From: Michael Roy Ames (email@example.com)
Date: Sun Feb 09 2003 - 12:43:02 MST
Gordon Worley wrote:
> Whether we need to play games or not is still at question. [snip]
I have answered that question postively, for the following reasons.
1) A game can be competitive, or non-competitive. The degree of
competitivenes can be arbitrarily set in the design of the game.
2) A game can be an analogy of reality, a model, within which:
a) a Seed AI can learn skills and techniques...
and can be *seen* to learn them (or not).
b) there are no negative real-world consequences
to success/non-success (winning/losing).
3) A game can be iterated, raising the level of difficulty at each
4) A game can be re-interpreted in a new modality, fostering the
building of analogies.
5) A game can embody an idea, or set of ideas, that can be referred back
to symbolically, assisting in the bootstrapping of language
> So, even assuming we can develop a game free of conflict (I think
> we can, but it's going to take a hell of a lot of work and a
> proof of noncompetitiveness), it's still not clear that a seed
> AI should be playing games at all.
Conflict and competitiveness in human societies are there for certain
reasons. Some of these human reasons will be totally unsuitable for
inclusion in Seed AI games, as the Seed AI will not have human
psychology. But to the extent that conflict/competition can be
effectively utilized by minds-in-general, these idea can be best
introduced to a Seed AI in the context of a game - with a positive-sum
purpose/reason **as an integral part of the game**.
> A seed AI certainly needs some kind of training, but I'm not sure
> that sports/games as we think of them are going to be the best
Your phrase, "as we think of them" leads me to doubt we are talking
about the same definition for the word 'game'. Here are a couple of
definitions I am using:
game n. (Form of) contest played according to rules and decided by
skill, strength or luck. <Concise Oxford>
game n. A competitive activity or sport in which players contend with
each other according to a set of rules. <American Heritage>
The key aspects being: a) a contest b) within some rules c) producing an
outcome. The human propensity to attach postive/negative judgements to
'winning' and 'losing' is not an elemental part of the game idea IMO.
An outcome of non-success, of 'losing', will often teach the contestent
more than winning. If we take away the emotional baggage, a 'losing'
outcome is often positively useful in learning. As long as the playing
of a game is taught to be a sub-subgoal of the subgoal: Learning, then
'winning' or 'losing' is incidental, and of little overall consequence.
As Ben Goertzel opined, teaching an AGI to play games could be entirely
a positive thing, as long as ve understands, "it's just a game."
At first, the competitiveness of the games presented should (IMO) be
very low or zero. But as the Seed AI gains understanding of a wider
context, competitive elements can be added with confidence that they can
(and will) be explained in that wider context. Competition can be a
learned technique that benefits all, players and non-players alike.
Instead of "Ha! I win! I win! Now I can squash you like a bug!" we would
have, "Hoohoo! I win! Now let's apply this new technique to solving the
protein folding problem."
Michael Roy Ames
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:41 MDT