Ordinary need not get lots of upvotes- I think I agree with you in that- but “not bad” shouldn’t be downvoted. More germanely, I think “better than average” is worth rewarding with a complement if you can catch it in the moment.
Some years ago I started teaching a handful of kids from my community basic rationality skills. The first thing I taught them was probability, drawing cards with and without replacement and trying to build up an idea of what statistics meant, and I clearly remember praising them when they realized that if 60% of the cards are red and 40% are blue, you always guess that the next card will be red instead of guessing red most of the time and blue some of the time. A couple of weekends ago my roommate (who is usually a couch potato) asked if it was okay if they went with me when I was heading out for a walk, and after we got back I told them it was nice to have company and that walking with them was fun, because I want them to do that more often.
When somebody does something you wish people would do, I think a quick complement or piece of praise is a fine way to <strike> classically condition your friends and family</strike> make them feel good about it. I would clearly distinguish “You’re the best writer!” from “you’re really good at writing!” from “I liked this thing you wrote, particularly this piece.” The first is the greatest complement from a literal, absolute level, but the last is the one that people seem to feel most and is also true more often.
Yup. Most compliments are recognizably fake. It takes more than a few words to give someone a sense of worth. You have to actually believe they are worthy.
And that’s only possible if your standards are reasonable. I believe there should be a set of traits that most of us could achieve (with some effort), that we hold to be sufficient for worthiness. I think our current set of standards is largely inherited from Eliezer’s writings, who derived them from what he held to be necessary to fix AI. Some of those standards might be simply too high to achieve.
And then of course the question becomes: which achievable standards could lead us to fixing AI, or at least maximizing our chance to do so?
Hrm. I find myself wanting to disagree with this comment while agreeing with your original post. I think there’s three distinct levels worth thinking about.
There’s worthiness as in “self-worth” or “worth helping.” The world is probably better if the bulk majority of people have this, and I have heard people express the idea that every human is worthwhile in this sense. That’s not to say me or you can’t prioritize who we care about, but “such and such people aren’t worth the air they breath” is a dangerous line of reasoning. Complements aren’t particularly useful here, as “You can use language, therefore your existence is positive” can frankly come off as a backhanded insult of sorts, since that’s a really low bar.
There are correct steps in the right directions well*, including most personal growth and including hill climbing towards better states. This is the place I think complements are best deployed; an adult human taking a ten minute walk outside is better than that human sitting on the couch watching reruns. (I recognize I’m making a value claim there that may not be globally correct.) Guessing “red” all the time in the probability question above is better than guessing “red” 60% of the time and guessing “blue” 40% of the time. Progress is worth appreciating, both on the personal level (“hey, congrats on beating your mile-run time!”) and on the group level (I watched a time-lapse map of malaria cases in a room full of EAs recently, and I am kinda disappointed that nobody cheered.) It doesn’t even have to be a new achievement! In martial arts, I eventually reached the point where every other session the instructor would nod and say “good stance” before moving on. This never stopped feeling good to hear, and it kept the basics in my mind even as I moved on to more advanced steps.
Then there’s being correct on an absolute scale. The kind of rightness that involves local validity and correct premises, the kind that gets measured against the real world and succeeds. A successful rocket launch, a healthy patient after surgery, an AI that does what we meant and not what we said. The universe does not grade on a curve and gives no awards for effort. I think if we as aspiring rationalists lose sight of this, then we will eventually go astray no matter how good we are at the first two*. Complements here are rare, but powerful.
My suggested heuristic for the community would be to complement someone when you know them and see them advance along the path, or when they do something which helps you advance*. I also offer complements when someone does something I want them and/or others to do more of even if it is not novel, and I suspect that this kind of complement is what you are seeking to encourage; if so, then we are in agreement. “Good stance” is important to hear, as is “good job updating” and even “hey, good job organizing the meet up yesterday! I think you pretty good moderating, you jumped in at the right moment when me and Bob were getting derailed.” Praise for getting things right, with the promise of more encouragement as we climb higher.
*To be clear, I don’t think there’s a single linear ladder we climb straight up from ignorance to superrationality. There are probably multiple paths to the summit, and there may well be more than one peak. That’s a different topic however.
I think this comment is more correct than not, so take the following as additions rather than disagreements per se:
First, the third case you mention—what you call “being correct on an absolute scale”. I think a more natural way to describe such things is simply “accomplishing tangible goals”, “getting things done”, “succeeding at tasks”, or something along these lines. (That is what we’re talking about, yes?)
More to the point, I think that to speak of “compliments” in such cases is to somewhat miss the point. After all, the outcome of your effort, in such scenarios—that is, the actual goal that was accomplished, the actual task completed, etc.—is good (valuable, beneficial, worthwhile, etc.), separately from the fact that you accomplished it. The accomplishment, result, etc., is an objectively verifiable good that has been created, which has whatever value it has regardless of what anyone says or thinks about it—whereas the notion of “compliment” implies that the worth of one’s actions are up to the complimenter to determine.
So if someone were to compliment me for creating something, doing something that needed to be done, or otherwise creating value, it would be slightly insulting; is this person suggesting that the value of what I made or did is determined by their evaluation of it? A compliment is aimed at a person, which is the wrong focus here; the right focus is the creation, or the result, or whatever.
No, the relevant concept here, I think, is not complimenting but recognition. When I make something of value, or do something useful, what I want isn’t for people to evaluate me, and to judge me positively for my actions; it’s to recognize that what I have done or made has value, and to credit me for that value. (Side note: Eric Raymond’s essay “Homesteading the Noosphere” explores the role of this desire in creating the “reputation economy” on which the free software movement is based.)
(One important distinction is that recognition—a.k.a. credit—is generally seen as being owed; we have a notion that recognition is something that creators, or those responsible for getting things done, deserve to be recognized for it; this is, of course, not the case for compliments, the giving of which is, at most, supererogatory.)
(comment snipped for length and clarity of discussion threading)
Yes. This. Very much this.
I get a sense that recognition, as opposed to positive judgment, is more durable. It’s not just that you did a good thing, it’s that the thing you did is a reflection of your good character, and we expect you to do more good things, and we want to keep you around and support you.
I hate to contradict someone when they’re agreeing with me, but… it seems like I didn’t get my point across very well. What you describe is actually the opposite of what I meant by “recognition”.
… I am at a loss as to how to better explain my point, however. Perhaps a brief self-quote:
When I make something of value, or do something useful, what I want isn’t for people to evaluate me, and to judge me positively for my actions; it’s to recognize that what I have done or made has value, and to credit me for that value.
“It’s not just that you did a good thing, it’s that the thing you did is a reflection of your good character” is the diametric opposite of that!
Huh. Okay. So it’s not about recognizing your worth, but about recognizing the worth of your work, and associating that work with you. Does that capture it better?
But that’s not far from recognizing that good work can be expected from you, which is the thing you actually game-theoretically want, right? That’s what I mean with “good character”, in the sense that the community is incentivized to support you and invest in you.
So it’s not about recognizing your worth, but about recognizing the worth of your work, and associating that work with you. Does that capture it better?
That’s it exactly, yes.
But that’s not far from recognizing that good work can be expected from you, which is the thing you actually game-theoretically want, right?
Well, that’s as may be. In my view, it’s important to cleanly separate these concerns, though. Conflating them gets you into the territory of people evaluating one another on fuzzy personal criteria, holistic “character judgments”, etc., etc. That’s not a good road to go down (as this discussion reminds us—and as we might also recall from, e.g., the work of Robyn Dawes).
The ethos I am defending, in other words, is “my work speaks for itself”. Don’t evaluate me—evaluate what I have done (or made, accomplished, etc.). Since you need not take my character into account when doing so, there is a strong and important sense in which evaluation of my work is objective. (And, of course, that it was I who did the work is simply a verifiable empirical fact, on which no debate is needed or possible.)
I insist on this point not out of a desire to be pedantic, but because I really do consider it a critical norm for any healthy, productive society, subculture, etc.
Edit: On the subject of “what I game-theoretically want”:
Edit2: Whoops, I hit the wrong button and posted this as a reply to myself instead of an edit. Could a moderator please move this comment up one level (i.e., make it a reply to toonalfrink)?
A norm that one’s work is evaluated, rather than one’s character, incentivizes people to do good work. A norm that one’s character is evaluated (with one’s work being only a component of this evaluation—if, perhaps, an important one) incentivizes one to ensure that one’s character is judged highly. If what we (collectively) want is for people to do good work, then it should be obvious that the latter norm causes the “good work” goal to be instantly Goodharted.
I don’t think the former is free from Goodharting either. My sense of a good community is one where we get the character judgment out of the way from the start. So indeed “people evaluating one another on fuzzy personal criteria”. In the sense of “hey we like you for the things about you you can’t change even if you tried”. So personal value is secured, meaning that the person can actually start to pursue the things they value truly for their own sake.
As I said in other places: If I got all of my needs out of the way, I would still work on AI Safety (which I value for it’s own sake), and have a lot more cognitive bandwidth to allocate to it too. All of which is now going to securing my worth. Which is essentially Goodharting, since I’m incentivized to skew everything I do to things that can be easily used for signaling.
An unsatisfied satisficer is a maximizer. I’m maximizing my status, and the useful work I’m doing is only a side effect. That doesn’t seem like a good thing. Especially with a security mindset.
How should we evaluate someone who is trying something they haven’t done before? If it’s by looking at how often they seem to succeed in general, that seems hard to distinguish from a “holistic character judgment.”
It seems to me that these are specific purposes, and we can make specific predictions for each of them. Even if we combine them into some sort of holistic judgment (which I think is worth trying to avoid, when possible, but may be unavoidable in some cases), we ought nonetheless to do that only after we have, at our disposal, the cleanly separate evaluations of a person’s work (with which we have properly credited them).
Second, I think there’s a sense of “worth” which you have left off, and it’s quite an important one. This is the notion of worth to someone. (This Cracked article frames the notion particularly starkly.)
Note that neither universal worth (in the “all people deserve love / dignity / etc.” sense), nor personal accomplishment (“I did 100 push-ups!”), nor absolute accomplishment (“I successfully launched a rocket”), capture this sense of “worth”. It’s not the case that all people are equally valuable to any given person or people; your personal accomplishments—while not entirely uncorrelated with your value to others—are largely unlinked therefrom; and while absolute accomplishments are, indeed, likely to be of value to someone (though there are notable exceptions even here), they may not be of value to the specific someones you care about, or want to impress, or desire recognition from, etc.
And the fact is that people are moved to give compliments, just as they are moved to do most anything else, by what they care about, what is of value to them, what they feel serves their goals, needs, plans, etc.
All of these are simply facts about the world—“is” statements, not “oughts”. Yet without a doubt there is a temptation, for some folks, to read what I’m saying as “suck it up; if you’re not valuable to others, they won’t compliment you, not matter how ‘deserving’ you may be in whatever way you feel is relevant”. That is not my intent. In fact, the policy I would recommend (and the one I follow in my own life) is, in many ways, the inverse of that harsh reply. But we have to recognize the facts before we can do anything about them.
Ordinary need not get lots of upvotes- I think I agree with you in that- but “not bad” shouldn’t be downvoted. More germanely, I think “better than average” is worth rewarding with a complement if you can catch it in the moment.
Some years ago I started teaching a handful of kids from my community basic rationality skills. The first thing I taught them was probability, drawing cards with and without replacement and trying to build up an idea of what statistics meant, and I clearly remember praising them when they realized that if 60% of the cards are red and 40% are blue, you always guess that the next card will be red instead of guessing red most of the time and blue some of the time. A couple of weekends ago my roommate (who is usually a couch potato) asked if it was okay if they went with me when I was heading out for a walk, and after we got back I told them it was nice to have company and that walking with them was fun, because I want them to do that more often.
When somebody does something you wish people would do, I think a quick complement or piece of praise is a fine way to <strike> classically condition your friends and family</strike> make them feel good about it. I would clearly distinguish “You’re the best writer!” from “you’re really good at writing!” from “I liked this thing you wrote, particularly this piece.” The first is the greatest complement from a literal, absolute level, but the last is the one that people seem to feel most and is also true more often.
Yup. Most compliments are recognizably fake. It takes more than a few words to give someone a sense of worth. You have to actually believe they are worthy.
And that’s only possible if your standards are reasonable. I believe there should be a set of traits that most of us could achieve (with some effort), that we hold to be sufficient for worthiness. I think our current set of standards is largely inherited from Eliezer’s writings, who derived them from what he held to be necessary to fix AI. Some of those standards might be simply too high to achieve.
And then of course the question becomes: which achievable standards could lead us to fixing AI, or at least maximizing our chance to do so?
Hrm. I find myself wanting to disagree with this comment while agreeing with your original post. I think there’s three distinct levels worth thinking about.
There’s worthiness as in “self-worth” or “worth helping.” The world is probably better if the bulk majority of people have this, and I have heard people express the idea that every human is worthwhile in this sense. That’s not to say me or you can’t prioritize who we care about, but “such and such people aren’t worth the air they breath” is a dangerous line of reasoning. Complements aren’t particularly useful here, as “You can use language, therefore your existence is positive” can frankly come off as a backhanded insult of sorts, since that’s a really low bar.
There are correct steps in the right directions well*, including most personal growth and including hill climbing towards better states. This is the place I think complements are best deployed; an adult human taking a ten minute walk outside is better than that human sitting on the couch watching reruns. (I recognize I’m making a value claim there that may not be globally correct.) Guessing “red” all the time in the probability question above is better than guessing “red” 60% of the time and guessing “blue” 40% of the time. Progress is worth appreciating, both on the personal level (“hey, congrats on beating your mile-run time!”) and on the group level (I watched a time-lapse map of malaria cases in a room full of EAs recently, and I am kinda disappointed that nobody cheered.) It doesn’t even have to be a new achievement! In martial arts, I eventually reached the point where every other session the instructor would nod and say “good stance” before moving on. This never stopped feeling good to hear, and it kept the basics in my mind even as I moved on to more advanced steps.
Then there’s being correct on an absolute scale. The kind of rightness that involves local validity and correct premises, the kind that gets measured against the real world and succeeds. A successful rocket launch, a healthy patient after surgery, an AI that does what we meant and not what we said. The universe does not grade on a curve and gives no awards for effort. I think if we as aspiring rationalists lose sight of this, then we will eventually go astray no matter how good we are at the first two*. Complements here are rare, but powerful.
My suggested heuristic for the community would be to complement someone when you know them and see them advance along the path, or when they do something which helps you advance*. I also offer complements when someone does something I want them and/or others to do more of even if it is not novel, and I suspect that this kind of complement is what you are seeking to encourage; if so, then we are in agreement. “Good stance” is important to hear, as is “good job updating” and even “hey, good job organizing the meet up yesterday! I think you pretty good moderating, you jumped in at the right moment when me and Bob were getting derailed.” Praise for getting things right, with the promise of more encouragement as we climb higher.
*To be clear, I don’t think there’s a single linear ladder we climb straight up from ignorance to superrationality. There are probably multiple paths to the summit, and there may well be more than one peak. That’s a different topic however.
I think this comment is more correct than not, so take the following as additions rather than disagreements per se:
First, the third case you mention—what you call “being correct on an absolute scale”. I think a more natural way to describe such things is simply “accomplishing tangible goals”, “getting things done”, “succeeding at tasks”, or something along these lines. (That is what we’re talking about, yes?)
More to the point, I think that to speak of “compliments” in such cases is to somewhat miss the point. After all, the outcome of your effort, in such scenarios—that is, the actual goal that was accomplished, the actual task completed, etc.—is good (valuable, beneficial, worthwhile, etc.), separately from the fact that you accomplished it. The accomplishment, result, etc., is an objectively verifiable good that has been created, which has whatever value it has regardless of what anyone says or thinks about it—whereas the notion of “compliment” implies that the worth of one’s actions are up to the complimenter to determine.
So if someone were to compliment me for creating something, doing something that needed to be done, or otherwise creating value, it would be slightly insulting; is this person suggesting that the value of what I made or did is determined by their evaluation of it? A compliment is aimed at a person, which is the wrong focus here; the right focus is the creation, or the result, or whatever.
No, the relevant concept here, I think, is not complimenting but recognition. When I make something of value, or do something useful, what I want isn’t for people to evaluate me, and to judge me positively for my actions; it’s to recognize that what I have done or made has value, and to credit me for that value. (Side note: Eric Raymond’s essay “Homesteading the Noosphere” explores the role of this desire in creating the “reputation economy” on which the free software movement is based.)
(One important distinction is that recognition—a.k.a. credit—is generally seen as being owed; we have a notion that recognition is something that creators, or those responsible for getting things done, deserve to be recognized for it; this is, of course, not the case for compliments, the giving of which is, at most, supererogatory.)
(comment snipped for length and clarity of discussion threading)
Yes. This. Very much this. I get a sense that recognition, as opposed to positive judgment, is more durable. It’s not just that you did a good thing, it’s that the thing you did is a reflection of your good character, and we expect you to do more good things, and we want to keep you around and support you.
YES. This.
I hate to contradict someone when they’re agreeing with me, but… it seems like I didn’t get my point across very well. What you describe is actually the opposite of what I meant by “recognition”.
… I am at a loss as to how to better explain my point, however. Perhaps a brief self-quote:
“It’s not just that you did a good thing, it’s that the thing you did is a reflection of your good character” is the diametric opposite of that!
Huh. Okay. So it’s not about recognizing your worth, but about recognizing the worth of your work, and associating that work with you. Does that capture it better?
But that’s not far from recognizing that good work can be expected from you, which is the thing you actually game-theoretically want, right? That’s what I mean with “good character”, in the sense that the community is incentivized to support you and invest in you.
That’s it exactly, yes.
Well, that’s as may be. In my view, it’s important to cleanly separate these concerns, though. Conflating them gets you into the territory of people evaluating one another on fuzzy personal criteria, holistic “character judgments”, etc., etc. That’s not a good road to go down (as this discussion reminds us—and as we might also recall from, e.g., the work of Robyn Dawes).
The ethos I am defending, in other words, is “my work speaks for itself”. Don’t evaluate me—evaluate what I have done (or made, accomplished, etc.). Since you need not take my character into account when doing so, there is a strong and important sense in which evaluation of my work is objective. (And, of course, that it was I who did the work is simply a verifiable empirical fact, on which no debate is needed or possible.)
I insist on this point not out of a desire to be pedantic, but because I really do consider it a critical norm for any healthy, productive society, subculture, etc.
Edit: On the subject of “what I game-theoretically want”:
Edit2: Whoops, I hit the wrong button and posted this as a reply to myself instead of an edit. Could a moderator please move this comment up one level (i.e., make it a reply to toonalfrink)?
A norm that one’s work is evaluated, rather than one’s character, incentivizes people to do good work. A norm that one’s character is evaluated (with one’s work being only a component of this evaluation—if, perhaps, an important one) incentivizes one to ensure that one’s character is judged highly. If what we (collectively) want is for people to do good work, then it should be obvious that the latter norm causes the “good work” goal to be instantly Goodharted.
Hm.
I don’t think the former is free from Goodharting either. My sense of a good community is one where we get the character judgment out of the way from the start. So indeed “people evaluating one another on fuzzy personal criteria”. In the sense of “hey we like you for the things about you you can’t change even if you tried”. So personal value is secured, meaning that the person can actually start to pursue the things they value truly for their own sake.
As I said in other places: If I got all of my needs out of the way, I would still work on AI Safety (which I value for it’s own sake), and have a lot more cognitive bandwidth to allocate to it too. All of which is now going to securing my worth. Which is essentially Goodharting, since I’m incentivized to skew everything I do to things that can be easily used for signaling.
An unsatisfied satisficer is a maximizer. I’m maximizing my status, and the useful work I’m doing is only a side effect. That doesn’t seem like a good thing. Especially with a security mindset.
How should we evaluate someone who is trying something they haven’t done before? If it’s by looking at how often they seem to succeed in general, that seems hard to distinguish from a “holistic character judgment.”
Why do we need to evaluate them?
Allocation of resources, attention, affiliation, and so on.
It seems to me that these are specific purposes, and we can make specific predictions for each of them. Even if we combine them into some sort of holistic judgment (which I think is worth trying to avoid, when possible, but may be unavoidable in some cases), we ought nonetheless to do that only after we have, at our disposal, the cleanly separate evaluations of a person’s work (with which we have properly credited them).
(continuation of sibling comment)
Second, I think there’s a sense of “worth” which you have left off, and it’s quite an important one. This is the notion of worth to someone. (This Cracked article frames the notion particularly starkly.)
Note that neither universal worth (in the “all people deserve love / dignity / etc.” sense), nor personal accomplishment (“I did 100 push-ups!”), nor absolute accomplishment (“I successfully launched a rocket”), capture this sense of “worth”. It’s not the case that all people are equally valuable to any given person or people; your personal accomplishments—while not entirely uncorrelated with your value to others—are largely unlinked therefrom; and while absolute accomplishments are, indeed, likely to be of value to someone (though there are notable exceptions even here), they may not be of value to the specific someones you care about, or want to impress, or desire recognition from, etc.
And the fact is that people are moved to give compliments, just as they are moved to do most anything else, by what they care about, what is of value to them, what they feel serves their goals, needs, plans, etc.
All of these are simply facts about the world—“is” statements, not “oughts”. Yet without a doubt there is a temptation, for some folks, to read what I’m saying as “suck it up; if you’re not valuable to others, they won’t compliment you, not matter how ‘deserving’ you may be in whatever way you feel is relevant”. That is not my intent. In fact, the policy I would recommend (and the one I follow in my own life) is, in many ways, the inverse of that harsh reply. But we have to recognize the facts before we can do anything about them.