Gilbert Harman

Major Contributions to Philosophy

In memory of Gil, we invited some of his friends and former students to summarize his major contributions to philosophy.  (If we’ve left out your favorite of Gil’s contributions, please let us know.)

Harman on Moral Relativism, by James Dreier

Harman on Character, by John Doris

Harman on Rationality and Justification, by Adam Elga

Harman on Inference to the Best Explanation, by Gideon Rosen

Harman on Intention, by Alison McIntyre

Harman on the Transparency of Experience, by Alex Byrne


Harman on Moral Relativism <top>

by James Dreier
Judy C. Lewent and Mark L. Shapiro Professor of Philosophy, Brown University

PhD in Philosophy, Princeton University, 1989

Ethics and Observation

One of Harman’s great contributions to philosophy is his argument for moral relativism in metaethics (the study of the nature of morality). The argument begins by noticing that in doing ethics, we typically employ thought experiments. Imagine a surgeon who could kill a patient so as to distribute their organs to others in the hospital whose lives the victim’s organs could save, for example: clearly we all think it is wrong to do so. Physics also employs thought experiments. But unlike physicists, ethicists never carry out their experiments in real life (for obvious reasons as well as some unobvious ones). So ethical theories are never tested against observations.

But that means we do not have the reasons to believe our moral theories that scientists have to believe theirs. Harman drew on some of his other work in epistemology to articulate the problem here. When a physicist observes a vapor trail in her cloud chamber, she may take the observation to confirm her theory, because the truth of the theory explains the observation. “Aha, there is a proton!” Her reasoning is inference to the best explanation: the best explanation of her observation includes the truth of her theory, so the observation is a reason to believe the theory. But now compare what we might think of as a moral observation. Harman’s famous example involves “a group of young hoodlums” you see torturing a cat. In some sense you may simply see that they’re doing something wrong. But, Harman argued, the truth of your moral view doesn’t explain the observation in any way. Rather, your reaction is explained by your upbringing, the presence of the animal and the boys and their activity, and so forth – but no reference to the actual wrongness seems to be necessary. Since inference to the best explanation is the main way (maybe the only way) we have of getting confirmation for our theories, it looks like we have no reason to believe in the moral facts we commonly take for granted.

But… that was not Harman’s conclusion.

Morals by Convention

The problem arises only because we started with the idea that for there to be moral facts, they would have to be like facts of physics. But they might be more like the facts of grammar: rules that are what they are because of our conventions and our psychology. That’s why we can study rules of our language by asking ourselves questions about which strings of words are grammatical sentences. Similarly, Harman argued, our moral rules are conventions, which we have internalized in the form of conditional intentions to act: I will refrain from harming you so long as you likewise refrain from harming me, for example. He thus defended a form of moral relativism, one of the views for which he was best known.

In the moral sense of ‘ought’, to say that someone ought to do something implies that the person has reasons to do it. This sense is to be distinguished from the evaluative sense, which we use when we say that something ought to be the case; for example, saying that someone ought to help the victims of a flood does not imply that any particular person has a reason to help, but only that, as we might otherwise put it, the world would be a better place if the victims were helped. A theory of the nature of morality has to explain what reasons people have to do what they morally ought to do, and that means explaining how they can be motivated to do it – at least when they act for these moral reasons.

Drawing on David Hume’s writings on the foundations of ethics, Harman’s view that moral rules are internalized conventions offered a way to explain how we can be motivated to act on our moral judgments. Of course, there was never any explicit convention in which one set of rules was agreed upon; the conventions in question are more like the conventions of language. We intend to use words with a certain meaning precisely because others adopt similar intentions; that’s how we manage to communicate. Moral rules are internalized, and they have the purpose, not of communication, but of resolving interpersonal practical disputes in a peaceful and predictable way. We intend to treat others fairly because we expect them to treat us fairly if we do, and our intentions are conditional on the continued forbearance of others.


What is right according to one moral framework may be wrong according to another, and there is no special privileged framework singled out among others – as Harman sometimes put it, there is no ‘single true morality’. Harman explicated his form of relativism by analogy to relativity in physics: velocity, length, simultaneity are fixed given an inertial frame, but there is no privileged inertial frame of reference, and these features of the world have a definite character only relative to a frame. Einsteinian relativity is counterintuitive, but nonetheless true! And moral relativism has counterintuitive consequences too. Suppose we are discussing someone who isn’t party to our conventions, and has no intention to participate in them. “He nevertheless ought to treat others with respect and honesty,” we insist. If relativism is true, then we are simply mistaken. He has no reason to treat us with respect, it may turn out. Harman’s explanation of what is happening here is famous. Think of how odd it sounds, he pointed out, to say, “Hitler ought morally not to have ordered the extermination of the Jews.” Some people do think it sounds odd. Harman’s claim is that it sounds odd because it implies that Hitler had some reason not to have ordered genocide, which quite possibly he hadn’t. (We could instead say, quite properly, that it is a terrible thing to have done – the relativism only shows up for judgments of what someone ought to do.)

I guess most people aren’t satisfied with this explanation – surely it’s just true, even if odd to say, that Hitler ought not to have ordered genocide. So that is indeed a counterintuitive consequence. But thinking of morality the way Harman did has a number of advantages, too.

First, it avoids the skepticism that the argument about ethics and observation has as its conclusion, since conventional rules do play a role in explaining our reactions to the bad behavior of others. Second, it explains how moral reasons motivate us: the conventions involve agreement in conditional intentions to act. Third, it makes good sense out of the primary methodology of ethics, which involves teasing out our own moral reactions to hypothetical scenarios. And finally, Harman claimed that it explains some otherwise puzzling features of the content of morality: for example, he argued that it is understandable why we would be willing to enter into a conventional agreement to refrain from harming others but not a convention that demanded very extensive aid to anyone who needs it, thus explaining the intuitive difference between, say, killing someone deliberately and merely declining to help those who will die without our help.

Harman’s work comprises lasting contributions to metaethics. The book and papers in which he published these arguments are still heavily cited and discussed in contemporary debates over moral epistemology, the semantics of moral language, and moral psychology. Indeed there are few topics studied today in metaethics that don’t draw in one way or another on Gil Harman’s contributions.

Harman on Character <top>

by John Doris
Professor of Philosophy, Peter L. Dyson Professor of Ethics in Organizations and Life, Dyson School of Applied Economics and Management, S.C. Johnson College of Business, Cornell University
Co-founder, with Harman and others, of the Moral Psychology Research Group

As we all know, Gil Harman was quite the character: funny, irreverent, and unexpected.  Philosophically, one of the unexpected things about Gil was that he didn’t have much use for character, despite being one himself. Ironic, that a person of such pronounced (admirable!) character was notorious (as part of a long string of his fecund philosophical notorieties)  for doubting that character matters much. How could that be?

In a word: evidence.  Or rather, lack of evidence.  In his classic 1999 paper, “Moral Philosophy Meets Social Psychology,” Gil surveyed experimental work, such as the infamous Milgram obedience experiments, documenting the formidable influence of circumstance on conduct.  This substantial social psychological tradition, he concluded, contained no evidence supporting the reality of character traits like honesty, compassion, and courage, as ordinarily understood – as robust and reliable determinants of behavior.  Instead, what the science repeatedly showed is that people could be easily – alarmingly easily – made to act in ways contrary to their character.

That’s social psychology, Gil’s critics were quick to observe; what about personality psychology?  Doesn’t that field repeatedly demonstrate the influence of distinctive individual differences on behavior?  Yes, but the problem is the magnitude of this influence: as most psychologists agree, the effect sizes associated with character variables – and indeed, most other psychological variables – tend to be pretty small, meaning that character traits – including virtues like honesty, compassion, and courage – will only be one ingredient among many in the “rainbow stew” of human psychology and conduct.

If so, Heraclitus was wrong about our character being our destiny, as are the countless people (just check the internet) who admiringly repeat his words.  Also compromised are the many philosophers, both historical and contemporary – often called “virtue ethicists” -- who build their ethical theories around the notion of robust character traits serving as bulwarks against moral failure.

If we desire a more humane destiny, Gil maintained, we should think a lot less about the content of our character, and a lot more about the character of our institutions.  Gil’s own distinctive character, he would have said, was itself a product of institutional support: the University where he taught continuously for over 50 years.  Without that support, he’d have said, you can’t be sure Harman would be Harman.

Gil’s skepticism about character made a lot of philosophers mad.  Doubly mad, actually.  Not only did Gil perturb the venerable philosophical tradition celebrating character, he also perturbed the tradition, ascendant in English-language philosophy for much of the 20th century, which instead that philosophical ethics had little to learn from scientific psychology.

While some philosophers were provoked, others were inspired.  Several generations of philosophers and psychologists, with Gil as mentor and role model, founded a thriving, and robustly interdisciplinary, field of  moral psychology, where philosophy and psychology are proudly practiced side by side.

Some of these moral psychologists celebrate character, and some don’t. The controversy  Gil fomented continues – a circumstance, I’m sure, that would have delighted him.  How can I be so sure he’d be delighted, if Gil is right about the frailty of character?   I shouldn’t be so sure, Gil would say.  And that surely would have delighted him too.

Harman on Rationality and Justification <top>

by Adam Elga
Professor of Philosophy, Princeton University

BA in Philosophy, Princeton University, 1996

According to one pervasive philosophical tradition (slightly caricatured), being reasonable or rational means starting from “foundational” beliefs (like the belief “I exist”) and using “laws of logic” to infer whatever follows from those foundational beliefs. And in drawing the dividing line between rational and irrational belief, one need not take into account “practical” limits on flesh-and-blood thinkers—such as that we have limited memories and limited time.

In a cluster of work highly influential to both philosophy and cognitive science, Harman argued against this tradition and put forward a replacement view. On the replacement view beliefs are “innocent until proven guilty”: it is by default reasonable for an agent to retain a belief unless there is some positive reason to give it up—such as its conflicting with other beliefs. So rather than asking whether some batch of beliefs is rational, we should ask whether a particular change in belief is rational.  Furthermore, the standards for rational belief change should take into account the practical limits on thinking mentioned above. For example, since real thinkers have

limited memories, it is rational to avoid cluttering up those memories with useless beliefs—even when such beliefs “logically follow” from their other beliefs. And since real thinkers have limited time, it can be rational to go eat lunch rather than continue to think about a problem—even if one’s current beliefs about the problem are logically inconsistent. On this view—and in contrast with much philosophical tradition—we should not expect principles of logic to have any very direct bearing on the principles of good reasoning.

Taken together, this cluster of Harman’s work suggests a significant perspective shift.  From the new perspective, some philosophical problems seem less pressing. For example, there is a philosophical tradition of constructing elaborate justifications for everyday beliefs about the future (such as the sun will rise tomorrow) or the external world (such as I am not a brain in a vat). But if it is changes in belief that need justifying rather than current beliefs, then it may be that no such justifications are needed.  In contrast, this perspective makes other problems and questions more pressing. If logic alone does not distinguish good from bad reasoning, we face the question: What principles of reasoning make sense for limited creatures like us? Harman gave several suggestions (for example, that good reasoning often increases the coherence among one’s beliefs—roughly, the degree to which the beliefs are compatible with, support and explain each other). But the question is far from settled, and researchers continue to attack it. The perspective shift that made this question pressing may be (in this area) Harman’s most important contribution of all.

Harman on Inference to the Best Explanation <top>

by Gideon Rosen
Stuart Professor of Philosophy, Princeton University

PhD in Philosophy, Princeton University, 1992

Gilbert Harman’s landmark paper, “The Inference to the Best Explanation” (1965) gave a name to a form of reasoning which philosophers had occasionally noticed but had rarely studied as a distinctive form of thinking in its own right.  It also made four points, all of which had significant implications for the subsequent history of epistemology. 

First some background. Philosophers had known since Hume (1739) that almost all of our reasoning in everyday life and in the sciences is non-demonstrative, in the sense that the evidence we start from does not guarantee that truth of the conclusion we arrive at.   In the first half of the 20th century, the study of this sort of reasoning focused almost exclusive on enumerative induction: reasoning that begins from patterns in the evidence (every swan we’ve examined has been white) and moves to a generalization (all swans are white) or a prediction (the next swan we encounter will be white).   The philosophical challenge was to describe the rules that govern cogent reasoning of this sort (and so to provide a so-called “logic” of induction)  and to explain why it is rational to reason in accordance with these rules (and so to provide a “justification” of induction).

Harman’s first point is that there must be more to non-demonstrative reasoning than induction.  A chemist takes in the laboratory evidence and concludes that matter is made of atoms;  a biologist  examines living things and fossils in the present and concludes that existing species derive from a common ancestor thanks to an evolutionary process spread out over eons.  In cases of this sort, our evidence does not include examples of the phenomenon that figures in our conclusion.  We don’t observe that many bodies are made of atoms, or that many species have evolved, and then infer that this holds in general, since atoms and evolution are totally unobservable (or at least they were for most of the history of science). So the inference can’t be a matter of extrapolating from patterns in the data. 

Harman’s second point is all of these inferences are naturally understood as having a different form: We collect evidence and then note that some hypothesis H provides a better explanation for the evidence than its rivals; and on this basis we conclude that H is (probably, roughly) true.   Since the best explanation for the observed facts may involve unobservable objects and processes (atoms, for example), this proposal makes good sense of the examples drawn from science.  But it also fits a host of everyday examples. Why is it reasonable for the detective to believe that the butler did it?  Because the hypothesis that the butler did it provides the simplest, most plausible explanation of the evidence (the bloody poker, the ash on the butler’s shoes, etc.)  The proposal also sheds light on longstanding problems of philosophy.  What justifies our belief in an external world filled with mind-independent objects that exist when we’re not looking?  What justifies our belief that other people have minds rather like our own?  These beliefs cannot possibly be justified by induction.  But they may be justifiable as inferences to the best explanation of our evidence.

Harman does not say what makes one explanation better than another, except to note that good explanations tend to be ‘simple’ and ‘plausible’.   For Harman this is a substantive problem at the boundary of philosophy and cognitive science, to be approached by studying the principles that scientists and other experts in good reasoning bring to bear when they identify the best, most plausible explanation of the evidence.   By the middle of the 17th century it was settled that heliocentric models of the solar system were “better” than models that placed the earth at the center, even though both could be made to fit the facts about the observed positions of the planets.  A detailed account of why this is so would permit a computer, for example, to identify the “best explanation” of the plantery phenomena by comparing the available models and applying an explicit criterion. It falls to historians of science and cognitive scientists to work out the details.   This move — to treat what would previously have been regarded as problems to be approached from the philosopher’s armchair as (at least in part) as a problem in cognitive science —  which would later become a central feature of Harman’s work, appears in embryo in this early paper.

Harman’s third point — and the most important from his point of view — is that Inference to the Best Explanation is not just a supplement to enumerative induction but a master rule which covers the extrapolatory inferences that philosophers have mistakenly classified as inductive.   Suppose I look around and notice that in a large sample, all swans are white.  The friend of induction instructs me to apply a rule of generalization: If in a large random sample, all Fs are G, conclude that all Fs are G.  But as stated, this is a terrible rule.  (In a large random sample every swan has been found to be within a mile of a human being; but no one should conclude that all swans have this property.)  And so historically the philosophical challenge was to sharpen up the inductive rule to avoid such problem (Goodman 1955).  According to Harman, this was all misguided.  If the inference from ‘All observed swans are white’ to ‘(Almost) all swans are white’ is any good, it is best understood as an explanatory inference:  The best explanation for the fact that all observed swans are white is that (almost) all swans are white.  By contrast, the best explanation for the fact that all observed swans have been within a mile of a human being is not that all swans have this property, but simply that observing a swan always involves within a mile of it.  That’s why the evidence supports the first generalization but not the second.  To put it bluntly: there is no such thing as enumerative induction, so there is no rule of induction and no problem of providing a justification for induction. There is only (in the examples we’ve considered) IBE.

Harman’s fourth point is not the point for which the paper is remembered, but nonetheless amounts to a striking anticipation of more recent developments in epistemology.  Suppose I see Jones hopping on one foot and wincing and conclude that  he’s in pain. We've already seen how the  friend of IBE would reconstruct the inference:  Jones is wincing, etc.; the best explanation for this is that Jones is in pain; So Jones is hopping because he is in pain; he’s in pain.   But the friend of induction has a different reconstruction:  Jones is wincing; in general (as we know by induction) when a person winces, she is in pain.  So Jones is in pain. These reconstructions impute different thoughts to the subject.  But of course the “reasoning” we’re talking about is automatic and unconscious.  We can’t tell be introspection (or by third-personal observation) whether I reasoned by IBE or by applying an inductive generalization to the case.  So is there any way to tell what I actually did?  

Harman’s ingenious proposal employs his solution to another philosophical problem.  Two years prior to Harman’s paper Edmund Gettier had reminded philosophy that knowledge cannot be analyzed as “justified true belief”, as many in the tradition had supposed (Gettier 1963).    Suppose I see Smith driving a new Tesla and infer that Smith has a new car.   In normal cases this would be a perfectly good way of coming to know that Smith has a new car.  But now suppose that unbeknownst to me, Smith is just test driving the Tesla—  it’s not her car —but that she has a new Toyota in the garage that I’ve never seen.  In that case my belief that Smith owns a new car is true and well-supported by my evidence; but I cannot be said to know that Smith has a new car.  And so the question arises: what distinguishes cases of justified true belief that amount to knowledge (the good cases) from those that don’t (the bad case)?   

Harman has an ingenious answer.   A justified true belief won’t count as knowledge if the subject has reasoned to that true belief by way of a false belief.   In the bad case I conclude the Smith has a new car (which is true) on the basis of my assumption that she owns the Tesla I saw her driving, which is false.   According to Harman, that is why my belief in that case does not amount to knowledge.

Now consider a different sort of case.  I see Jones hopping on one foot and wincing and conclude that he’s in pain, which he is. In a normal case this would be a good way to come to know that Jones is in pain.  But suppose that while Jones really is in pain — he has a headache — he’s hopping on one foot because he’s an actor practicing for a role.  In that case I do not know that Jones is in pain, even though I have a warranted true belief that he is.  But why not?  The inductivist  — that is, the psychologist who thinks we in fact reason by induction —reconstructs my reasoning as follows:  Jones is hopping, etc. Normally when people are hopping like that, they’re in pain.  So Jones is in pain.   But there is no false premise in this inference, so this reconstruction would not explain why my belief does not amount to knowledge.   The “explanationist” — the psychologist who says in in fact reason by IBE —reconstructs my reasoning differently:  Jones is hopping etc.  The best explanation for this is that Jones is in pain.  So Jones is hopping because he is in pain. So Jones is in pain.   But note that this reconstruction has me reasoning to my conclusion from a false premise — that Jones is hopping because he is in pain.  So this reconstruction can explain why my justified true belief does not amount to knowledge.  According to Harman, that is a reason to  favor the empirical hypothesis that everyday reasoning of this sort involves quick and unconscious applications of IBE.

This style of argument, which uses data from normative epistemology — claims about what we can and cannot know — to support a claim in the empirical psychology of reasoning, is unprecedented in the history of either subject.  Philosophers have subsequently come to appreciate the value of treating an unanalyzed concept of knowledge (rather than, say, belief) as a basic concept for inquiry in the philosophy of mind and language, and hence for psychology (Williamson 2000). This powerful idea traceable to the closing pages Harman’s 1965 paper and to the subsequent development of the point in Thought (1970).


E. Gettier (1963), Is Justified True Belief Knowledge?  Analysis 23:6, pp. 121-23

N. Goodman (1955), Fact, Fiction and Forecast. Harvard University Press.

G. Harman (1965).   The Inference to the Best Explanation. Philosophical Review 74:1, pp. 8-95.

G. Harman (1970). Thought.   Princeton University Press.

T. Williamson (2000). Knowledge and Its Limits. Oxford University Press.

Harman on Intention <top>

by Alison McIntyre
Virginia Onderdonk ’29 Professor of Philosophy, Wellesley College

PhD in Philosophy, Princeton University, 1985

In “Practical Reasoning”, Harman presented an account of practical reasoning that incorporated his own distinctive coherentist and holistic account of theoretical reasoning.  He displayed the advantages that would accrue to a view that analyzed intentions as mental states that could not be reduced to beliefs and desires, though an intention involves – and rationally requires – a corresponding belief that one would do what one intends and intentions are formed to promote one’s “intrinsic” or non-instrumental desires.  Harman proposed novel and ingenious solutions to existing philosophical problems and clearly formulated some new ones that continue to be discussed.  The field of experimental philosophy has explored his observation that we sometimes say that someone brought about a result “intentionally” even if they did not intend to do it, and his hypothesis that this linguistic oddity occurs when a person brought about a result “in the face of what ought to be a reason not to do so.”  Although the article was published in 1976, it continues to influence current research and has been cited more than 150 times since 2017 in work in linguistics, epistemology, and psychology. 

Harman on the Transparency of Experience <top>

by Alex Byrne
Professor of Philosophy, Massachusetts Institute of Technology

PhD in Philosophy, Princeton University, 1994

How can clumps of cells, like ourselves and other animals, think and perceive? One very popular answer among philosophers and cognitive scientists is that running some software is the magic ingredient. Of course some hardware is also needed, and in our case that’s the brain. Wire up billions of neurons, load the latest mindware, add a body and an environment to live in, and you have a cognizing sentient creature.

Gil was an important proponent and elaborator of this view—more exactly, of a generalization of it called functionalism. In a paper called “The intrinsic quality of experience,” published in 1990, he defends functionalism against those who object that it “must inevitably fail to account for the most important part of mental life, namely, the subjective feel of conscious experience.”

Gil says at the start, “I will say little that is original and will for the most part merely elaborate points made many years ago”; to judge by the paper’s impact this was false modesty. In any event, the paper is classic Gil: brisk, to the point, clearly written with minimal jargon, and rich in ideas.

Functionalism makes your mental life all about relations: causal relations between your brain and your behavior, causal relations between objects in your external environment and your brain, causal relations between states of your brain. The “subjective feel of conscious experience,” according to the functionalist, is explained by this complex web of causal interactions involving your brain, body, and environment. But some philosophers have thought that cannot be right, because consciousness is not a relational matter. The subjective feel of experience is an intrinsic quality, “a quality something has in itself, apart from its relations to other things,” as Gil puts it. And if that is right, functionalism is wrong.

That is one of three objections Gil examines. Part of his reply is in this passage:

When Eloise sees a tree before her, the colors she experiences are all experienced as features of the tree and its surroundings. None of them are experienced as intrinsic features of her experience. Nor does she experience any features of anything as intrinsic features of her experience. And that is true of you too. There is nothing special about Eloise’s visual experience. When you see a tree, you do not experience any features as intrinsic features of your experience. Look at a tree and try to turn your attention to intrinsic features of your visual experience. I predict you will find that the only features there to turn your attention to will be features of the presented tree, including relational features of the tree “from here.”

This is not a completely novel observation, but Gil expresses it so clearly in a look-the-emperor- has-no-clothes sort of way. It’s also unusual as a piece of academic writing—Gil addresses the reader, inviting us to look at a tree and to attend to intrinsic features of our experience. We won’t find any intrinsic features, he predicts. The subjective feel of experience is not an intrinsic quality, after all.

(Why Eloise? I don’t know, but she is a fictional character in children’s books by Kay Thomson. Appropriately, the character was based on Thomson’s childhood imaginary friend, and Gil’s paper has plenty of imaginary things, including unicorns, the Fountain of Youth, and Macbeth’s dagger.)

The significance of this passage is profound, and goes way beyond functionalism. Whether the alleged features of experience are intrinsic or not makes no difference. Try to attend to your experience or its features when you look at a tree. It’s rather like trying to attend to a perfectly transparent pane of glass in a window frame, through which you can see the tree. It is as if the frame is empty, and the glass isn’t there at all. You end up staring at the tree, swaying in the wind. Somehow the very item that you might have thought was the most rock solid, the most evident—your own conscious experience—has vanished. But the experience has to be there!

This seeming elusiveness or transparency of experience has been deployed in many arguments about the nature of mind and mental representation, and Gil’s paper is invariably cited and the passage above is often quoted. When I recently opened up David Papineau’s The Metaphysics of Sensory Experience, which came out this year, the title alone made me confident that Papineau was going to mention Gil on seeing trees. Which he did: “Here is a well-known quotation from Gilbert Harman.” This being philosophy, Papineau does not agree with Gil. But it is indicative of Gil’s monumental contribution to philosophy that, three decades later, Eloise is still going strong.