Re: Friendliness not an Add-on

From: Chris Capel (pdf23ds@gmail.com)
Date: Tue Feb 21 2006 - 11:44:11 MST


On 2/21/06, BillK <pharos@gmail.com> wrote:
> Our top board player was in a different class to all the other team
> members. He would win games consistently against us mere mortals and
> we really couldn't see what he was doing. We had a record of the games
> to analyse. We weren't making 'wrong' moves and he wasn't making
> brilliant moves that we had overlooked. It was much more subtle than
> that. More like a game of Go, where you have to 'shape' the game.

You weren't making wrong moves that you could understand were wrong
with your limited-depth, shallow-search knowledge of the board
position, but that he could plainly see were wrong with his
expandend-depth knowledge of the board. His moves may have been
inscrutable (while not obviously bad) to you, but that doesn't mean
they weren't brilliant, at least compared to your less capable
playing. So, in a sense, you *were* making wrong moves, and he *was*
making brilliant moves.

Or, did you mean, you weren't making blunders (isolated moves of much
lower quality than average), and he wasn't making isolated, especially
good moves?

> The point is that an advanced AI will be planning and producing
> schemes that we have no chance of understanding. We will see the moves
> but that won't help much.

We'll see the actions, but we won't know how they could possibly fit
into a larger plan. Indeed, if we were told the plan we might not even
be able to keep it all in our heads at once, let alone begin to
understand it.

I do think, though, that the experience of being good at chess, of
being able to win consistently against other knowledgable players, and
then playing someone much better than you, who squeezes your position
quickly and inexorably, and who has a tactical trap around every
corner, and who can beat you consistently by the midgame, is probably
something that induces the same sort of attitude we ought to take
towards a superintelligence, only much more so. Chess has a closed
search space, unlike real life. In real life, rules are provisional,
and "breaking the rules" just means discovering and exploiting holes
in your opponent's knowledge of reality.

And, yes, we should be careful with our chess metaphors.

Chris Capel

--
"What is it like to be a bat? What is it like to bat a bee? What is it
like to be a bee being batted? What is it like to be a batted bee?"
-- The Mind's I (Hofstadter, Dennet)


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:55 MDT