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).
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).