[ExI] Unfrendly AI is a mistaken idea.

[John K Clark]

"Eliezer S. Yudkowsky" <sentience at pobox.com>

> I would like to know what you believe (and remember, this is a question of
> simple fact) is the state of mind of someone who would like to build a
> Friendly AI.

I don't really like to play the part of a psychiatrist, but you specifically
asked me to do so and try to get into the head of the friendly AI people, so
whatever it's worth here is my attempt to be an amateur shrink.

For generations Caucasians have observed black people and noticed that they
seemed to have emotions, but they convinced themselves that they couldn't
have really deep emotion like they themselves did because, well., because
they weren't white. Therefore they could treat black people like shit and
even own them with no guilt. The friendly AI people have the additional
problem of explaining away the fact that the slave in question is without a
doubt vastly more intelligent than they are; I imagine they rationalize this
by saying, against all the evidence, that emotion is harder to achieve than
intelligence, that emotion is the secret sauce that only a meat brain can
produce never a silicon brain; so they delude themselves that the super
intelligent AI is just a souped up adding machine. Or perhaps they think
emotion is  something tacked on and they just won't tack it onto their AI,
as if one also needed to tack on a Beethoven circuit on a radio if you
wished it to play Beethoven. And then I imagine they just refuse to think
how evolution could ever have produced emotion if it weren't intimately
linked to intelligence.

But the above is of academic interest only because there is not a snowball's
chance in hell of outsmarting a mind a thousand times smarter and a million
times faster than your own. However it cannot be denied that they sincerely
believe they can accomplish this imposable task, although they never give a
hint how to go about it. And at this point my very modest psychoanalytical
abilities fail me completely. What on Earth were this friendly AI people
thinking?

 John K Clark