Eliezer S. Yudkowsky wrote:
>>recent development has not amounted to a
>>wholesale rewriting of our moral programming.
>Whose moral programming are we talking
>about? Mine is fairly well rewritten - I wrote
>the specs for an AI and I try to make mine
>match the specs, to the limits of what you
>can do with software alterations.
I was referring to people in general. The imprint of our evolutionary heritage is still very visible the behaviors and choices of 20th century humans, although the cultural evolution of the past few thousand years has begun to create many subtle and interesting variances.
We should be cautious in evaluting how thoroughly any of us have separated ourselves from much of this moral programming, which includes all the broad imperatives of survival. Suicide is one way to prove that you're no slave to your genes, but so is using condoms and remaining childless. We may find independent, *philosophical* rationales for all of our actions, but that doesn't mean that the genenetic legacy isn't still still playing an influential, covert role in our many of choices (which isn't necessarily bad).
>>To give robots Asimov-type laws would
>>be a planned effort to do to them what
>>evolution has done to us - make them
>>capable of co-existing with others.
>Wrong-o. Nothing that can learn - nothing
>that has a data repository that can
>change - can have an irrevocable moral
Certainly not irrevocable, but highly resistant to drastic change. A "conservative" set of moral instincts might be responsible for the failure of the human race to have committed mass suicide by now. Where once the proto-human ate, surived, and reproduced for no conscious reason, now we stare at the sky and ask "what's it all about?" to no certain answer. So we create meaning systems like religion that give us the intellectual justification to carry on. Would we have bothered doing the same absent a billion years of genetic survival instincts?
>We aren't talking about human moral
>systems, which derive from game theory
>and evolution. We're talking about AI moral
>systems, which are quite different.
I'm not sure I understand that comment. AI morality will be designed and programmed by conscious humans, unlike our own morality which was designed mostly by the unconscious workings of mother nature. This is a big advantage, since we'll be able to get rid of vestigal instincts like "use agression against conspecifics to ensure the maximum proliferation of your offspring," along with many other outdated tendencies that linger to find new outlets of expression in the modern world.
But in creating a moral system de novo (for AI's, for ourselves, or for both) we can only put into it what we bring to the table. Our knowledge of our own moral programming and its strenghts and weaknesses will have to be the jumping off point for any new design.