In 2017, Dan Kahan, Ellen Peters, Erica Dawson, and Paul Slovic published a study in Behavioural Public Policy with a deliberately uncomfortable design. Participants were given a numerical reasoning task built on a table of results. When the correct interpretation of the table was politically neutral, people with higher numeracy were better at reading it. When the identical numbers were framed so that the correct answer contradicted the participant’s ideological group, the pattern inverted: the most numerate participants were the most likely to reach the wrong conclusion in the direction their group preferred. Skill at reasoning did not make people more accurate in the cases where accuracy was costly. It made them more effective at defending their side.
That finding compresses the subject of this article into a single experiment. The intuition behind “intelligence is not wisdom” is usually conveyed with stories about clever villains and well-meaning fools. The research literature offers something more useful: measurements of how reasoning ability and sound judgment come apart, and a long philosophical tradition that anticipated the split. Separating the demonstrated findings from the plausible engineering and the open questions is what makes the distinction usable rather than merely rhetorical.
What intelligence is, and what wisdom would have to be
Intelligence, as psychometricians measure it, is broad competence at reasoning, learning, and abstract problem-solving—the capacity to manipulate symbols, detect patterns, and acquire new skills. It predicts academic performance, job performance, and the speed of learning reasonably well. It says nothing directly about what a person does with those capacities.
Practical wisdom is a different sort of excellence. In Book VI of the Nicomachean Ethics, Aristotle distinguishes scientific knowledge, which concerns what cannot be otherwise, from practical wisdom (phronesis), which concerns what can be otherwise—the variable, particular, and uncertain domain of human action. He defines the practically wise person as one able to “deliberate well about what is good and expedient for himself… not in some particular respect… but about what sorts of thing conduce to the good life in general.” Crucially, Aristotle insists that practical wisdom concerns particulars as much as general principles, and that it cannot be reduced to a body of knowledge that could simply be handed over. His illustration is medical: knowing that light meat is digestible is useless unless one also knows which meats are light. Knowing the rule is not the same as being able to act well on it.
From the beginning, then, the two are conceptually distinct. Intelligence is general cognitive power; practical wisdom is situated excellence in deciding and doing what is good. That they are different concepts does not by itself show they come apart in practice. The empirical work does.
Biases that don’t track intelligence
Keith Stanovich and Richard West spent decades testing the common assumption that smarter people think more rationally. Their 2008 paper in the Journal of Personality and Social Psychology, “On the Relative Independence of Thinking Biases and Cognitive Ability,” reported that several classic biases are largely uncorrelated with cognitive ability. Myside bias—the tendency to evaluate evidence in ways that favor one’s prior beliefs—showed little relationship to intelligence. Framing effects, anchoring, and base-rate neglect were similarly independent of the ability measures.
A companion paper that same year in Thinking & Reasoning focused directly on myside and one-sided thinking, and found that cognitive ability failed to predict these biases. The result is easy to misread. It does not claim that intelligence is unrelated to all reasoning; it claims that a specific and consequential class of errors—the ones involving favor to one’s own position—is not corrected by being smarter. A high-IQ person is not thereby protected from motivated reasoning.
This is a genuinely counterintuitive result, and it survives replication better than most findings in the field. The folk model treats intelligence as a general-purpose error-canceling device: the smarter you are, the fewer mistakes you make, including mistakes of judgment. The data say the relationship is selective. Ability helps with some problems and is nearly orthogonal to others, and the ones it fails to help with are often the ones that matter most for conduct.
When skill sharpens motivated reasoning
The Kahan study shows the mechanism behind that independence: skill can be turned against accuracy. Numerate people are better at extracting meaning from quantitative data. When the data threaten their identity, the same skill lets them construct a more sophisticated justification—identifying an alternative reading, questioning a method, demanding a standard they would not apply to a friendlier result. The skill is real; its direction depends on motivation.
This is why the popular framing of misinformation as a deficit of intelligence is incomplete. In the Kahan design, the failure was concentrated among the more numerate. The relevant deficiency was not processing power but the disposition to seek accuracy even when accuracy is unwelcome—and that disposition is not the same thing as ability. A person can be equipped to reason well and still not be inclined to use that equipment against their own interests.
Thinking dispositions and the missing ingredient
If intelligence does not supply the required disposition, the question is what does. In a 2008 paper in the Journal of Educational Psychology, West, Toplak, and Stanovich tested whether “thinking dispositions”—measured traits such as actively open-minded thinking and need for cognition—predict rational performance beyond cognitive ability. They did: dispositions explained unique variance in performance on heuristics-and-biases tasks after controlling for intelligence.
This is the empirical cousin of a point Aristotle made about moral will. Actively open-minded thinking is the disposition to consider evidence that might overturn one’s view, to hold conclusions provisionally, and to seek reasons for positions one dislikes. It is trainable to some degree, it varies between people, and it is distinct from how quickly or flexibly someone processes information. A student can be gifted and closed-minded; a person of modest measured ability can be scrupulous about testing their own beliefs. The two dimensions move somewhat independently, and it is the disposition, not the ability, that predicts whether reasoning is used to seek the truth.
What Aristotle meant by practical wisdom
That weak correlation between ability and sound judgment maps onto the structure of phronesis. Aristotle treats practical wisdom as inseparable from moral virtue: he argues that the person with practical wisdom is not merely clever but aims at the right ends, and that virtue and practical wisdom come as a package. The clever person who can achieve any goal, indifferent to whether it is good, is in the Ethics the very case of skill without virtue. Cleverness can serve any end; practical wisdom serves the good one.
Barry Schwartz and Kenneth Sharpe make this distinction central to the modern reading. In Practical Wisdom (2010) and in earlier work, they separate moral skill—the judgment to perceive what a situation calls for—from moral will—the motivation to do it. Their argument is that institutions have increasingly optimized for rules and incentives, and that this can crowd out the discretionary judgment good practice requires. A teacher compelled to follow a scripted lesson and a doctor required to satisfy a checklist can lose the capacity to respond well to the particular student or patient. Skill and will are both necessary, and systems that reward only rule-compliance can erode both.
This is not an argument that rules are useless. It is the observation that rules cannot contain the whole of good practice—the same conclusion the Dreyfus model of expertise reaches from the psychology of skill, where the expert acts from holistic recognition that no rule list captures, and the transition from rule-following to judgment requires involved experience rather than more instruction.
Wisdom as a measured construct
Psychology has moved from debating whether wisdom exists to trying to measure it, with important caveats. Igor Grossmann’s 2017 review, “Wisdom in Context,” gathered evidence that wise reasoning—intellectual humility, recognition of change, consideration of others’ perspectives, willingness to compromise—varies substantially across situations. It is not a stable trait displayed uniformly. People reason more wisely about other people’s conflicts than about their own, and self-focused contexts inhibit wise reasoning. The capacity is real, measurable, and heavily situational.
A 2020 consensus effort, often called the common wisdom model, converged on a definition of wisdom as “morally grounded excellence in social-cognitive processing.” Its two components are metacognitive processes—such as intellectual humility and perspective-taking—and moral grounding, an orientation toward the good. Notably, the researchers reported that these features form a factor distinct from general cognitive ability. Wisdom, as measured in this literature, is not a proxy for intelligence, and no amount of additional processing power would turn one into the other.
The caveats matter as much as the findings. Much of this work relies on self-report or scenario-based assessment, which is subject to social-desirability bias, and wisdom is judged partly by moral standards that psychology cannot derive from data. Grossmann himself has warned that person-centric measurement can misrepresent a construct that is inherently social and contextual. The honest summary is that the science establishes the separateness of wisdom-like reasoning from intelligence more firmly than it establishes a precise metric for wisdom.
Why optimization does not supply goals
The practical stake for artificial intelligence is that a system can be highly intelligent—better than most humans at modeling, predicting, and achieving objectives—without containing any answer to the question of which objectives are worth pursuing. An optimizer is defined by an objective function, and that function is an input. Increase the system’s capacity to maximize it and you increase its effectiveness at whatever it was pointed at, whether deep or shallow, healing or destructive. This is the control-theoretic version of the point: capability is orthogonal to value.
Grossmann’s context-sensitive account sharpens this. If the propensity for wise reasoning depends on situation and setting, then a system that ignores context and maximizes a decontextualized metric is structurally unlike wisdom, however capable it is. The dimensions wise judgment attends to—whose interests are affected, what could change, what remains uncertain—are precisely the dimensions a naive optimizer tends to treat as noise.
None of this implies that intelligent systems cannot assist moral deliberation. They can surface consequences we overlooked, make tradeoffs explicit, and expose inconsistencies between our stated principles and our intuitions. The claim is narrower and harder to escape: helping to think is not the same as supplying the ends of thought, and no increase in processing power closes that gap.
Cleverness in the service of any end
It is tempting to treat the gap between intelligence and wisdom as an artifact of measurement, or as a problem that better education or better technology will erase. History suggests otherwise. Sophisticated knowledge has been used for both healing and destruction for as long as it has existed, and the most technically capable actors in any era have not been noticeably more virtuous than their contemporaries. The capacity to reason well is a tool; the disposition to aim it at the good is something else.
This has an uncomfortable corollary for the automation of judgment. If a decision depends on skill, a machine may perform it better than a person. If it depends on will—on caring which ends are chosen and accepting responsibility for them—then delegating it does not solve the problem; it relocates the problem to whoever set the objective. The moral weight does not disappear when the decision is automated. It moves to the specification, which is a human artifact with human authors.
The gap between knowing and being able
There is a third element beyond skill and will that the philosophical tradition names and the psychology literature confirms: virtue is acquired by practice, not by instruction. Aristotle’s account in the Ethics is explicit that moral virtue comes from habit—we become just by doing just acts—and that knowing what courage requires is not the same as being disposed to act courageously. Practical wisdom, on his account, is not a body of propositions to be memorized but an integrated capacity built through repetition.
This matters for the AI question because it identifies what no transfer of information can supply. A system can describe the courageous action, the fair distribution, or the kind response. It cannot perform the repeated acts that make these dispositions a person’s own, and it cannot take responsibility for having them. The mechanism that turns knowledge into character is not retrieval or reasoning; it is practice under consequence, which is why the Dreyfus model ties the advance from rule-following to judgment to emotional investment in outcomes.
On this view, the relevant learning is not the acquisition of knowledge about the good but the formation of the person who reliably does it. That is a claim about how character develops rather than a measured finding about AI, and it should be labelled as such. It does, however, explain why the intelligence-wisdom gap persists across so many domains: more information about what a good person does does not produce a good person, for the same reason that reading about swimming does not produce a swimmer.
This is also where the argument touches its moral and religious frameworks rather than resting on data. Virtue traditions—Aristotelian, Confucian, Christian, and others—treat the formation of character as a project with its own practices and its own standards of success. A scientific account can describe the psychology of habit; it cannot adjudicate between the conceptions of the good life that different traditions propose. Those remain frameworks chosen for reasons that are not themselves experimental results.
What would count as evidence
Several claims here are measured, several are reasonable engineering, and some are speculation.
Demonstrated: individual differences in reasoning ability are largely independent of several important biases; numeracy can amplify identity-protective reasoning; thinking dispositions predict rational performance beyond ability; wise reasoning varies with context and is distinct from general cognitive ability.
Plausible engineering: designing systems that preserve human judgment—exposing uncertainty, showing evidence, making value choices visible rather than burying them in a score—can reduce the risk that optimization crowds out deliberation. This follows from the findings but has not been tested at scale as a general intervention.
Speculation: predictions about the moral trajectory of advanced AI—whether it will produce wisdom or simply more efficient instrumental reasoning—do not follow from the data. This is the part of the discussion where evidence ends and a framework, including a moral or theological one, must be chosen. A framework can guide decisions; it is not itself a finding.
The separation matters because much public argument treats the question as if data would settle it. What data can settle is whether smarter people reason better. Greater cognitive ability helps on many reasoning tasks. It does not reliably eliminate motivated reasoning, especially when identity or loyalty makes an accurate conclusion costly.
Sources and further reading
- Aristotle. Nicomachean Ethics, Book VI (trans. W. D. Ross). Wikisource. https://en.wikisource.org/wiki/Nicomachean_Ethics_(Ross)/Book_Six
- Stanovich, K. E., & West, R. F. (2008). “On the Relative Independence of Thinking Biases and Cognitive Ability.” Journal of Personality and Social Psychology, 94(4), 672–695. https://doi.org/10.1037/0022-3514.94.4.672
- Stanovich, K. E., & West, R. F. (2008). “On the failure of cognitive ability to predict myside and one-sided thinking biases.” Thinking & Reasoning, 14(2), 129–167. https://doi.org/10.1080/13546780701679764
- Kahan, D. M., Peters, E., Dawson, E. C., & Slovic, P. (2017). “Motivated numeracy and enlightened self-government.” Behavioural Public Policy, 1(1), 54–86. https://doi.org/10.1017/bpp.2016.2
- West, R. F., Toplak, M. E., & Stanovich, K. E. (2008). “Heuristics and Biases as Measures of Critical Thinking: Associations with Cognitive Ability and Thinking Dispositions.” Journal of Educational Psychology, 100(4), 930–941. https://doi.org/10.1037/a0012842
- Grossmann, I. (2017). “Wisdom in Context.” Perspectives on Psychological Science, 12(2), 233–257. https://doi.org/10.1177/1745691616672066
- Grossmann, I., Nusbaum, H. C., et al. (2020). “The Science of Wisdom in a Polarized World: Knowns and Unknowns.” Psychological Inquiry, 31(2), 103–133. https://doi.org/10.1080/1047840X.2020.1750917
- Schwartz, B., & Sharpe, K. (2010). Practical Wisdom: The Right Way to Do the Right Thing. Riverhead Books.
Loading comments…