Complete.Org: Mailing Lists: Archives: freeciv-ai: June 2003:
[freeciv-ai] Re: learning from experience

[freeciv-ai] Re: learning from experience

[Top] [All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index] [Thread Index]
To: "Alexander L. Vasserman" <avasserm@xxxxxxxxxxxxxxxxxxxxxx>
Cc: freeciv-ai@xxxxxxxxxxx
Subject: [freeciv-ai] Re: learning from experience
From: "Per I. Mathisen" <per@xxxxxxxxxxx>
Date: Mon, 9 Jun 2003 19:33:05 +0000 (GMT)

On Fri, 6 Jun 2003, Alexander L. Vasserman wrote:
> I was wondering if anyone has excercised/implemented the idea of AI
> learning from experience. For eample, for the science selection could
> assign a weight to each technology, which would be combined with current
> game factors (want?) to figure out which technology is the most valuable.
> The weights then can be adjusted deppending on wether the game was won or
> lost. For a lost game technologies that have been researched early would
> be penalised, for a won game they would be encouraged. Thisis just
> one possible way of doing it. I am sure I could come up with more
> interesting strategies. My question is more about the concept itself. Some
> randomization of these decisions would be required, but I think this
> would be a good thing. In addtion to hopefully improving the AI it would
> also descourage a single winning strategy. If a strategy used too often
> the AI would adjust to it, and at least will make you look for a new one.

So the AI should adapt to whoever it plays against? I guess it could check
which username the opponent has...

In any case, there is a problem: Changing context. A strategy that works
fine in, say, gen 1, may not work as well in gen2. Some slight changes in
server options might change context enough to throw its accumulated
weights into question.

What you describe would be quite useful in order to optimize the AI. We
could put researched weights into the rulesets. However, I am pessimistic
about the state of AI research when it comes to being able to write an AI
that can figure out on the fly and on its own new strategies and read
something meaningful into accumulated statistical data. But I would be
happy to be proven wrong.

  - Per

[Prev in Thread] Current Thread [Next in Thread]