Steven Pinker
What this covers
Tyler Cowen interviews Steven Pinker on the mechanisms underlying language, mind, and historical progress. Pinker defends a vision of human nature grounded in evolutionary structure—not as raw determinism, but as an innate architecture that enables learning and cultural accumulation. The conversation spans linguistics, psychology, philosophy, and history, circling repeatedly between how much of our capacities are wired versus built, and whether reason and Enlightenment values actually deliver measurable improvement in human welfare. Pinker argues yes on both fronts: that inherited cognitive machinery works with learning to generate language and thought, and that measurable declines in violence and rises in well-being demonstrate reason at work despite our irrational passions.
The discussion covers wide ground in language science—how irregular verbs erode through children's incomplete learning, why combinatorial rules coexist with rote memory, and why attempts to engineer language change (like gender-neutral pronouns) fail against the grip of closed-class items learned early and embedded in mass usage. He defends a version of universal grammar as innate learning bias rather than fixed structures, and levies a sustained critique of Chomsky for moving targets and lack of empirical rigor. On consciousness, Pinker distinguishes self-knowledge from subjective experience (qualia), and argues the hard problem of consciousness may reflect a limit of human self-reflection rather than a genuine scientific puzzle. On progress, he addresses why Spontaneous Order in language mirrors historical trends—violence declines faster than genes change, taboo breakdown rather than attitude shift explains recent political shifts, and Collective Rationality can function even amid individual irrationality. The claims are precise and often contentious; Pinker positions himself as empiricist against both genetic determinism and linguistic determinism, and against skeptics of reason itself.
Pinker argues that human language, mind, and social progress are best understood as products of evolved structure interacting with learning and culture, and that the Enlightenment commitment to reason and science genuinely drives measurable human improvement despite our irrational quirks.
- Language arises from two competing mechanisms (rote memory and combinatorial rules), illustrating spontaneous, unplanned order rather than design or pure innateness
- The nature-nurture dichotomy dissolves once you specify the innate learning mechanism that does the learning
- Violence has declined and well-being has risen historically, evidence of reason and Enlightenment ideals working even against our evolved irrational passions
Language is a system combining arbitrary memorized words with compositional rules enabling human cooperation.
- Strong linguistic determinism (Sapir-Whorf), the view that we can only think thoughts for which our language has words, must be wrong because if it were true language could never have originated—it wasn't given by Martians; we developed language because we had pre-linguistic ideas for which we coined words, and the constant invention of circumlocutions and jargon shows language is not the medium of thought.
“If that were true, then you’d have to ask how did language originate in the first place. It wasn’t given to us by Martians. We developed language because we had ideas that existed prior to our being able to articulate them, for which we coined words.”
- Language is made possible by two mechanisms: brute-force rote memorization of arbitrary sound-meaning pairings (words), and a combinatorial rule system that lets you compute the meaning of phrases from the meaning of their parts and arrangement; the tension between these two systems produces much of language's quirkiness, including irregular verbs.
“We store by brute force, rote memory, arbitrary pairings between a sound and a meaning... We don’t just blurt out words but we also combine them into phrases and sentences using rules that allow you to predict... the meaning of a combination from the meaning of the parts”
- There is a defensible version of universal grammar: the child is innately biased to analyze heard speech in particular ways, pulling combinatorial rules out of the speech stream rather than recording sentences verbatim; commonalities across languages arise because language is reconstructed anew each generation by children's minds from the data their parents and peers provide.
“the child is biased to analyze the speech that he or she hears in particular ways. It does not simply record sentences verbatim... the algorithmic or computational or rule-governed half tries to pull out combinatorial rules from the speech stream”
- Grammatical language—where the meaning of a combination depends on the arrangement of the meanings of its parts and the number of such combinations is unlimited (recursion)—is unique to humans; though other species communicate, birdsong is combinatorial but has no semantics, and primate calls show only minimal ordering, all different enough from human grammar that, without Procrustean stretching, human language is unique.
“grammatical language in which the meaning of the combination depends on the arrangement of the meaning of the parts and, moreover, that the number of such combinations is unlimited — this goes back to the idea of recursion... is something that... I don’t think you see in other species.”
- Language is central to being human because it enables cooperation: it lets you make agreements for delayed and different paybacks impossible in barter, and because information is a nonrivaled good that can be shared without deprivation and multiplied, it is the ideal medium of reciprocity—conferring large benefits to others at small cost—which lubricates the hyperdeveloped cooperation and cumulative technological know-how of our species.
“information is the ultimate trade good because it is a nonrivaled good. You can share it with someone else without being deprived of it yourself. It can be multiplied, and that makes it the ideal medium of reciprocity”
Violence has declined sharply on timescales too short for natural selection to explain.
- Declines in violence occur on timescales far too short for Darwinian natural selection—Germany shifted from the world's most militaristic to its most pacifist culture in roughly a generation, and the American homicide rate halved in eight years—so we need not invoke any genetic change to explain historical reductions in violence, even if a slow genetic contribution (as Gregory Clark speculated) cannot be entirely ruled out.
“you can see declines of violence that take place on time scales that couldn’t possibly be due to Darwinian natural selection. For example, the fact that Germany went from the world’s most militaristic culture to the world’s most pacifist culture in pretty much a generation, or that the American homicide rate fell in half in eight years.”
- Historically, despite human irrationality, bad ideas get dropped over time—we no longer practice human sacrifice, throw virgins into volcanoes, or maintain hereditary monarchies—and progress in science, wealth, health, longevity, and the documented decline of violence are all fruits of the Enlightenment, achieved not linearly or inexorably but genuinely.
“some bad ideas get dropped by the wayside. Not necessarily quickly, not necessarily absolutely, but we don’t have human sacrifice anymore. We don’t throw virgins into volcanoes to get better weather. We don’t have too many hereditary monarchies anymore.”
- Pinker's optimism does not consist of prophesying the future or extrapolating curves; it consists of observing what has actually happened and noting that the pessimistic belief that we live in unusually violent times is factually incorrect, while how to project that into the future involves many unknowns he is not arrogant enough to claim to answer.
“my optimism consists of looking at what has happened and noting that, first of all, the pessimistic view is factually incorrect. Namely, people believe that we’re living in unusually violent times and we’re not.”
Humans possess rational capacity through frontal cortex override of passions, enabling self-control and progress.
- Humans are not only irrational passions; we have a large frontal cortex grafted onto the brain that can now and again override passions—we can exert self-control, count to ten, save for a rainy day—not uniformly or reliably, but enough to be worth celebrating and encouraging.
“We do have this big frontal cortex grafted onto a brain which now and again can override our passions. We can exert self-control. We can count to 10. We can save for a rainy day.”
- Skepticism about the power of reason is self-undermining: by the very act of posing the question and debating it rather than arm-wrestling, you have already committed yourself to reason, so reason cannot coherently be rejected through argument.
“by the very act of even posing this question you’re committed to reason... It’s too late. You’ve already committed yourself to reason.”
- Hume's claim that reason is a slave to the passions was primarily a conceptual point—that deduction from A to B cannot tell you what B (your goals) should be, since goals are logically distinct from beliefs—rather than a psychological claim that people are permanently irrational; people can be made more rational through inquiry, science, and exposing fallacies.
“He’s partly making a conceptual point that the ability to go from A to B using reason doesn’t tell you what the B should be... you can’t through a chain of deduction identify what you ought to aim for. That’s just a category mistake.”
Mind has specialized structure like biological organs without strict modularity, and nature-nurture is a false dichotomy.
- Even traits with a heavy genetic component have their genetic influence distributed over thousands of genes, each incrementing or decrementing the trait by a smidgen and many with mixed positive and negative effects, so the idea of inserting a single 'gene for musical ability' into a child is factually incorrect—that is not how genes work.
“the genetic influence is distributed over thousands and thousands of genes, each of which increment or decrement the trait by a smidgen, and many of which have a mixture of positive and negative effects.”
- The 'innate or learned' dichotomy is a false one resolved not by splitting the difference but by going to a finer-grained mechanism: there is an innate structure that does the learning, because learning cannot happen by magic; characterizing the computational architecture of that learning mechanism—what it does well and poorly—is the actual solution to the nature-nurture problem.
“there is an innate structure that does the learning, because learning doesn’t happen by magic. There has to be something in place that does the learning.”
- Large-scale genetic enhancement of offspring is extraordinarily unlikely because of biological tradeoffs (e.g., a too-dense neural network is actually stupider, and brain-power gains may raise cancer risk) and because humans are so risk-averse about genetic manipulation—people won't even eat genetically modified tomatoes—that they will not take such risks editing thousands of genes in their children.
“We don’t know how much boost in brain power you can get without an increase in the chances of brain cancer or of other tradeoffs. A neural network that’s too dense actually is stupider.”
- The mind is neither a homogeneous neural network nor a set of strictly walled-off snap-in modules with limited communication; rather it has specialization and structure like biological organs and tissues (e.g., blood that suffuses all tissues yet is still an organ), so there are different faculties or mindsets without them being literally independent.
“the mind can have a specialization, and structure, and different components, without them literally being independent.”
Advancing civilizations develop conflict-reducing mechanisms making war obsolete through norms and world government.
- There is no natural arc toward civilizational self-destruction; rather, as civilizations advance they tend to develop mechanisms that make conflict less likely, because war is stupid—you destroy things, kill people, and gain nothing you couldn't get otherwise, while belligerence gives others an incentive to destroy you—so a plausible trajectory is that advancing civilizations make war obsolete through world government or international norms, as a response to the Fermi paradox.
“as you become more advanced, civilizations develop mechanisms that make conflict less likely, which I think is the trajectory we have gone on so far... war is a pretty stupid thing to do.”
Nuclear weapons persist unused because human desire to cause destruction is the rate-limiting factor, not technology.
- It is not true that every form of physical accomplishment follows an exponential curve of getting cheaper; for example, plane travel did not keep getting faster—it leveled off after about 1957 and may even be slower—so there is no necessity that a $10,000 nuclear weapon will ever exist, since proliferation still requires thousands of centrifuges.
“every form of physical accomplishment follows an exponential curve of getting cheaper and cheaper. For example, plane travel hasn’t gotten faster and faster. If you extrapolate from the Wright brothers to say 1957, then it just totally leveled off.”
- The amount of violence we see is not limited by the cost of technology but by the small number of people who think it a good idea to cause massive damage for no reason beyond publicity; the rate-limiting step on terrorist destruction is human desire, evidenced by how few attacks occur despite no technological or economic impediment to a pressure-cooker bomb anywhere.
“The amount of violence that we see is not limited by costs of technology. It’s limited by the number of people who think that it would be a good idea to blow a lot of stuff up for no reason other than attracting publicity.”
- One reason no nuclear weapon has been used in war for more than 70 years is that they are not useful weapons for accomplishing battlefield goals—leaving a huge radioactive crater is not a coherent war goal—they are useful only to deter an existential threat or all-out invasion, which is why North Korea wants them.
“they haven’t been used on the battlefield because leaving a huge radioactive crater is just not a very coherent war goal.”
Language change through deliberate policy faces natural resistance in closed-class grammatical items learned early.
- Politically motivated language change is easier for open-class items (nouns and verbs) than for closed-class grammatical items like pronouns, articles, and tense markers; gender-neutral pronoun proposals since the 1970s have never caught on because grammatical items are learned early, highly frequent, and distributed across millions of conversing people, creating natural resistance to deliberate engineering.
“it’s easier to do it with what linguists call open-class vocabulary items, nouns and verbs, than with closed-class or grammatical items like pronouns.”
- The euphemism treadmill occurs because changing a label cannot change underlying attitudes: as long as an entity carries a negative connotation, a new label will simply absorb the emotional aura of the concept rather than the reverse, so labels for stigmatized groups cycle (negro to black to African American) until the underlying prejudice declines—evidenced by African American persisting 20+ years, reflecting declining prejudice.
“As long as there’s still some kind of negative connotation to an entity, then changing the label for it will just result in the new label picking up the emotional aura of the concept rather than the other way around.”
Markets and collective enterprises achieve rationality despite individual cognitive quirks and behavioral anomalies.
- Classical economics' rational-actor model is not obsolete just because individuals show irrational quirks, because a collective enterprise governed by certain rules can have a rationality that none of its individual minds possess; moreover, people experienced in real market transactions, with real money at stake, don't fall for the lab fallacies behavioral economists highlight and cannot easily be turned into a 'money pump.'
“even though every individual human brain might have its quirks and be irrational, it is possible for a collective enterprise that works by certain rules to have a kind of rationality that none of the individual minds has.”
Passive voice and other linguistic structures persist because they serve communicative purposes despite prescriptive criticism.
- The passive voice is unfairly maligned by style manuals; no construction survives in a language for over 1,500 years without serving a purpose, and the passive is the better choice when the topic already in the spotlight is the acted-upon entity, since good style begins a sentence with given information (the topic) and ends with new information (the focus).
“no construction could have survived in the language for more than 1,500 years if it didn’t serve some purpose. There are circumstances in which the passive is the better choice. In particular when the topic of a conversation... is the done-to or acted-upon.”
- There is nothing inherently better about the grammatical rules of standard English versus African American English vernacular; on dialect merit Pinker is a relativist, but it is good to settle on a single standard for formal writing to maximize communication and efficiency—standardization is valuable without one standard being objectively superior.
“there’s really nothing inherently to choose between the rules in black English vernacular and any other English vernacular. There I’m also a relativist.”
- No one planned language to be optimal; multiple competing tradeoffs (memory frequency, phonological euphony, ease of pronunciation) act as pushes and pulls, and as millions of speakers and learning children make small adjustments, the language spontaneously evolves into a balance—a Hayekian spontaneous order.
“That is Hayekian in the sense that no one planned the language to be optimal in satisfying one criterion. There are tradeoffs... As millions of speakers make little adjustments... the language itself spontaneously evolves with some balance.”
Consciousness conflates self-knowledge with subjective experience, distinct phenomena that need not co-occur.
- Consciousness conflates two distinct phenomena: self-concept or self-knowledge on one hand, and subjective experience (qualia—what it is like to feel or taste something) on the other; you could have subjective experience without any concept of yourself, or a robot that monitors and presents its own state with no one home feeling anything.
“You could have subjective experience of redness and sourness and warmth and so on without it including some concept of yourself. Conversely, you could imagine an intelligent system, a robot... where there’s no one home”
- Dreams cannot prophesy the future because that would violate much of what we know about physics; dreams are a kind of screen saver, and our understanding of physics is good enough to rule out prophetic dreams.
“I think dreams are a kind of screen saver. It would violate much of what we know about physics for a dream to be able to prophesy the future.”
- The hard problem of consciousness (why information-processing neurons feel like something) may not be a genuine scientific puzzle but a quirk of how our minds reflect on themselves; the mind examining itself is naturally puzzled because the part doing the articulating is part of the very thing it is trying to explain, so this is a case where human intuition hits a wall like the nature of time or before the Big Bang.
“there may just be some facts about the universe that are true, and we’ll never be satisfied that we intuitively understand them. Not because there is some mystery... but just that our very way of grasping reality might make certain things puzzling to us.”
Irregular verbs erode through children's generalization as frequency declines, a Darwinian process filtered by learning.
- A verb can resist erosion even when uncommon if it is similar to other well-memorized verbs, because it can parasitize the memory strength of phonologically nearby forms; this is why analogically-formed new irregulars like 'snuck' (from dig/dug, sing/sang/sung) can take hold and spread.
“if it’s similar to other things that are well memorized, it can parasitize the memory strength of something nearby in phonological space”
- Irregular verbs must be memorized because no rule generates them, and since memory thrives on frequency, when a verb declines in usage a generation of children may fail to master it and default to the regular add-'ed' rule, causing the verb to become regular permanently; this is a Darwinian erosion of the stock of irregular verbs filtered through children's minds.
“Since memory thrives on frequency... if any verb declines in frequency... then you could have a generation that never successfully masters it. They’ll default to the all-purpose add-ed rule, and then the verb will go from irregular to regular”
Intellectual tribalism and speech-limiting mechanisms threaten science's self-correcting norms and progress.
- Trumpism does not show that attitudes have become more misogynistic or racist; Google search data on racist or sexist joke terms (per Stephens-Davidowitz) shows no sudden U-turn in such searches, so the change is that people who kept ugly attitudes to themselves now feel allowed to express them because certain benign taboos have broken—taboos that legitimate or de-legitimate public expression.
“if you Google for various racist or sexist terms that are used in jokes, you get a pretty good barometer of racism... you don’t see a sudden U-turn in the popularity of racist jokes in the last, say, six months.”
- There is an intellectual equivalent of tribalism in which we treat disagreements like Red Sox versus Yankees—it is pleasurable to read arguments supporting beliefs we hold and annoying to encounter challenges—so the ideal of science and intellectual discourse is an arena whose rules make wrong beliefs eventually too embarrassing to hold, but disturbing signs (campus speech restrictions, Internet-fostered conspiracy theories) show the process can worsen.
“We tend to think of intellectual disagreements like the Red Sox versus the Yankees. It’s deeply pleasurable to read arguments that support a view that you already hold. It’s really annoying to read something that calls one of your beliefs into question.”
Linguistic homogeneity is unlikely despite language extinction because translation, media, and mother-tongue inertia persist.
- We are almost certainly heading toward more linguistic homogeneity—in the midst of a mass extinction of languages—but even dire predictions of 90% extinction leave around 600 languages, and major languages like Spanish, Hindi, Russian, and Chinese will not be given up, because translation software, national media, national pride, and the inertia of one's mother tongue prevent a single language driving out all others.
“We almost certainly are. We’re in the midst of a mass extinction of languages... even under the most dire predictions, say 90 percent of languages go extinct, that leaves 600.”
Chomsky's linguistic theory became unscientific dogma through charisma rather than testable, precise falsifiable claims.
- Chomsky's theory has been an unscientific moving target—revised every decade on a principle of perpetual revolution, never specified precisely nor tested against a dataset of language variation—because linguistics was polarized by a charismatic figure and his opponents rather than proceeding toward more precise, testable theory.
“it was neither specified in a precise way nor field-tested against a dataset of language variation, which I think is unfortunate in terms of ordinary scientific practice.”
Group selection is a faulty analogy without Darwinian mechanism for differential group replication.
- The idea of group selection is a big blunder; there is no Darwinian process of differential survival of replicators that applies to groups the way it applies to genes—it is a bad analogy—so explanations of declining individual aggressiveness should not be grounded in group selection.
“I think that the idea of group selection is a big blunder... I don’t think that there is a Darwinian process of differential survival of replicators that applies to groups in the way that it applies to genes. I think it’s a bad analogy.”
Mind is physically determined yet practically unpredictable through chaos and nonlinearity, not mystically free.
- The mind is physically determined—there is no miracle in the brain when we decide—but because of chaotic, nonlinear processes (Brownian motion, possibly quantum phenomena) the system behaves in ways that are for all practical purposes random and unpredictable, like weather taken to higher degrees, so only a Laplace's-demon-level theory could predict our choices.
“I don’t think there’s any miracle that goes on in the brain when we make a decision. In that sense, we’re determined. On the other hand there is so much unpredictability, nonlinearity that for all intents and purposes we’re not determined.”
Fashion operates via frequency-dependent selection where rarity signals status until the trend trickles down.
- Fashion is driven by frequency-dependent selection borrowed from biology: elites differentiate via a distinctive look, but once being rare confers status it paradoxically becomes common as it trickles down (e.g., to Target), forcing elites to jump to a new rare look—a perpetual churn; unlike language this may carry no semiotics, since all the unusual look communicates is 'it's different and I'm getting away with it.'
“in the competition for status, in differentiating yourself as an elite... you want a look that’s different enough that distinguishes you... When being rare is an advantage, paradoxically it starts to become more common, meaning that you then have to look to something completely different that’s rare again.”
Music taste and dialect merit are matters of relativism without objective superiority.
- The fact that Pinker doesn't have rap music on his iPod is not an argument about the objective merit of rap compared to any other music; on musical taste he is a relativist.
“The fact that I don’t have any rap music on my iPod is not an argument for the objective merit of rap music compared to any other kind of music. There I’m a relativist.”
Critical periods for language acquisition vary by component: phonology is sensitive early, vocabulary lifelong, syntax mid-teen.
- There are likely several critical or sensitive periods in language acquisition: phonology (accent, including Chinese tones) is most sensitive to early acquisition, vocabulary has no critical period since we learn words all our lives, and syntax/grammar mastery probably begins declining in the teenage years around age 15, making native mastery harder if learning starts late—though no specific brain change has been identified to explain it.
“There is evidence for at least probably several critical periods, or at least sensitive periods, in language acquisition. In particular phonology — that is, the sound pattern of the language, the accent... is the most sensitive.”
Photography works because visual perception reconstructs three-dimensional reality from two-dimensional light arrays.
- Photography is possible because visual perception does not know the external world directly but makes hypotheses from a two-dimensional array of reflected light; duplicating that array with pigment or LEDs fools the perceiver into seeing the actual thing, creating a tension between conveying something out there and making the two-dimensional patch itself aesthetically pleasing.
“visual perception doesn’t consist of knowing the external world directly, but rather making hypotheses about it via a two-dimensional array of light that’s reflected from it.”