[gurps] Failing autopilots and RVO

[Knapp]

> It's some of both -- I think the real good Go programs tend to run on
> multiple-core machines, and even then they can only beat pros occasionally
> and only if the computer gets a handicap (sometimes a really big handicap).
>
> (I'm not a computer professional but I'm taking some AI courses at
> university and we talked about games a lot last year.)
>
>
> --
> chuk

All basically true but a bit behind the current state of the art.
Handicaps are falling to something normal and pros are loosing more.

There have been some big improvements in the AI logic that has
resulted in pros getting beat a lot. Mostly this has come in the form
of heavy Monti Carlo Tree Searches. Still big handicaps and small
boards and fast computers against the pros. At the current rate of
improvement the pros may fall soon and this is mostly due to software
not hardware.

This leads be to believe that RPG robots and ship AIs should be much
better than they are in most game worlds that I have played in.

I think that when we look back at AI and robots of today in fiction,
it will be just like looking back at computers in fiction 70 years
ago. The impossibly hard part is saying how to do it right, so I have
been trying to look at how not to do it, at least.

-- 

Douglas E Knapp

Why do we live?