Re: group-based judgement

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Thu May 30 2002 - 12:28:07 MDT


Wei Dai wrote:
>
> On Fri, May 24, 2002 at 11:01:53AM -0400, Eliezer S. Yudkowsky wrote:
> > However, there are ethical implications to altering our treatment of people
> > based on prior probabilities rather than observed behavior. Bayes's theorem
> > necessarily influences our estimate of the prior probabilities, but does not
> > necessarily control whether we choose to alter our treatment based on those
> > probabilities. Arguably the situation is viewed as an iterated Prisoner's
> > Dilemna in which we agree not to prejudge in exchange for not being
> > prejudged by others.
>
> No, prejudgement is not analogous to iterated Prisoner's Dilemma. Here's
> a simple example that shows why: Consider a two-player game with two types
> of players. Type A are those who, if they were prejudged, would be
> prejudged favorably. Type B are those who would be prejudged unfavorably.
> Then a type A player has no incentive to not prejudge, since the
> "retaliation" he might get, being prejudged in return, doesn't hurt him;
> in fact it actually helps him. Now consider a type B player. If he faces a
> type A player he might as well prejudge since he's being prejudged anyway.
> If he faces another type B player, he can "not prejudge" but that's now
> meaningless.

But there are many different factors on which any of us can be prejudged. I
can be prejudged as having more than two Jewish grandparents, as a white
male, as a Georgian (male), as a twenty-something (male), or many other
stereotypes which would be utterly misleading to anyone trying to sketch out
a picture of Eliezer Yudkowsky. There are not "A" players and "B" players.

> > To put it another way, being judged only by your
> > actual, personal actions and not your statistical associations with the
> > actions of others is a public good; it helps preserve the ethical structure
> > of reciprocal altruism, which requires that you adjust your treatment of an
> > agent based on those actions that are subject to the judged agent's control.
>
> That doesn't make sense. Just because someone has been prejudged, that
> doesn't mean his actions can't further affect other people's judgement of
> him.

For humans, it usually *does* mean exactly this. Especially if the
prejudgement is such as to rule out further interaction with that person.
If we had evolved as Bayesian reasoners instead of prejudiced snap
stereotypers there would be much less of a problem. If someone prejudges me
I am not likely to stick around and try and alter that person's opinion; I
am likely to walk out and look for someone more reasonable to talk to.

-- -- -- -- --
Eliezer S. Yudkowsky http://singinst.org/
Research Fellow, Singularity Institute for Artificial Intelligence



This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 09:14:31 MST