taking a helicopter to a mountain summit and climbing it. Both get you to the top. Only one makes you a climber. The person who took the helicopter has the view; the person who climbed has the capacity.

Students who extract key points from texts they never actually read are taking the helicopter. They can report the view. They cannot do what climbers can do. And in a world where AI can take anyone to any summit instantly, the capacity to climb becomes the differentiating factor.

What We Lose

Let me be concrete about what’s lost when slow reading disappears.

We lose the ability to follow complex arguments. Arguments that unfold over many pages, that require holding premises in mind while working through implications, that build to conclusions that can’t be grasped in isolation—these become inaccessible to readers who can only extract.

We lose sensitivity to rhetoric. The ability to notice how a text is working on you, what techniques of persuasion it’s deploying, where it’s being careful and where it’s eliding—this comes from close attention to language, not from summaries that strip the rhetoric away.

We lose the experience of being changed by a text. The books that matter most are the ones that rearrange something in us, that we come out of differently than we went in. This requires submission to the text’s temporality, letting it work on us at its own pace. Extraction prevents this.

We lose the capacity for boredom, which is also the capacity for depth. The person who cannot tolerate the slow parts of a long book cannot tolerate the slow parts of a complex project, a long relationship, a difficult organizational transformation. The discipline of staying with something that isn’t immediately rewarding is transferable.

We lose, ultimately, the ability to read at all—in the deepest sense of that word. We become capable only of processing, scanning, extracting. The texts remain closed to us even when we’ve captured their key points.

The Case for Required Difficulty

In my years as a university president and now as a board chair, I’ve thought a lot about curricular requirements—what we ask students to do, and why.

There’s always pressure to make requirements less demanding, to accommodate students who are busy, who work, who have competing obligations. These pressures are real and often legitimate. But they can also erode exactly what makes education transformative.

Some things are valuable precisely because they are difficult and time-consuming. You cannot get the benefits of deep reading without doing deep reading. There’s no shortcut, no hack, no efficient alternative. The difficulty is not an obstacle to the benefit; it is the benefit.

I recognize that this is hard to defend in a culture that valorizes efficiency and questions anything that takes time. But it’s true, and we do students no favors by pretending otherwise. The capacity for sustained attention, developed through practice with difficult texts, will serve them for decades. The time saved by summarizing or giving in to their unwillingness or inability to engage with substantive texts will not.

Reading and Leadership

I want to make one more connection, because it matters for how we think about the practical value of deep reading.

Every leadership role I’ve held has required the ability to read complex situations carefully. Not to extract key points and move on, but to sit with ambiguity, notice what’s not being said, understand how different constituencies see the same situation, resist the pressure to resolve complexity prematurely.

This is reading in the broadest sense—the interpretive disposition that close reading develops. A leader who can only skim will skim their organization, their people, and their challenges. They’ll extract key points and miss everything that matters.

The leaders I’ve most admired are deep readers in this sense. They pay attention. They notice. They resist the simplification that loses essential complexity. They’re willing to stay with difficulty until understanding emerges.

These capacities were developed somewhere. For many of them, they were developed in exactly the kind of sustained engagement with difficult texts that we’re now told is a luxury we can’t afford. Those texts may not have been Donne or Milton, or even texts as traditionally understood. They may be complex problems that resist easy solutions. The experience is very similar to what I’ve described, and just as important.

Indeed, we can’t afford not to afford it.

An Invitation

I’ll end with an invitation rather than an argument.

Choose a book you’ve been meaning to read—something substantial, something that will take weeks rather than hours. Commit to reading it slowly, without skipping, without extracting. Give it your full attention for a period each day. Let it work on you.

Notice what happens. Notice the resistance, the boredom, the frustration, and then—if you stay with it—the breakthrough into something else. The sense of another mind becoming available to you. The capacity for attention strengthening like a muscle under load.

This is what deep reading offers. It cannot be summarized. It can only be experienced. And it remains, in an age of infinite summaries, more valuable than ever.

The Human Element is a monthly newsletter on humanities and durable skills in an age of artificial intelligence. Next month: “The Ethics Bottleneck”—why every AI deployment decision is fundamentally an ethical decision, and why we’re short on people equipped to make them.

Misconception of the Month

“Why spend years studying literature or history when you can just ask ChatGPT? AI makes all that knowledge instantly accessible to anyone.”

This is perhaps the most seductive misconception of all—seductive because it contains a kernel of truth. AI does make information accessible. You can ask it about the French Revolution, the themes of Middlemarch, the arguments in Plato’s Republic, and receive competent summaries in seconds.

But confusing access to information with the capacity to use it well is a profound category error. And it reveals a fundamental misunderstanding of what humanistic education actually provides.

The Prompt Is Not the Skill

Here is a simple experiment. Give two people access to the same AI system and the same task: research a complex topic and produce a useful synthesis. One person has spent years developing their capacities for inquiry, interpretation, and judgment. The other has not.

Watch what happens.

The first person formulates precise questions that cut to the heart of what they need to know. They recognize when the AI’s response is superficial, incomplete, or subtly wrong. They probe further, refine their queries, triangulate across sources. They know what to do with the information once they have it—how to weigh conflicting accounts, how to situate claims in context, how to construct an argument that holds together. The output is genuinely useful.

The second person asks vague questions and accepts the first response. They can’t tell whether the AI has given them something accurate or plausible-sounding nonsense. They don’t know what they don’t know, so they don’t know what else to ask. The output looks like information but doesn’t function as knowledge.

Same tool. Same access. Vastly different results.

The difference isn’t the prompt—it’s everything behind the prompt. The years of developing judgment about what questions matter. The practice in evaluating sources and arguments. The interpretive skill that recognizes nuance, limitation, and implication. The capacity to think with information rather than merely receive it.

This is what humanistic learning develops. And it’s precisely what determines whether AI access translates into actual capability.

The Illusion of Frictionless Knowledge

We’ve been here before. Every information technology has prompted predictions that education would become unnecessary because knowledge would be freely available.

The printing press was supposed to make teachers obsolete—why sit in lectures when you could read books? Radio and television would bring the world’s best instructors into every home. The internet would democratize learning completely; everything humanity knew would be available to anyone with a connection.

While each technology did expand access, none eliminated the gap between access and capability. If anything, each expansion made the gap more consequential. When everyone has access to the same information, the differentiating factor becomes what you can do with it.

I watched this unfold across my years as a university president. Our students arrived with more information at their fingertips than any previous generation. They could look up any fact, access any text, find any dataset. What they often couldn’t do—what they came to us to learn—was formulate good questions, evaluate what they found, synthesize across sources, construct original arguments, and communicate their understanding to others. The information was free. The capability was not.

AI amplifies this dynamic dramatically. The floor of accessible information has risen higher than ever. A competent summary of almost any topic is now available instantly. This makes the ceiling—what you can do with information, how you think with it and beyond it—more important than ever.

What Questions Require

Let me dwell on the question-formulation problem, because it’s more profound than it first appears.

Good questions are not obvious. They are achievements. They require understanding a domain well enough to know where the interesting problems lie. They require recognizing what you don’t know—which itself requires knowing quite a lot. They require a sense of what would count as a satisfying answer, what evidence would be relevant, what level of precision is appropriate.

When I taught Donne’s poetry and Milton’s Paradise Lost, students often arrived thinking they knew how to ask questions. They’d been asking questions their whole educational lives. But their questions tended to be either too broad to be answerable (“What does this poem mean?”) or too narrow to be interesting (“What year was this written?”). Learning to ask good questions—questions that opened the text rather than closing it down, questions that actually could be pursued—was some of the hardest work they did.

With Paradise Lost, I’d watch students transform over the course of a semester. They’d arrive asking whether Satan was the hero—a reasonable starting point, a question that has animated centuries of readers. But that question, asked naively, goes nowhere. Asked well, it opens into deeper inquiries: What does Milton mean by heroism? How does the poem’s structure shape our sympathies? What happens when we recognize that we’ve been seduced by Satan’s rhetoric in exactly the way Eve was? The movement from the naive question to the sophisticated one is the movement from access to capability. No AI summary can make it for you.

This capacity transfers. In every complex domain, the ability to formulate productive questions separates those who can navigate the complexity from those who flounder. For what should we be optimizing? What are we not seeing? What would change our mind? What assumptions are we making? These are not queries you can simply type into a system. They emerge from disciplined thinking about thinking, from practice in inquiry itself.

AI can answer questions. It cannot yet ask the right ones. That remains a human capacity—and a trained human capacity, not an innate one.

The Evaluation Problem

Even more challenging than formulating questions is evaluating answers.

AI systems produce confident-sounding outputs that are often correct, sometimes incomplete, and occasionally fabricated. They present information without the epistemic markers that help humans calibrate confidence—the hedging, the citation of sources, the acknowledgment of uncertainty that a knowledgeable human would provide. Everything comes with the same polished assurance.

Evaluating these outputs requires independent knowledge. You need to know enough about a domain to recognize when something is wrong, or missing, or subtly misleading. You need the interpretive skill to notice what isn’t being said. You need judgment about what level of verification is required for different stakes.

This is precisely the capacity that humanities education develops through years of engagement with texts, arguments, and interpretive problems. A student who has spent a semester evaluating primary sources in a history seminar has practiced, dozens of times, the work of assessing reliability, identifying bias, recognizing gaps, and triangulating across accounts. A student who has written papers that required defending interpretations against counterarguments has internalized the habit of asking “but is this actually true?”

Without this training, AI becomes an oracle—a source of answers that are accepted because they sound authoritative. With this training, AI becomes a tool, powerful but requiring judgment to use well.

The Iteration Gap

Sophisticated use of AI is not a single query but a process—an iterative dialogue of question, evaluation, refinement, and deeper inquiry.

Watch an expert use AI for research. They don’t ask once and accept. They probe, test, redirect. They notice when a response is generic and push for specificity. They catch inconsistencies and demand clarification. They use the AI’s outputs as starting points for their own thinking, not as conclusions.

This iterative capacity is a skill. And it looks remarkably similar to what happens in a good seminar. A student offers an interpretation. The professor probes: What’s your evidence? How do you account for this counterexample? What are the implications of that claim? The student refines, extends, sometimes abandons the original position. Understanding emerges through dialogue, through the friction of engaged inquiry.

AI can simulate this dialogue, but only if the human brings genuine intellectual engagement to it. The AI won’t spontaneously challenge your assumptions or point out that you’re asking the wrong question. That requires a human mind trained in critical inquiry—someone who has internalized the probing voice that says, “but is that really true?” and “what am I missing?”

The prompt is the visible part of this process. But the skill lies in everything the prompt represents: the thinking that shaped it, the evaluation that follows it, the iterative refinement that improves it. These are not things AI provides. They’re things you bring to AI. And they’re developed through exactly the kind of sustained humanistic learning that AI supposedly makes unnecessary.

The Integration Challenge

Perhaps the deepest skill is knowing what to do with information once you have it.

Information is not knowledge. Knowledge is not wisdom. The gap between them requires integration—connecting new information to existing understanding, situating it in context, grasping its implications, determining what should be done about it.

I think often about the difference between a student who has summarized a historical period and one who can think historically. The first can recite events and dates. The second understands contingency—that things could have gone otherwise, that outcomes we take for granted emerged from choices that were not inevitable. The second can use historical understanding to illuminate the present, recognizing patterns and possibilities that the first would miss.

AI can provide summaries. It cannot provide the integration that makes information useful. That requires a mind prepared to receive it—a mind with frameworks, contexts, and habits of connection that transform data into understanding.

Milton himself was a master of integration. Paradise Lost synthesizes classical epic and Christian theology, contemporary politics and cosmological speculation, the literary traditions of Homer and Virgil with the theological concerns of Genesis. Milton moved horizontally across domains that others treated as separate, creating something that transcends any single category. Studying how he did this—how he brought together what had been apart—shaped how I see intellectual problems, how I recognize opportunities for connection that specialists might miss.

My preparation as a Renaissance scholar gave me such frameworks. The Renaissance was itself a period of integration, of horizontal connection across domains that had been separate. Studying how Donne brought together theology and science, how Milton wove politics into his epic, how courts and academies became spaces where knowledge circulated across fields—this shaped how I see the world. When AI provides me with information, I have somewhere to put it. The information enters a prepared mind. This is what education provides that access alone cannot.

The Honest Case for AI in Education

None of this is an argument against AI. It is an argument for understanding what AI actually changes.

AI should transform how we teach, but not by making humanistic learning obsolete. It should make us more intentional about developing the capacities that matter—the question-formulation, the evaluation, the integration, the iterative refinement that determine whether AI access translates into actual capability.

Used well, AI can accelerate learning. It can provide practice at evaluation when students fact-check its outputs. It can enable more ambitious research by handling mechanical tasks. It can serve as a tireless interlocutor for those developing arguments.

But “used well” is doing a lot of work in that sentence. Using AI well is itself a sophisticated skill, and it’s built on exactly the capacities humanities education develops. The students best positioned to leverage AI are those who know how to think—who can formulate questions, evaluate answers, iterate productively, and integrate information into understanding.

The ones without this training will be at AI’s mercy, unable to distinguish its genuine capabilities from its confident confabulations. They’ll have access to the tool without the capacity to wield it.

The Leverage Paradox

Here is the paradox at the heart of this moment: the better AI gets at providing information, the more valuable the human capacities around information become.

When summaries are free, judgment about what to summarize becomes precious. When answers are instant, the quality of questions becomes decisive. When content can be generated endlessly, the ability to evaluate and curate becomes essential.

This is the leverage paradox. AI provides leverage—it amplifies human capability. But leverage works both ways. It amplifies the capability of those who have developed strong foundational skills. It also amplifies the confusion of those who haven’t.

We’re already seeing this divergence. Some people are using AI to become dramatically more effective—researching faster, writing more, thinking more ambitiously. Others are using AI to avoid thinking, accepting its outputs uncritically and producing work that looks complete but collapses under scrutiny. The technology is the same. The human capacities differ.

The most practical thing a student can do right now is develop the capacities that will determine on which side of this divergence they end up. Those capacities are not AI skills—there’s no course in prompt engineering that will substitute for judgment. They are thinking skills, interpretation skills, inquiry skills. They are what humanities education has always provided.

The Real Skills Gap

The conversation about AI and education has focused on the wrong gap.

We worry about whether students will use AI to cheat—to generate essays without learning, to shortcut the work of understanding. This is a real concern, but it’s a symptom of a deeper problem: we’ve oriented too much of education around outputs that AI can now produce, rather than capacities that AI cannot replicate.

The real gap is between those who will be able to use AI as a powerful tool for genuine intellectual work and those who will be used by it—manipulated by its confident errors, dependent on its outputs, unable to think independently when the tool fails or misleads.

Closing this gap requires doubling down on exactly what makes humanistic learning valuable: the development of judgment, the practice of inquiry, the cultivation of critical evaluation, the integration of knowledge into wisdom.

The prompt is not the skill. The skill is everything behind the prompt—everything that determines whether technology serves human purposes or substitutes for human thinking. That skill is developed through years of practice, guided by those who have developed it themselves, in exactly the kind of educational environments we’re being told are obsolete.

They are not obsolete. They are essential. More essential than ever.

The Human Element is a monthly newsletter on humanities and durable skills in an age of artificial intelligence. Next month: “Reading in an Age of Summaries”—on the irreplaceable value of slow, careful reading when everything can be compressed.

Misconception of the Month

“Sure, communication and teamwork matter, but those are soft skills—the hard skills are what really count in the job market.”

You hear this everywhere: in career advice, in curriculum debates, in the quiet hierarchy that places engineering above English, accounting above art history. “Soft skills” are acknowledged as nice to have, a social lubricant that makes the real work go more smoothly. But the real work—the hard skills—is where the value lies.

This framing has it exactly backward. And the language itself is part of the problem.

What We Talk About When We Talk About “Soft Skills”

Let’s start with the terminology, because terminology shapes perception.

“Soft skills” implies something yielding, secondary, vaguely feminine in the old dismissive sense. Soft versus hard. Optional versus essential. The stuff you can’t quite measure versus the stuff that shows up on a test. The language itself encodes a hierarchy that has little to do with actual difficulty or actual value.

Consider what gets categorized as “soft”: communication, critical thinking, collaboration, adaptability, ethical reasoning, judgment under uncertainty, the ability to navigate ambiguity and work across difference. Now consider what these capacities actually require. Years of practice. Constant calibration to context. The ability to read situations, audiences, and implications that are never fully explicit. Failure modes that are subtle and consequential.

These are not soft. They are, in fact, brutally hard—which is precisely why so many people and organizations struggle with them, and why their absence causes so much damage.

I spent years teaching students to read John Donne’s poetry. Donne’s verses are dense, syntactically complex, full of conceits that yoke together wildly disparate domains—astronomy and love, cartography and the soul, death and sleep. To read his poetry well, students had to slow down, tolerate confusion, hold multiple interpretive possibilities in mind simultaneously, and construct meaning through patient attention. None of this was soft. Arguably, it was among the most demanding cognitive work they encountered in their education.

And it was preparation—though they didn’t always recognize it at the time—for every complex situation they would later face. Every negotiation that required reading what wasn’t being said. Every leadership moment that demanded judgment without complete information. Every ethical dilemma where the right answer wasn’t obvious and the stakes were real.

The Measurement Trap

Part of why “hard skills” get privileged is that they are easier to measure. You can test whether someone knows Python. You can verify a credential. You can count the outputs.

But ease of measurement is not the same as importance or difficulty or relevance. And our obsession with the measurable has distorted how we think about capability.

In my work with a national partnership connecting business and higher education, I’ve watched this distortion play out repeatedly. Employers tell us they desperately need people who can think critically, communicate clearly, collaborate across teams, and adapt to change. Then their hiring processes screen far too often for technical keywords and credentials—the things for which they can easily filter. They are optimizing for what they can measure, not what they actually need.

This sounds practical, but the result is predictable. Organizations full of people who passed the measurable tests but can’t run a meeting, can’t write a clear email, can’t navigate disagreement productively, can’t exercise judgment when the situation doesn’t match the template. The soft skills gap turns out to be a performance gap, a leadership gap, an organizational effectiveness gap. It’s not soft at all.

Durable Skills: A Better Frame

Rather than “soft skills,” I prefer the term “durable skills.” It captures something essential that “soft skills” obscures.

Durable skills are the capacities that remain valuable across technological and economic shifts. They are transferable across roles, industries, and contexts. They don’t depreciate when the tools change—they become more valuable, because they are what allow people to adapt to new tools.

Technical skills are often perishable. The programming language you learned five years ago may be declining in relevance. The software platform you mastered gets replaced. The specific procedures of your first job become obsolete. This isn’t a criticism of technical training—we need people with deep technical capabilities. But technical training alone leaves you vulnerable to the very changes for which it was supposed to prepare you.

Durable skills compound over a career. They are evergreen rather than depreciating. The judgment you develop in your twenties becomes the leadership capacity of your forties. The communication skills you honed early make you more effective at every subsequent stage. The ability to learn, interpret, and adapt—capacities the humanities develop systematically—turns out to be the most practical preparation for a world that won’t stop changing.

When I became a university president for the first time, I drew on my training as a Renaissance scholar in ways I hadn’t anticipated. Not the content—I wasn’t quoting Donne or Milton in budget meetings—but the capacities. The ability to interpret complex texts translated into the ability to read complex institutional situations. The tolerance for ambiguity I’d developed wrestling with seventeenth-century poetry served me in circumstances where the data pointed in multiple directions and decisions couldn’t wait for certainty. The practice of understanding how meaning is constructed and contested helped me navigate constituencies with different values and vocabularies.

None of this was soft. It was the hardest and most consequential work of my career.

The AI Amplification

The current moment makes the durable vs perishable distinction more stark than ever.

AI is exceptionally good at the tasks we’ve traditionally labeled “hard skills.” It can code, calculate, process, analyze, and generate outputs that once required significant technical training. The floor has risen. Tasks that used to differentiate candidates are becoming commoditized.

What AI cannot do—and this is not a limitation that will be engineered away—is exercise judgment about what matters, interpret situations in their full human context, communicate in ways that build trust and move people, navigate ethical complexity, or adapt to circumstances that don’t match its training data. These remain human capacities. And as AI handles more of the mechanical work, these human capacities become the differentiating factor.

The irony is sharp. The skills we’ve dismissively called “soft” are the ones that remain hardest to automate. The skills we’ve privileged as “hard” are increasingly handled by machines. Our vocabulary has been pointing us in exactly the wrong direction.

I see this in my advisory work with a skills assessment company that started by measuring technical capabilities. They’ve found that technical skill alone is an incomplete predictor of success. The people who advance, who lead, who create value over time, are the ones who combine technical competence with what we’ve been calling soft skills—and what we should be calling durable skills. The company is now investing in finding ways to assess and develop these capacities, because their clients are demanding it. The market is correcting the vocabulary, even if slowly.

The Development Problem

If durable skills are so valuable, why don’t we develop them more systematically?

Part of the answer is the terminology problem—we’ve labeled them “soft” and thereby signaled they’re secondary. Part of it is the measurement trap—we invest in what we can easily assess. But part of it is a genuine difficulty: durable skills are hard to develop.

You don’t acquire judgment by taking a course in judgment. You develop it through practice in situations that require judgment, ideally with feedback from people who have good judgment themselves. You don’t learn to communicate by memorizing principles of communication. You learn it by communicating in contexts with real stakes—writing that will be read, arguments that will be challenged, presentations that need to move actual audiences.

This is what the humanities have always provided, even when we haven’t described it in these terms. A seminar where you must articulate an interpretation and defend it against counterarguments. A paper where you must write for a reader who won’t fill in your gaps. A difficult text that doesn’t yield to a first reading or a fifth, teaching you that understanding is earned through sustained attention. A historical case that shows how people in the past made decisions under uncertainty, revealing the contingency of outcomes that seem inevitable in retrospect.

These experiences develop capacities that can’t be acquired through content delivery alone. They require the kind of guided practice that higher education, at its best, provides. And they produce graduates who are prepared for the long game—not just the first job, but the fifth job, the career pivot, the leadership role, the challenges we can’t yet anticipate.

The Translation Challenge

Here’s what keeps me up at night: we are failing at translation.

Humanities programs develop durable skills systematically, but we don’t always help students articulate what they’ve gained in terms the professional world recognizes. A philosophy major who has spent years constructing and evaluating arguments often can’t explain in an interview how that translates to strategic thinking. A literature student who has learned to read difficult texts with care doesn’t know how to connect that to the interpretive demands of organizational life.

Meanwhile, employers claim to want durable skills but have hiring processes that screen for proxies and credentials. They are looking for “critical thinking” but filtering resumes for keywords. They say they need “communication skills” but evaluate candidates through technical tests that don’t involve communication.

Both sides need to get better at this. Humanities programs need to be more intentional about helping students translate—not dilute but translate—their capacities into language that resonates in professional contexts. Employers need to examine whether their processes actually select for what they say they value. The skills assessment company I advise is working on this problem: creating ways to surface durable skills, to make them visible in hiring decisions, to help both candidates and employers see past the misleading vocabulary.

This is solvable. But it requires abandoning the soft/hard binary that has distorted our thinking for too long.

What’s Actually Hard

Let me be concrete about what I mean when I say these durable skills are hard.

It is hard to write clearly. Most people cannot do it, including many who have advanced degrees and prestigious positions. Clear writing requires clear thinking, and clear thinking is uncommon.

It is hard to listen well—to actually hear what someone is saying rather than waiting for your turn to speak, to understand the concerns beneath the stated position, to make others feel genuinely understood.

It is hard to exercise judgment under uncertainty, to make decisions when the information is incomplete and the stakes are real, to act without the comfort of certainty.

It is hard to give feedback that is honest and kind, to disagree without damaging relationships, to navigate conflict productively.

It is hard to adapt—to let go of what you know when circumstances change, to learn continuously, to remain effective as the ground shifts beneath you.

It is hard to lead, to take responsibility for outcomes you don’t fully control, to hold people to standards while treating them with dignity, to make the final call when reasonable people disagree.

These capacities take years to develop. They require practice, feedback, and often failure. They separate those who advance from those who plateau. They are in short supply in almost every organization in which and with whom I have worked.

Soft? I don’t think so.

A Proposal

Let’s retire “soft skills.” The term has done enough damage.

Let’s talk instead about durable skills—the capacities that transfer, that compound, that remain valuable as the world changes. Let’s recognize that these skills are genuinely difficult to develop and that developing them systematically is one of the most important things higher education does. Let’s insist that employers align their hiring processes with what they claim to value, and that educators help students translate their capabilities into language the world understands.

And let’s stop apologizing for the humanities as if they were a luxury. They are not. They are where durable skills have always been developed, even when we didn’t call them that. In an age when AI handles more and more of the mechanical work, what remains is the human work—the judgment, the interpretation, the communication, the ethical reasoning, the horizontal thinking that connects domains and sees what specialists miss.

That’s not soft. That’s the whole game.

The Human Element is a monthly newsletter on humanities and durable skills in an age of artificial intelligence. Next month: “The Prompt Is Not the Skill”—why the ability to use AI well is itself a product of exactly the training AI supposedly makes obsolete.

Misconception of the Month

“In the age of AI, we need more STEM graduates, not more people studying philosophy or literature. The humanities are a luxury we can no longer afford.”

This sentiment appears constantly—in op-eds, in state legislature debates over university funding, in university board rooms, in parental advice to college-bound students, in corporate commentary on “workforce needs.” It sounds pragmatic. It sounds like clear-eyed adaptation to a changing world.

It is almost perfectly wrong.

The Misread Obituary

The humanities have been declared dead or dying so many times that the genre has become its own literary tradition. The obituaries follow a predictable pattern: a new technology or economic shift arrives, commentators declare that we now need “practical” skills, and the study of history, literature, philosophy, and languages is deemed a quaint indulgence from a less serious era.

This happened when typewriters arrived. When television arrived. When the personal computer arrived. When the internet arrived. Each time, the prediction was that the new technology would render humanistic education obsolete, replaced by technical fluency in the tool itself.

Each time, the prediction proved backward.

What actually happened was that the new technology increased the value of the capacities humanities education develops—the ability to communicate clearly, to think critically about information, to understand context and audience, to make judgments under uncertainty, to grasp what matters and why. The technology handled the mechanical part. As a consequence, the human part became more important, not less.

We are now living through the most dramatic version of this pattern. Artificial Intelligence (or AI) systems can generate text, summarize documents, answer questions, write code, and produce analysis. The mechanical floor has risen enormously. And this has led, unsurprisingly, to another round of obituaries for the humanities.

But look more carefully at what’s happening, and you see the opposite story emerging.

The Horizontal Mind

I spent my early professional years as a scholar of John Donne—the seventeenth-century English poet, priest, lawyer, diplomat, and occasional pirate-voyage participant. Donne is remembered now mostly for his poetry, those dense, startling verses where lovers become compasses, parting becomes gold beaten to airy thinness, and death itself is told it shall die. But what made Donne extraordinary wasn’t mastery of a single domain. It was his capacity to move acrossdomains—to bring the language of cartography to love, theology to seduction, legal reasoning to devotion.

This wasn’t incidental to his genius. It was the source of it.

Donne operated within a coterie—a network of poets, theologians, statesmen, and scholars who circulated ideas, manuscripts, and problems across what we would now call disciplinary boundaries. The Renaissance, for all its romanticization, was fundamentally an era of horizontal connection. Knowledge moved. People who could move with it—who could see how insights in one domain illuminated problems in another—were the ones who shaped their world.

We are entering another such era, though we don’t always recognize it. AI handles vertical expertise remarkably well. It can go deep. What it cannot do is move horizontally—cannot see that a problem in organizational design rhymes with a problem in Renaissance patronage networks, cannot recognize that a question about AI ethics is also a question about human finitude that poets and theologians have been wrestling with for centuries. The horizontal mind, the mind trained to make connections across disparate domains, is precisely what the humanities develop. And it is precisely what the current moment demands.

The Bottleneck Has Moved

When a technology automates certain tasks, it doesn’t eliminate the need for human judgment—it concentrates that need at different points in the process.

Consider what it takes to use AI systems well. You need to formulate questions that elicit useful responses, which requires understanding what you’re actually trying to learn or accomplish. You need to evaluate the outputs, which requires enough knowledge to recognize quality, accuracy, and relevance. You need to interpret results in context, which requires understanding of situation, audience, and stakes. You need to make decisions about what to do with the information, which requires ethical reasoning and strategic judgment.

Every one of these capacities is precisely what humanities education develops. The bottleneck hasn’t disappeared; it’s moved upstream to the distinctly human activities of framing, evaluating, interpreting, and deciding.

I saw this vividly during my years as a university president—twice, at two very different institutions. The hardest problems I faced were never technical. We could find people who understood budgets, facilities, enrollment management, IT systems. What we needed—and what was always in short supply—were people who could interpret ambiguous situations, communicate across constituencies with different values and vocabularies, make judgment calls when the data pointed in multiple directions, and hold complexity without rushing to false resolution. The students I’d taught years earlier in my Renaissance literature courses, the ones who’d learned to sit with Donne’s paradoxes and find meaning in difficulty, turned out to be unusually well-prepared for this kind of work. They’d been trained to think horizontally, to tolerate ambiguity, to interpret.

Organizations everywhere are discovering this in real time. The problem isn’t generating content—AI can produce unlimited content. The problem is knowing what content should exist, whether the content is good or incomplete, what it means, what is left out, and what should be done about it. These are humanistic questions, and they turn out to be the hard part.

The Durable and the Perishable

There’s a useful distinction between perishable skills and durable skills. Perishable skills are tied to specific tools, platforms, or technical configurations. They have immediate practical value but depreciate as the technology changes. Durable skills are transferable across contexts and remain valuable regardless of which tools are in use. They are evergreen and adaptable.

Knowing how to use a particular software application is a perishable skill. Knowing how to learn new software quickly, how to evaluate whether a tool serves your purposes, and how to communicate effectively regardless of medium—these are durable skills.

The humanities have always been in the durable skills business, even when we didn’t use that language. Reading difficult texts carefully, constructing and evaluating arguments, writing clearly for different audiences, understanding how context shapes meaning, grappling with ethical complexity, recognizing patterns across historical situations—these capacities don’t expire when the technology changes. They become more valuable when the technology changes, because they’re what allow people to adapt, learn, and grow.

My work with a national partnership between business leaders and higher education has made this painfully clear. Employers consistently say they want critical thinking, communication, collaboration, adaptability. They struggle to hire for these qualities, partly because they don’t know how to assess them and partly because they’ve been conditioned to read resumes for technical keywords rather than durable capacities. Meanwhile, humanities graduates often have exactly what employers claim to want—but can’t translate their training into language the hiring process recognizes.

This is a translation problem, not a value problem. And it’s one we can solve.

The Translation Layer

I’ve recently been advising a skills assessment company that began by helping employers evaluate technical capabilities—coding, data analysis, the measurable and testable competencies. What they’ve discovered is that technical skill alone predicts much less than anyone assumed. The people who succeed and advance are the ones who can also communicate clearly, work across teams, exercise judgment, and adapt when circumstances change.

So now they’re exploring something harder: how to make durable skills visible, how to help students and job-seekers articulate what skills and competencies their humanities education gave them in terms that resonate in professional contexts, and how to help employers recognize the very capacities they claim to value.

This approach matters because the obituary narrative has consequences. When students absorb the message that humanities degrees are impractical, they avoid them. When employers absorb the message that humanities graduates lack “real skills,” they screen them out. The result is a self-reinforcing cycle that deprives organizations of exactly the capacities they need, and deprives students of training that would serve them well across a long career.

Breaking this cycle requires work on both sides. Humanities programs need to be more intentional about helping students translate their capacities—not dilute them, but translate them. And employers need to become more sophisticated about what they’re actually seeking and how their selection criteria and processes support those skills and competencies. A hiring process optimized for keyword-matching will produce keyword-matched employees. Whether it will produce people who can lead, adapt, and exercise judgment in a rapidly changing environment is another question.

What the Obituarists Miss

The recurring prediction that humanities education is obsolete rests on a fundamental misunderstanding of what that education actually provides.

The misunderstanding goes something like this: humanities students learn “content”—facts about history, interpretations of novels, philosophical positions. AI can now provide this content instantly. Therefore, humanities education is unnecessary.

But content was never the point. The content is the material through which students develop capacities—capacities for attention, analysis, interpretation, discernment, expression, empathy, and judgment. A history student isn’t primarily learning facts about the past; they’re learning how to evaluate sources, construct narratives from fragmentary evidence, understand causation in complex systems, and recognize how the present emerged from contingent choices. A literature student isn’t primarily learning plot summaries and literary periods and forms; they’re learning how to read attentively, how language works, how meaning is constructed and contested, how emotion is evoked, and how to inhabit perspectives different from their own.

When I taught Donne’s Holy Sonnets or Milton’s Paradise Lost, I wasn’t trying to produce Donne or Milton experts. I was using those poems—their difficulty, their strangeness, their use of language and argument, their refusal to yield easy meanings—to develop students’ capacities for close attention, for sitting with uncertainty, for finding coherence in apparent contradiction. A student who genuinely has wrestled with Donne’s “Batter my heart, three-personed God” or the character of Milton’s Satan, has practiced something that will serve them in every subsequent complex situation they encounter: the discipline of staying with difficulty until meaning emerges, and the ability to resist easy judgement and meaning for the complexity that lies beneath the surface.

These capacities can’t be downloaded. They can’t be prompted into existence. They develop through sustained practice with difficult material, ideally guided by people who have developed these capacities themselves and can model what mastery looks like.

AI makes the content more accessible. It also makes the capacities more valuable. The obituarists see only the first half of this equation.

The Practical Paradox

There’s a deep irony in the call to abandon the humanities for a more “practical” education. The most practical preparation for a world of rapid technological change is precisely the kind of agile, flexible, transferable, judgment-intensive training and capacity development that the humanities provide.

The students who learned to code in a specific language five years ago are now watching that language become less relevant. The students who learned to think clearly, write well, evaluate arguments, and adapt to new contexts are finding those capacities in higher demand than ever.

Now, as a university board chair, I watch this play out at the governance level. The institutions that will thrive in our current rapidly changing environment are not the ones that chase every technological trend, retooling their curricula to match the current moment’s demands. They’re the ones that hold fast to what has always worked—deep training in thinking, communicating, and making meaning—while at the same time adapting how they deliver such an education, engage today’s learners, and translate that broad education for new contexts.

The Renaissance offers a model here. It was an era of extraordinary technological disruption—the printing press alone transformed how knowledge was created, stored, and shared. The people who flourished weren’t the ones who abandoned classical learning for the new technology. They were the ones who brought classical learning to the new technology, who used the new tools to extend and amplify capacities that remained fundamentally human. Horizontal minds, moving across domains, making connections that narrow specialists couldn’t see.

We need those minds now more than ever.

What We Should Actually Be Worried About

None of this means humanities programs are beyond criticism or that they’re optimally configured for the current moment. There are real questions worth asking.

Are humanities programs helping students translate their capacities into language that resonates in professional contexts? Often not well enough. Are they providing enough scaffolding for students to connect their training to concrete applications? Often not. Are they adapting their methods to account for new tools and new contexts? Unevenly.

These are problems of communication and adaptation, not problems with the underlying value of what humanities education provides. The core offering—deep training in interpretation, judgment, and expression—is more relevant than it has ever been. The challenge is making that relevance legible to students, parents, employers, and policymakers who have absorbed the obituary narrative.

That’s part of what this newsletter aims to do. Not to defend the humanities as a nostalgic preserve, but to make the case—clearly and practically—that humanistic education is exactly what a rapidly changing, increasingly automated world demands.

The obituaries are wrong. They’ve always been wrong. And the sooner we recognize this, the better prepared we’ll be for what’s coming.

The Human Element is a monthly newsletter on the humanities and durable skills in an age of artificial intelligence. Next month: “What We Talk About When We Talk About ‘Soft Skills’”—why we need better language for the capacities that matter most.

The humanities have been declared dead so many times that the genre has become its own literary tradition. Each technological shift—typewriters, television, computers, the internet—prompted predictions that humanistic training was now obsolete.

Each time, the prediction proved backward.

We’re living through the most dramatic version of this pattern. AI can generate text, summarize documents, answer questions, produce analysis. The mechanical floor has risen enormously. And the obituaries are back.

But look carefully at what’s actually happening, and you see the opposite story. The bottleneck has moved. The capacities that matter most—framing, evaluating, interpreting, deciding—are precisely what humanities education develops.

More on January 14. Subscribe to receive it directly.

Subscribe now

Subscribe now