Evolutionary psychology has made impressive progress in understanding the origins of morality. Along with the many posts about these origins on Less Wrong I recommend Robert Wright's The Moral Animal for an excellent introduction to the subject.

Guilt does not naturally fall out of these explanations. One can imagine a mind design that although often behaving morally for the same reasons we do, sometimes decides a selfish approach is best and pursues that approach without compunction. In fact, this design would have advantages; it would remove a potentially crippling psychological burden, prevent loss of status from admission of wrongdoing, and allow more rational calculation of when moral actions are or are not advantageous. So why guilt?

In one of the few existing writings I could find on the subject, Tooby and Cosmides theorize that "guilt functions as an emotion mode specialized for recalibration of regulatory variables that control trade-offs in welfare between self and other."

If I understand their meaning, they are saying that when an action results in a bad outcome, guilt is a byproduct of updating your mental processes so that it doesn't happen again. In their example, if you don't share food with your sister, and your sister starves and becomes sick, your brain gives you a strong burst of negative emotion around the event so that you reconsider your decision not to share. It is generally a bad idea to disagree with Tooby and Cosmides, but this explanation doesn't satisfy me for several reasons.

First, guilt is just as associated with good outcomes as bad outcomes. If I kill my brother so I can inherit the throne, then even if everything goes according to plan and I become king, I may still feel guilt. But why should I recalibrate here? My original assumptions - that fratricide would be easy and useful - were entirely correct. But I am still likely to feel bad about it. In fact, some criminals report feeling "relieved" when caught, as if a negative outcome decreased their feelings of guilt instead of exacerbating them.

Second, guilt is not only an emotion, but an entire complex of behaviors. Our modern word self-flagellation comes from the old practice of literally whipping one's self out of feelings of guilt or unworthiness. We may not literally self-flagellate anymore, but when I feel guilty I am less likely to do activities I enjoy and more likely to deliberately make myself miserable.

Third, although guilt can be very private it has an undeniable social aspect. People have messaged me at 3 AM in the morning just to tell me how guilty they feel about something they did to someone I've never met; this sort of outpouring of emotion can even be therapeutic. The aforementioned self-flagellators would parade around town in their sackcloth and ashes, just in case anyone didn't know how guilty they felt. And we expect guilt in certain situations: a criminal who feels guilty about what ey has done may get a shorter sentence.

Fourth, guilt sometimes occurs even when a person has done nothing wrong. People who through no fault of their own are associated with disasters can nevertheless report "survivor's guilt" and feel like events were partly their fault. If this is a tool for recalibrating choices, it is a very bad one. This is not a knockdown argument - a lot of mental adaptations are very bad at what they do - but it should at least raise suspicion that there is another part to the puzzle besides recalibration.


Suppose you need a lawyer for some important and very lucrative legal case. And suppose by a freak legislative oversight, your state has no laws against legal malpractice and unethical lawyers can get off scot-free. You are going to want to invest a lot of effort into evaluating the morals of the many lawyers anxious to take your case.

One lawyer you meet, Mr. Dewey, has an unusual appearance. A small angel, about the size of a rat, sits on his right shoulder holding an electric cattle prod. This is remarkable, and so you remark upon it.

Mr. Dewey scowls. "That angel has been sitting there for as long as I can remember," he tells you. "Every time I do something wrong, she pokes me with her prod. If it's a minor sin like profanity, maybe she'll only poke me once or twice, but if I lie or swindle, she'll turn the power up on max and keep shocking me for days. It's a miserable, miserable existence, and I'm constantly scared to death I'll slip up and make her angry, but I can't figure out how to get rid of her."

You express some skepticism about this story, so Mr. Dewey offers to demonstrate. He says a mild curse word, and sure enough, the angel pokes him with the cattle prod, giving him a mild electric shock.

Suddenly, Mr. Dewey is a very attractive candidate for your lucrative case. You can be assured that he won't swindle you, because whatever gains he might take from the swindle are less attractive than the punishment he would get from the angel afterwards.

Surgeon Paul Brand considered pain so useful to the body's functioning that he called it "the gift nobody wants". Mr. Dewey's angel is also such a gift, even though he might not appreciate it: clients worried about ethical issues will bring their patronage to his law firm, giving him a major advantage over the competition.

Whereas normally we must trust a lawyer's altruism if we expect em not to con us, in Mr. Dewey's case we need only trust him to pursue his own self-interest. This, then, is the role of guilt: it provides assurance to others that we will be punished for our misdeeds even if there is no external authority to punish us, avoiding Parfitian hitchhiker  dilemmas and ensuring fair play. The assurance of punishment ensures fair play and makes mutually beneficial transactions possible.


The big difference between Mr. Dewey and ourselves is that where Mr. Dewey has unquestionable evidence of his commitment to self punishment in the form of a very visible angel on his shoulder, for the rest of us guilt is a private mental affair and can be faked. It would seem to be a winning strategy, then, to claim a tendency to guilt while not really having one.

Ms. Wolfram is Mr. Dewey's main competitor, and is outraged at her rival's business success. In an attempt to even the scales, she buys a plastic angel figure from the local church and glues it to her shoulder. "Look!" she tells clients. "I, too, suffer pain when I commit misdeeds!" Her business shoots up to the same high levels as Mr. Dewey's.

One day, the news comes that Mr. Dewey was spotted whipping himself in the town square. When asked why, he explained that in a moment of weakness, he had overcharged a customer. His angel, who had lost its cattle prod, was mind-controlling him into the self-flagellation in place of its more usual punishment.

This provides an impressive bar for Ms. Wolfram to live up to. Sure, she could just whip herself like Mr. Dewey is doing. But it wouldn't be worth it - she just doesn't like the money enough that she would whip herself after every swindle just to drum up business. If she's going to have to whip herself to fake remorse whenever she commits wrongdoing,  her best policy really is to genuinely stop swindling people.

Mr. Dewey has found an unfakeable signal. Even though whipping himself in public is one of the most unpleasant things he could do, in this case it is good business practice. It once again differentiates him from Ms. Wolfram and restores his status as the city's most desirable attorney.

In evolutionary terms, guilt becomes more credible the more it requires publicly visible behavior that no reasonable cheat would want to fake. Hurting oneself, avoiding pleasurable activities, lowering your own status, and withdrawing from social activities are all evolutionary costly and therefore good ways to prove you are experiencing guilt; the usual vocal, postural, and facial cues of being miserable are also useful.

There's no reason people should evolve an all-consuming sense of guilt. If an opportunity comes along where the benefits of cheating are greater than the social costs, an organism should still take it. Therefore, guilt has to be unpleasant but not infinitely unpleasant. A person who committed suicide in response to even the slightest moral infraction would be trustworthy, but they'd miss out if an excellent opportunity to win major gains for cheating happened to fall into their lap.

The conspicuous experience of guilt is an evolutionarily advantageous way of assuring potential trading partners that you will be punished for defection. The behaviors associated with guilt are costly signals that help differentiate false claims of guilt from the real thing and add to public verifiability of the punishment involved.


If you kill your brother in order to inherit the throne, you probably deserve whatever guilt you feel. But in the phenomenon of "survivor's guilt", people feel guilt for events that weren't even remotely their fault. Maybe you go hiking with your brother, and through no fault of your own he trips and falls down a crevasse and dies, and now you feel guilty. Why?

Hunter-gatherer societies were more violent than our own; statistics differ but by some estimates around 30% of hunter-gatherer males died of homicide. Even as late as the Bronze Age, Biblical figures who killed their brothers comprise a rather impressive list including Cain, Solomon, Ammon, Abimelech, and Jehoram; Jacob's sons merely attempted to do so. So the priors for suspicious death must have been very different in the olden days.

Further, in such a crime-ridden culture, there may have been more incentives to blame an enemy for a death, even if that enemy was not responsible. A person whose brother has accidentally died on a hiking trip with no witnesses would be very targetable.

And even in less drastic situations than blaming survivors for a death, there may be other possible threats to reputation. If there is only one survivor of a battle, he may be suspected of cowardice; if there is only one survivor of a disaster, she may be suspected of running away without helping others.

Therefore, it would be advantageous to have a method of proving your innocence. Suppose that you would gain benefits X from killing your brother and covering it up, but that you would suffer losses Y if you were suspected of the crime and punished. A precommitment to a policy of experiencing a level of guilt between X and Y provides a tool for proving your innocence. It would no longer be in your self-interest to kill your brother, because you will suffer so much guilt that you won't be able to enjoy the benefits of your crime; your would-be accusers realize this and admit your innocence, saving you from the still worse outcome Y.

In this case, guilt would be an entirely adaptive response to a disaster with which you were associated, even if your own actions were beyond reproach. A level of unhappiness worse than any benefits you could get by profiting the tragedy, but less than any punishment you might receive if you were suspected of profiting from the tragedy, would be helpful in clearing your name of any wrongdoing.

(The proposed mechanism is almost identical to one cited in Thornhill and Palmer's controversial and unpleasant evolutionary account of post-traumatic stress after rape.)

This theory makes some testable predictions, which as far as I know have not been tested:

- People should feel guiltier about events for which reasonable suspicion might exist that they played a part; for example, if your brother slipped and fell while you were hiking alone with him rather than in a large group with many witnesses.
- People should feel guiltier about events for which they might profit; for example, if you stood to inherit money from your brother, or never liked him much anyway.
- People may be suspicious of people who come out of a disaster feeling no survivor's guilt.


Guilt, like pain, is "a gift nobody wants". Because people with guilt are known to punish themselves for moral wrongdoing, their social group considers them more trustworthy and they gain the advantages of trade and cooperation. In order to prove that their guilt is real rather than feigned, they use costly signals like deliberate self-harm and self-denial to display their punishment publicly

When one has done nothing wrong, it can sometimes be advantageous to paradoxically display guilt in order to prove one's lack of wrongdoing. These costly signals demonstrate that it is not in one's self-interest to lie about these matters, while still being less costly than the punishment for defection.

Although this could theoretically be mediated by the behavioral strategies of a sufficiently intelligent and Machiavellian unconscious mind, it fits within the framework of evolutionary psychology and can also be interpreted in evolutionary terms.

New Comment
103 comments, sorted by Click to highlight new comments since: Today at 12:40 PM
Some comments are truncated due to high volume. (⌘F to expand all)Change truncation settings

I believe that my dog felt guilt, because when I would come home and find he'd gotten into the garbage and strewn it around the kitchen, he would use body language very similar to that of a human who was feeling guilt. (The dog was not picking up on my emotions, because I usually hadn't seen the mess yet, and didn't know what he'd done until his body language told me.)

This body language has much in common with submissive body language, which is partly shared between canines and primates.

So ancestral "guilt" could have been "the emotion you feel after disobeying the alpha", which helps you display submission and also not do it again so you don't get beaten again. It could originally have just been a type of fear. (Perhaps it still is!)

This is compatible with Yvain's explanation - it would have the effect Yvain described; consistent guilty behavior would provide evidence that you would be likely to obey the alpha even in his/her absence. But the origin could be simpler than that.

Testable predictions:

  • you should feel measurably less guilty doing things that harm subordinates, than things that harm superiors

  • if you are the absolute alpha, guilt should disappear

  • if you are a local alpha, guilt should disappear in that local context, while reappearing in a context where your superior disapproves

Some of the existing psychological results seem consistent with such claims: http://lesswrong.com/lw/dtg/notes_on_the_psychology_of_power/
That seems remarkably similar to most people's actual behavior within dominance hierarchies, as I have observed them.
The dog behaviorist Alexandra Horowitz says that dogs don't feel guilt or act guilty; they just act submissive to appease the alpha. She's run experiments that seemed to indicate this. Your anecdote is exactly the kind of thing she says shouldn't happen, although maybe he was anticipating your anger and trying to compensate in advance.
0Alex Flint13y
I observed the same thing in a dog we had several years ago. Sometimes we never found out what she'd done wrong, but we knew it was something she believed was pretty bad because she'd slink around with her stomach almost dragging along the ground for hours.

You drew an analogy to pain as an unwanted gift: I think an even better analogy is with rage. On Steven Pinker's account, a hot-temper is a way to signal that you're unconditionally pre-committed to wreak havok if anyone harms you, even if, after having been harmed, it is no longer in your interest to do so. Temper is a signal that you have mind-control angel on your shoulder that sends you uncontrollably crazy when you are wronged, and is hence a deterrent to anyone that might harm you.

That's an interesting idea. Where can I read more about it? Any specific Pinker book?
0Alex Flint11y
This was from How The Mind Works: http://www.amazon.com/dp/1469228424

I'm currently reading I Thought It Was Me (but it isn't), a book about women and shame-- the author distinguishes between shame (the feeling of deserving to be of extremely low status because of personal defect) and guilt (the feeling that one has not lived up to one's standards and can change).

I can't remember if there's a theory about why shame exists, but the author says it never leads to useful change even though it's pervasive.

She mentions having done some research about men and shame (doesn't seem to have written any books about it) and says that th... (read more)

I remember someone making a distinction between shame-based cultures (they claimed that Mideastern, Indian, and Asian cultures were shame-based) and guilt-based cultures. The difference, they claimed, was that shame is social: You feel shame only if you're caught. Guilt, however, is something you feel even if you're not caught. I think this distinction between shame and guilt is useful, even if there aren't really "shame-based cultures". Also, you should feel shame for things you were accused of doing even if you didn't do them, or which you did under extenuating circumstances that others are unaware of. "Shame-based culture" sounds like it would appear the same as what some people call an "honor culture", eg., 18th-century Scotland, 19th-century West Virginia, most 21st-century Islamic nations, where defending your honor has extremely high importance.

This is slightly off-topic (as it doesn't help distinguish between Yvian's hypothesis and T&C's) but anyway:

People who feel guilty sometimes give to charity, right?

Is the social purpose of giving (in this case) therefore to punish yourself financially rather than actually help anyone?

I believe it's to make atonement, which is more closely related to punishment than to helping others. There is a component of both present in atonement, however. Note that you will not see many people punishming themselves in ways unrelated to their crime (hair shirts, self-flagellation, etc) outside of religious communities that believe the true problem of wrongdoing is distancing oneself from a deity rather than harming others. But you will also not see many people atoning for wrongdoing by finding the cheapest possible way to create sufficient good to counterbalance the harm they've committed. Rather, most people try to find hard/painful ways to do good. Often this involves nonfinancial contributions.
1Scott Alexander13y
Maybe it's to signal that your misdeed was a one-time lapse rather than coming from any deep inner tendency. For example, if you run over a cat, you might donate money to an animal shelter, which shows that despite what happened you really like animals and would never run one over intentionally. The financial punishment angle can't hurt, though.
Interesting. My introspection says no... for what that's worth. Usually I feel more good than bad when I give to charity.
Good! That's how it should be. I was thinking about only the people who give as a response to guilt. Do you give mainly when you feel guilty, or at other times? Also, have we addressed whether in general guilty people who self-punish feel good doing so (even as it puts them at a material/social disadvantage)?

If we were trying to engineer a corporate entity to behave morally, we might precommit the entity (using a contract) to generous settlement terms with any whistleblowers providing even relatively minimal evidence of the appearance of unethical or antisocial behavior.

I'm not completely sure, but possibly you would also (or instead?) have to require similar precommitments of the humans (CEO? Directors? Shareholders?) who have power over the entity, so that their incentives align with the corporation's.

I like this idea a lot. I think some elaboration might be necessary to remove possible system-gaming scenarios, though. For example, imagine a collusion between an accountant who can manipulate the books and a "whisteblower" employee. If the accountant could alter the books without it being traced back to them, then they could arrange to split the settlement and win big, at the expense of the rest of the corporation.
Nick Szabo has written some about "group controls", which are (I believe) the state of the art for achieving Friendly organizations: http://szabo.best.vwh.net/groupcontrols.html

I find evolutionary-psychological reasoning always a bit suspect because it seems a bit too good to explain just about anything. Having said that -- sometimes dogs seem to have a sense of guilt -- does that imply they have a kind of morality too(*)? Or is it just some kind of 'act' due to co-evolution with humans?

(*) De Waal in Primates and Philosophers argues that some animals have a certain degree of morality (Robert Wright slightly disagrees with De Waal, in the same book).

9Scott Alexander13y
I sort of agree with this; it's temptingly easy to explain away complex higher-level behaviors as adaptations. But unless I am missing something, the alternative to an evolutionary theory of guilt isn't a non-evolutionary theory of guilt. It's to say "Guilt? Well, obviously if you do something immoral, then you feel bad afterwards, because on some level you know it was wrong." As far as I know once you've reached the point where you really feel like emotions require explanations and you understand that the explanations cannot themselves be mental, evolutionary psychology is pretty much the only game in town. I'm not saying the particular formulation of guilt presented here is right - maybe it's Tooby and Cosmides' model, maybe it's something else no one's ever thought of, but I think any accurate model of guilt would sound just as reductionist as this one. Guilt in dogs seems to be mostly illusory.
I don't doubt that owners have an inaccurate view of their pets' mental life, but I wouldn't draw the line between human guilt and dog seeming-guilt too sharply. Specifically, the article says: (correction - here's a better quote: So, dogs show a guilty face, not in response to their own past actions, but in response to human reactions. But this is not entirely unlike how guilt works in humans. As you yourself observed: Humans are perceptibly better than dogs at remembering what they did and planning what they're going to do, and this aspect of guilt - awareness of what the dog had done - seems to be missing. But other aspects of guilt seem to be present.
That isn't what happens with dogs, at all! Whoever claimed this has never had a dog. I've had a dog, and usually, the first clue you get that your dog has done something wrong is that your dog comes up to you, face downcast, ears and tail drooping, then gives you that big-eyed "I'm so sorry" pleading look. Before you have any clue that the dog has misbehaved. There was no "misinterpreting" my dog's guilty look, because it was binary. We often punished him by telling him to go stand in the closet for a short time. So if we came home, and he had done something he knew we'd be upset about, he'd greet us at the door, and then go open the closet door and walk into it. Hard to misinterpret.
On second look, the experiment does not look very compelling, specifically because the misdeed that being used may be too weak to trigger guilt even if the dog can feel guilt. The misdeed is this: This trivial misdeed ranks pretty low on the list of things that I would find troubling. Higher up would be making a mess, higher still would be destroying something. I can understand why the experimenters might want to choose a trivial misdeed - they want to keep costs down. As it happens I've read an alternative explanation of guilty behavior in dogs, which is that the dog is reacting not to a memory of having committed a misdeed, but to the presence of some situation (i.e. the aftermath of the misdeed) that the dog knows makes you upset, and that the dog would be acting equally guilty regardless of whether this aftermath was the product of the dog's own behavior or not. Now I'll be the first to say that this sounds almost like whoever came up with that theory is trying very hard to come up with any excuse to deny dogs a little bit of memory and self-awareness. But still, I thought I'd mention the theory.
As Yvain pointed out in the main post, the same thing also applies to humans.
Yes, agreed -- for common behaviours, it's hard to think of anything but evo-psy. Hypothetically, there could some global environmental factor that influences behaviour -- think gay bomb -- but that's probably not very realistic. The trouble with evo-psy is indeed how it's hard to distinguish between alternative hypotheses. Regarding the BBC-article - I think it is more about the dog owners than the animals themselves. But of course, if the only 'evidence' for dog-guilt is the bias of their owners, the case gets very weak.
Dogs are only recently extracted from wolves, which are social animals whose behavior affects their status in the group. They would have experienced similar pressures even before their co-evolution with humans. Observing guilt in non-social animals would probably falsify the hypothesis though.
My unofficial conclusion after a few years of paying attention to my dog in this space is that mostly what she's doing is giving me the signals that work. The "I'm sorry, please forgive me, see: I'm lovable!" face/dance is really quite effective at defusing my irritation, especially given that she can often pick up on me being irritated before I quite notice it myself.
Also see Wild Justice, a book that surveys animal moral systems. I interviewed one of the authors here.
I'm not sure there's that strong a difference between these scenarios. What's the difference between an act that is usually followed and a morality that is occasionally breached? Couple this with self-deception and the description of a mind as multiple interacting agents...
I usually interpret the intended difference to have to do with the relevance of belief in observers. That is, when my X is described as "merely an act," I understand the speaker to be suggesting that if I believed myself unobserved, I would not demonstrate X. (There's also a related implication having to do with sincerity, but that's much trickier to express in a succinct coherent way.) Of course, as you point out, in reality it's more complicated than that, and something can be in some meaningful sense "an act" while also being something I do for my own benefit.

So if you loved your brother dearly and everybody else knew this, you would feel less guilt if he died while you were on a hiking trip together with a group of other people?

Thinking back to how the cliff ledge where he'd been standing suddenly began to collapse, and everyone else had simply stood there, frozen, and you instantly lunged towards him, and actually managed to just just brush his finger tips... but by that time you had fallen over the edge yourself.

Then WHUMPH! an out-jutting tree broke your fall, knocking the wind out of you like the fist of... (read more)

The one thing that caught my eye in your scenario is the stipulation that, in the case of your actual brother, there are no other people present. Is this because you would be made somehow to feel guilt if there were witnesses or because the presence of others who also failed to save your brother somehow mitigates the guilt? It is an interesting situation given the dynamic between your brother and yourself but the witness factor is what intrigues me.
No, I would feel the same if there were witnesses to my failure to risk myself to save my actual bad-blood-brother in such a hypothetical situation. I just stipulated the 'no witnesses' to contrast with the first situation fully. I'm just saying: "okay, here's a hypothetical situation where none of the reasons proposed in this article for the existence of guilt apply, and yet I'd expect to feel great guilt. By contrast, here's a situation where they all apply but I wouldn't expect to feel guilt." The question is, do you other people feel my expectations are wrong? Would you expect differently for yourselves? How do you think most people would [expect to feel/actually feel if such a thing happened]? Do you think I'm mistaken about how my hypothetical situations relate to the arguments in this article? Etc Oh, and actually, I already do feel that I did miss an opportunity to save him. If I had been more mature myself while we were growing up, I could have been an awesome big brother. Preventing him from ever getting near being as lost as he is now probably would have been possible through behavioral means, I mean But of course I couldn't teach him not to be an idiot, because I hadn't learned yet myself, because nobody had thought to teach me in any deliberately optimized way, and I had to learn it from a long streak of experiences that just happened through good luck to be sequenced in an order that stopped me from getting hopelessly lost myself. Which took a long time. But I don't feel guilty for not being lucky enough to have an environment that would have taught me the level of maturity I would have needed to save him, soon enough that I could have done so. I feel sad, remembering the little boy that was me carrying the little boy that was him through the forest on my shoulders, and knowing that only one of those little boys was lucky enough to become a man who is at least trying to be a nice person and a positive force in the world. And I feel a great need
See also the Markdown syntax guide. Posts (as opposed to comments) use a wysiwyg editor or HTML, not Markdown.
Thanks, although I had just found the answer myself on the FAQ page when I saw the little red 'you got a reply' envelope come on. Would have found it sooner, but of course this is 'less wrong of the many interesting links', and if you aint focused, you will be distracted. How I missed seeing the "help" button in the first place though, I don't have any excuse for. :P
You are not the first!
Part of it could be that she was an adult when things were going wrong with your brother.
Well, yes, you're right of course. There are lots of different factors at work there. What I'm wondering is, does rationality seem likely to be one of them? That's a horribly uncontrolled situation for actually answering the question, but that's the context in which I thought of it. Actually, if aspiring rationalists should be better at not feeling 'undeserved guilt', and guilt serves the 'purpose' Yvain is saying it might, that could be another defecting by accident situation, couldn't it?
No, I would feel the same if there were witnesses to my failure to risk myself to save my actual bad-blood-brother in such a hypothetical situation. I just stipulated the 'no witnesses' to contrast with the first situation fully. I'm just saying: "okay, here's a hypothetical situation where none of the reasons proposed in this article for the existence of guilt apply, and yet I'd expect to feel great guilt. By contrast, here's a situation where they all apply but I wouldn't expect to feel guilt." The question is, do you other people feel my expectations are wrong? Would you expect differently for yourselves? How do you think most people would [expect to feel/actually feel if such a thing happened]? Do you think I'm mistaken about how my hypothetical situations relate to the arguments in this article? Etc
Rationality can often allow us to overcome otherwise debilitating emotional responses. I think a non-rationalist in the same situation who let their psychopathic brother die ... they would probably feel a lot of guilt. A LOT of guilt. Finally, evolution isn't always fine-tuned, especially in social contexts. Guilt simply may not be a fine-tuned enough emotion to make you feel pain directly proportional to the odds of you being suspected for a crime; it's more likely that you feel guilt differently in different classes of situations rather than in different levels of the same scenario.

My question is:

Why is guilt often so bad a de-motivator? There are people who know they will feel guilty before the fact, do in fact feel guilt after the fact, and perhaps even continue to live with that guilt every day, but still continue the behavior. Guilt seems like it evolved exactly for the purpose of preventing people from acting in a way they consider immoral or unethical, so why does it so often seem so bad at its job?

I think the point is that it didn't evolve to stop us from doing things we consider immoral - it evolved to stop us from getting punished by other people for doing things that are generally considered immoral. This also explains why people feel relieved (and, as is implied, less guilty) when the thing they're feeling guilty about is found out - it seems quite likely to me that in the ancestral environment, if you were going to be punished for something, that punishment would happen pretty promptly after your indiscretion was discovered, and guilt would be pretty pointless after that point in time.
"I think the point is that it didn't evolve to stop us from doing things we consider immoral - it evolved to stop us from getting punished by other people for doing things that are generally considered immoral." This seems a question begging explanation. It simply pushes the issues back, "why did it evolve just to signal a self-punitive tendency and not to signal a more-adaptive restitutionary tendency?" (Perhaps it only looks more adaptative, but that must be shown.)

Is guilt really a good signal?

In a precommitment case, the person takes on burdens which are imposed by purely external forces. The surety of punishment comes from the duty other people have to honor the agreement and take a pound of flesh. In showing you this agreement, bearing an appropriate seal of recognition, I demonstrate that I will suffer negative consequences for cheating.

Guilt, however, doesn't operate in that way because I can't show you my guilt function and so demonstrate my commitment. If we expand it to say that my reputation for guilt is an... (read more)

A positive benefit of having guilt (for the person suffering it) that seems to have been overlooked here is that we punish people less when they display signs of guilt. This is so prevalent that we even see it in court rooms. In criminal procedings prosecutors try to talk up any sigh of lack of guilt or repentance to get longer sentences and defendents claim they are alreay suffering a lot of guilt to get sentences lower. This also applies to fake guilt.

  • People may be suspicious of people who come out of a disaster feeling no survivor's guilt.

How does this theory interact with the idea (untested folk theory?) that helping others when there's been a disaster results in not feeling guilty about that disaster?

Bravo, this is a nice analysis.

I think you should carry through the experiments you suggest. I'm assuming you don't want to actually kill your brother, so perhaps you could gather some test subjects, ask them to imagine being in the situations listed, and ask them to report the magnitude of their hypothetical guilt. Imperfect, but I think this kind of experiment is fairly widely accepted in the relevant journals. Then publish in a top evo-psych journal.

It doesn't look like undeserved guilt gets you anything that plain vanilla guilt doesn't. Vanilla guilt suffices to make it so that it's not in your self-interest to kill your brother, arrange for a disaster etc.

What do you mean by "vanilla" guilt?
Feeling bad for actually having committed a crime.

I think the implication is that undeserved guilt serves to signal to others - once the event has actually happened - that those things were not in your self interest. Everyone can see they weren't, because you are publicly paying a higher cost than the benefit you would receive for having caused them.

Vanilla guilt perhaps helps, but it's a much less powerful signal to the community. First, you must have committed a crime in the past, so they can know that you are a guilt-feeling person. Second, they have to trust that your personality hasn't changed since then. Third, they have to know about/remember the incident in question, and fourth they have to make the mental leap of using that memory as evidence in this situation.

Those reasons do make sense. I think the post should mention them.
It's the kind with the lovely bits of vanilla bean in, comes in a tin with the white flowers on the label, you know the stuff.
And then of course there's chocolate guilt. I used to get that all the time at Chanukah.
Ah, if only it were true

It seems to me that guilt is inherently paradoxical. In light of the misdeed, if you believe you're bad, your good. If you believe you're good, you're bad. Seems a lot like the liar's paradox.

I don't think that's paradoxical. Consider table of the possible combinations: (Did a bad deed) (Feel guilty) (Morally correct?) * Yes, Yes, Yes * No, No, Yes * Yes, No, No * No, Yes, No* In other words, guilt is morally correct IFF it's a true indication of whether you actually did anything bad. That's in line with the OP's interpretation of guilt as a signalling mechanism; not being guilty when you ought to be is a kind of lying. *I'm less certain about being guilty when you haven't done anything wrong; I think this isn't considered particularly immoral, but at least in the cultures I'm used to it's considered pointless and somewhat egotistical, a kind of self-pity.
Guilt is not merely the acknowledgement of a mistake, is it? Isn't it self-punishment? It's self-punishment that seems paradoxical to me. Punishment only makes sense if the punisher and punished are different people. Another way to look at is that It takes a good judge to accurately judge someone as bad. So which are you when you're feeling guilty, judge or judged?
The OP disagrees with you; it points out that self-punishment can be a good idea if you're doing it in front of other people, because it signals that you have genuine regret over your actions, and makes it seem less likely that you'll do something bad again because you're precommitted to punish yourself again if you did.
I agree it may make sense to feign it. But to do it sincerely, to actually inflict the emotional pain of guilt on yourself, and not just feign it, seems irrationally paradoxical. When I isolate for consideration just the interior subjective phenomenon of feeling guilty, I can't see how to escape the paradox that believing you're bad makes you good and vice versa.
If everyone feigned it, then nobody would believe anybody else's feigning. It's a typical game theory pattern; cheating the system is possible, but if everyone or nearly everyone cheated then there wouldn't be a system. Plus, there are forms of self-punishment that are hard to fake, and these accordingly seem to get more respect. So are you imagining a kind of rapid back-and-forth state change here? I don't think that necessarily has to be the case, because what we think of as somebody's moral state has to be a sum, taking into account all the things they've done and are doing and adding them together. Being guilty over having done something bad is itself a moral good, no doubt, but it doesn't replace the bad moral state that was arrived at by doing the bad thing, it's just added to it.
Thanks very much for your help on this DSimon. I really appreciate it. You say "Being guilty over having done something bad is itself a moral good, no doubt, but it doesn't replace the bad moral state that was arrived at by doing the bad thing, it's just added to it." Are you saying that one can be in two moral states in the same moment? How can that be?
Sort of. I'm saying that one's moral state, as we typically think about it, can be modeled (in a simplified way) as a sum of the god and bad things you've done. So let's say you do a bad thing, pushing you from morally neutral at 0 down to -6. But, you feel correctly guilty about it, which is +2 good. Now your moral state is at -4, which is better but still not up in the positive numbers that we'd call "overall morally good". This is also why there isn't any flip-flopping back and forth: guilt in this model is about a particular historical action, not your current overall moral state. I can also approach this from the other side and show that a simple good/bad binary doesn't work as a model of our sense of moral evaluation. If you imagine the various combinations in the table I made earlier, it feels right (at least to me) to be able to sort them from most to least moral like so: * Didn't do anything bad, doesn't feel guilty * Didn't do anything bad, feels guilty * Did something bad, feels guilty * Did something bad, doesn't feel guilty If that (or any other single ordering) also feels about right to you, then it follows that you'd need more than just one bit of input (i.e. whether or not a person feels guilty) to morally evaluate a person; with only one bit of information, you can only sort people into two groups, you couldn't come up with an ordering for four people like above. Therefore your earlier statement that "believing you're bad makes you good and vice versa" can't be correct, because it only takes as input that one bit of information.
0Swimmer963 (Miranda Dixon-Luinenburg) 13y
Isn't this true all the time? Say you took money from your sister's wallet and lied to her about it...that's morally wrong...but you donated the money to charity, which is morally right.
Maybe that's the point. Maybe the way we evolved to demonstrate group loyalty and therefore, trustworthiness, was to forfeit our rationality as a rite of initiation. We agree to be irrationally loyal. If so, can't we dispense with that by now? I like Drescher's derivation of a logical foundation for the Golden Rule even when it benefits no one to abide by it. It's essentially the same logic that leads one to forgo the $1,000 in Newcomb's problem. If I know you're committed to rationality, then I can trust you to abide by the golden rule.

I suspected your interpretation of C&T to be uncharitable, since they seem to be so clearly wrong [to me]. But reading at that link, you appear to have their position correct. Importantly, they characterize the recollections one has about the guilt-inspiring event, and the conscious pondering one often does about alternative ways to have approached that event, as part of the actual mechanism of the learning process for the guilt-module. That sounds unlikely to me. If the mechanism that computes the affective state change is unconscious, why would it's... (read more)


Great post :)

Guilt does not naturally fall out of these explanations. One can imagine a mind design that although often behaving morally for the same reasons we do, sometimes decides a selfish approach is best and pursues that approach without compunction. In fact, this design would have advantages; it would remove a potentially crippling psychological burden, prevent loss of status from admission of wrongdoing, and allow more rational calculation of when moral actions are or are not advantageous. So why guilt?

You could also hypothesize that not being ... (read more)

[This comment is no longer endorsed by its author]Reply

Should "all evolutionary costly" be "all evolutionarily costly"?

The second essay in Nietzsche's 'Genealogy of Morals' is a worthwhile read for anyone interested in guilt. There is an even mixture of dubious/spurious claims here and brilliant insights, which makes it an ideal text for applying rationality to, and sparring with one's own cognitive biases. Second, I hope the kind of person who reflexively/unthinkingly privileges 'analytic' philosophy over the continental kind (this seems to be a recurring theme at lesswrong.com) might think again, or at least become acquainted with a good example of their nemesis. The ef... (read more)

[This comment is no longer endorsed by its author]Reply

This post rings true for me. I have a pretty overactive guilt reaction, and I think I'm considered to be trustworthy and reliable (except when it comes to group academic projects, as per previous posts!). For example, if I make plans with somebody, even the thought of cancelling makes me feel guilty...so I don't cancel on people. Thus I have a reputation as someone who doesn't cancel. More to do with self-behaviour modification than with overt displays of guilt, and I don't think most people know that the reason why I'm reliable is because I feel guilty if I'm not.

So for you guilt is less of a signal in and of itself but the origin of your modified-behavior signal? Several others have raised the point of guilt as an evolutionary facilitator and it makes sense to me that individuals who self-corrected due to a guilt response would be the most socially accepted and thus most likely to procreate. The shoulder-angel doesn't need to be there if the behavior remains the same as if it had been.
0Swimmer963 (Miranda Dixon-Luinenburg) 13y
It makes sense to me, too.

First, guilt is just as associated with good outcomes as bad outcomes. If I kill my brother so I can inherit the throne, then even if everything goes according to plan and I become king, I may still feel guilt. But why should I recalibrate here?

In an evolutionary explanations for an emotion, the outcomes are summed over evolutionary history.

If I kill my brother so I can inherit the throne, then even if everything goes according to plan and I become king, I may still feel guilt.

I think this should be changed to clarify that essentially all human evolution occurred in the context of small bands. There was no evolution to optimize adaptations to account for such a disproportionately large gain as a monarchy with a throne. Maybe: "If I kill my brother so I can inherit more wealth and status, then..."

The disproportionate gain is mostly in currency that evolution doesn't care about anyway. You do get some reproductive advantages from being a monarch but nothing remotely like in proportion to the wealth and power. In reproductive terms you would be better off being the playboy second son of one of the wealthier minor nobility. Lower risk with a lot more time and attention to devote to promiscuity rather than those pesky things like maintaining power, and public image.
I agree with your main point, but don't ignore the evolutionary value of your offsprings' inherited status and how that caches out in terms of their reproductive success.
You're confusing Christian monarchs with monarchs in general.
No, I'm not. The reproductive advantages do not scale with wealth and power in general. You seem to be confusing monarchs in general with a particular instance of a conquerer that was not born of a monarch, the a founder of a monarchy or in any way a king or prince.
Rulers who used their power and wealth to acquire enormous harems and then reproduced like crazy, leaving hundreds or even thousands of children, are not at all uncommon historically. Furthermore, the children of regular royal concubines were typically not in danger of starvation, and thus had a decent chance of reproducing themselves, while the most favored sons would normally become powerful enough to amass their own harems with time. This seems like pretty good scaling. The Christian idea that the ruler is bound by the same moral standards of monogamy as his ordinary subjects is a huge outlier among human cultures. (In fact, I can't even think of any other similar historical example, though someone more knowledgeable could probably find it.) Certainly, if you look at almost any other place and time, you'll find rulers reproducing at rates unthinkable even to their high-ranking subjects.
IIRC, there's a kind of a cultural precedent in the Old Testament notion that a king ought not have too many wives or horses... though I can't remember exactly where that comes from.
Deuteronomy chapter 17. The standard scholarly explanation puts this text as being written at the end of or right after the destruction of the first Temple, so they've had a few hundred years to see all the bad things that kings can do.

If you kill your brother in order to inherit the throne, you probably deserve whatever guilt you feel.

Which is probably relatively little. You don't need nearly as much in the way of guilt when you have that much power. Not to mention if daddy was overly squeamish you probably wouldn't have ended up a prince in the first place.

I find the "don't need nearly as much in the way of guilt when you have that much power" part important, because it sounds plausible and sensible and could be another testable prediction.
And, like most of the other possible tests here, probably not ethical.
But, even if we can't ethically test it as a prediction, we can at least try to test it as a postdiction. Consider people who've committed atrocities ranging from gang violence to the Holocaust, such that punishment comes late and unexpectedly. From the accounts I've read, it seems to me the following are both true and what we would expect based on this theory: 1. While the perpetrators are riding high, in their circle of fellow perpetrators, subject to no disapproval and apparently safe from punishment, they don't feel particularly guilty. 2. Later, when it all comes crashing down and they're in the power of a wider society that despises them and is likely to punish them for their crimes, they often feel - not just consciously fake, but genuinely feel - a great deal of guilt.
6Scott Alexander13y
That sounds more like an argument in favor of Tooby and Cosmides' theory; at the very least it doesn't seem kosher that I used the opposite as support for my theory in the paragraph starting "guilt is just as associated with good outcomes as bad outcomes". Maybe I didn't think it through enough.
Hmm. Thinking about it a little more myself, it seems to me the social is much more important than the material in this regard. In other words, someone who secretly commits a crime and gains great material benefit therefrom, but will be punished if ever caught, is likely to feel guilty. But if the crime is known and approved of by his social circle - even if that's only a gang or terrorist group - and even if he gained no material benefit, he's much less likely to feel guilty (until and unless that social circle is broken and he finds himself in prison).
Maybe this is a big reason why recidivism of imprisoned people is so high. After committing a crime, they get removed from the society in which they'd experience guilt and placed in with people who've done similar things. Or worse things. So the guy who's in prison for selling a kilo of cannabis hears the stories of a hardened home robber, and absorbs the robber's ability to rob guilt-free. Hmm, so, considering the way guilt really plays out with modern adults, I don't think guilt is much more than a conditioned response of submission learned in childhood. It feels bad to be forced to be submissive, and we internalize that bad feeling as a conditioned response to doing something we know is bad.
As another interesting data point, in the early centuries of the Ottoman Empire, the regular succession procedure involved the new sultan killing off his brothers to eliminate any chance of usurpation preemptively. Sultans would often have numerous children with their harem of concubines, so an ascension to the throne often involved a mass killing of the new sultan's half-brothers. As far as I know, the sultans were not known to be tormented by fratricidal guilt.
I don't think this is a disproof -- if you're expected to do something you don't have to feel guilty about it because you could still be trustworthy.
Trustworthy? When you're the one with all the power (and your subjects don't have the option of conveniently relocating away) what you need in the Machiavellian sense is for your subjects to trust that you'll kill anyone who makes themselves a threat.
Fascinating and horrifying. But how could that be stable? Wouldn't the brothers figure out ahead of time that they were going to be killed, and all try to kill each other as soon as they realized it?
I suspect that's a feature. Part of the genius of the Ottoman method is the only sons that become Sultans are survivors.
8Scott Alexander13y
It's good to know you still take your responsibility to defend the Ottomans seriously.
:D I do actually enjoy a lot of the Ottoman system, and don't enjoy playing them in Diplomacy. It's such a nasty little corner, and their administrative innovations don't come into play (and weren't helping them very much at that point in history). The main thing I enjoy about the Ottoman method of finding successors is it makes the most of good genetics (the leader can select the best women and probably has a good supply himself) and education (if you only have 30 potential leaders, you can afford to tutor all of them extensively, which is prohibitively expensive to do for the general population at earlier tech levels) while also minimizing the variance, which is the main problem with normal hereditary rule (if the eldest son gets the throne, you get one shot at a good ruler; if the best son gets the throne, you get the best of 15 shots; if the best child gets the throne, you get the best of 30). I do have a healthy Burkean respect for monarchism- speaking loosely, the theory that no ruler can be worse than a civil war- and trying to ensure that succession conflicts happen within the confines of the palace means that as little collateral damage is done as possible. Of course, this doesn't always pan out (there were a few wars of succession that were proper civil wars).
From what I understand, in the earlier rough and martial times, sultans were mostly busy campaigning and had relatively few sons, who would occupy various military and government posts when they reached adulthood. Upon sultan's death, the sons were expected to fight it out to determine the successor, and this occasional period of instability or even civil war was considered normal. I suppose the brothers would prepare for that moment as best they could. As the rulers became more decadent and started spending more time siring numerous children with their harems, while the government system became less hands-on, their sons effectively became prisoners of palace factions, which were the real powers clashing over succession on sultan's death. Once in charge, the new sultan would have no problem killing his captive and powerless brothers. (Sometimes, if he was without sons, he'd keep them alive until he got a son to make sure the dynasty doesn't die out.) Often the slaughter was extended to other family members as well, just to make sure. Later the custom was made more merciful, and the brothers would be kept imprisoned for life instead. This institution also served as a reserve of potential successors in case the sultan and his sons died out unexpectedly.
I interpreted this to mean that once a sultan takes the throne, he has his brothers killed, rather than the other way around. Not sure which is right!
You kill all the brothers, then get the throne because of it. Not the other way around.
Ok, I'll try that next time it comes up :)
Heh. Tell me how that turns out. ;)
I have the impression that this was relatively common among rulers when it came to succession in general. Almost as common as brothers actually usurping their sibling to claim power. With great power comes great inclination to get more! Seeing this topic discussed always reminds me of Stardust (the movie or book). The king was outright encouraging his sons to defenestrate each other. He was a little disappointed in them - he had killed all his brothers by this age and didn't think they were living up to his standards.
If the crime is approved by his social circle, as far as his mind is concerned it's not really a crime. After all, if you help massacre the rival tribe and take their stuff, there's no reason to feel guilty about this among one's tribe. One only needs to display guilt when dealing with other tribes that are considering forming an alliance against your tribe.