[gurps] AI's in GURPS (was Failing autopilots and RVO)

[Knapp]

On Thu, Mar 5, 2009 at 2:08 PM, Johannes Trimmel
<a9300617 at unet.univie.ac.at> wrote:
> On Thu, 5 Mar 2009, Knapp wrote:
>
>> > What is creativity as far as GURPS goes?
>>
>> That is a great question, period, even without the GURPS bit!
>> In is right in there with what is intelligent.
>>
>> A modern Monti Carlo based go program comes up with moves that no
>> human or computer has ever played before and it will not play those
>> same moves again, yet they are good moves. Is that creativity?
>>
>
> That i would call unpredictability.
>
> Propably you could have a difference in creativity in 2 Monte Carlo
> programms, with the more creative program having a better scoring
> function. Or propably by calculating longer pathes.
>
> Mostly creativity seems to be about recognizing that some non standard
> approach actually has merits in the given situation. The prime reason for
> a Monte Carlo program failing at that would be either to fail to notice
> that this approach would be successfull in the end and reject it, because
> some transient state would have a bad score. So it could be fixed by a
> better scoring function, that recognizes the potential the transient state
> has, or by searching paths that go beyond that transient state.
>
>
>
>
> One mans groundfloor is an other mans earthmissle
> @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@
>
> Johannes Trimmel

In go programs they found that scoring did not work well so they don't
use it. They start at the current next move and then play random moves
for both sides to the end a bunch of times. The next move that has the
highest percentage of wins is the next move picked. Thus your problem
is not a problem at least in the go example although they did try that
scoring idea.

-- 
Douglas E Knapp

Why do we live?