On Wed, 11 Jul 2001, Ian Phillipps wrote:
> On Tue, 10 Jul 2001 at 13:12:37 -0300, Raul Dias wrote:
> >
> > How can a markov based R-P-S can be unbeatable without cheating?
>
> It chooses each of R, P and S with probability 1/3. This is unbeatable.
> Fairly trivial as Markov processes go.
[snip]
> http://www.research.ibm.com/journal/sj/393/part2/orwant.pdf
Actually, if you read the text carefully, it doesn't. Choosing randomly
is indeed the optimal strategy in RPS, but most human players find it
very difficult to play optimally, since we're notoriously bad at coming
up with truly random sequences.
This, of course, leaves us vulnerable to a counter-strategy, which is
what the Orwant program takes advantage of. Of course, that leaves it
in turn vulnerable to yet _another_ counter-strategy, but the adaptive
nature of the algorithm should ensure that it won't lose very much, as
it should settle into a random strategy if playing against an opponent
it can't successfully predict.
--
Ilmari Karonen - http://www.sci.fi/~iltzu/
"This is why Denver had a gigantic red-light district downtown in the early
days. A man may love his horse, but a man can't really ... _love_ his horse."
-- Charles Martin in rec.arts.sf.science
