[computer-go] What is All-as-first-heuristic?

[Chrilly]

Jan Lowmans and Heli Weigels law are of course simplifications. In case of 
the modern very deep searches the search has also significant impact on the 
playing style. Generally the programm prevers lines which are extended. This 
can be explained that the evaluation has a correct part and a noise-term. 
The expected value of the noise-term is in variations where the programm 
extends the lines higher. This is alongside the results of Don Beal that a 
5-ply searcher with a purely random eval beats a 1-ply searchere 200:0.
There are indeed some decisions which are much deeper than current searches. 
Pawn-structures and especially passed pawns are an example. But even if a 
shallow searcher gets some positional advantage the deeper one usually 
starts do defend quite well and its difficult to win for smaller searcher. 
The win must be "forced". There are no reasonable alternatives to the 
winning moves.

Jan Lowman knew of course that a deeper search is beneficial. But especially 
the shape of the search has in a 5 ply search not so much influence. 
Especially the statistic based methods have not so much effect. I think the 
problem in Go is not, that search does not help. The problem is that its 
currently to shallow that one can really play around with it. Additionally 
there is the problem that the of moves are deeper.

Chrilly