[computer-go] Abstract analysis of Monte Carlo playout

[Eric Boesch]

On 7/28/07, forrestc at adnc.com <forrestc at adnc.com> wrote:
> > So even though you the playout agent has only 50% probability of
> > playing correctly, the probability that after 2 plys the position is
> > still won is 75%!
>
> going toward a limit of 66.6% as the number of plies increases

That is a parity artifact, as Antti's second chart showed. But it
reflects a truth: the player who moved last had the last chance to
make a mistake, and therefore the next player may have better than
even odds to win. This effect will be severe in a Nim game between two
players who don't understand the position -- if you are not sure you
chose a winning move then you probably chose a losing one. In go,
however, advantages can accumulate and the leader is more stable, and
the game has an element of random walk to it as blunders push one the
score further in one player's favor or the other.