Re: computer-go: Computer Go survey paper - request for comments

[Robert Jasiek <jasiek@xxxxxxxx>]

Martin Mueller wrote:
> I realize that there are still a number of shortcomings and omissions in my
> draft, so I'm asking you all for your feedback in order to improve it.

Comments:
- 1.1: Probably you omitted any complete rules
intentionally?
- 1.1: the ko rule: This is a bad term. The move
rules provide variation to the game while the
restriction rules (a theoretically more appropriate
term) restrict variation of the game. OTOH, using
the term ko rule simplifies matters for the reader
who is not so easily overwhelmed by other terms
including the word ko.
- 1.1: basic ko: nice term
- 1.1: two or three consecutive passes end the game:
Better omit "two or three" because it just restricts
details without reason. E.g. it might also be four
passes.
- 1.1: place several stones in a row: This is not
5-in-a-row:) Better speak of consecutive black moves.
- 2.3: related fields: You might add "[mathematical]
go rules theory".
- 2.3: To the literature you might add references to
go rules theory if you define those in the appendix.
- 2.4.2: It is very misleading to mention search space,
number of board positions, and estimates for that number
together and to use just a "simply speaking" for a
restriction. The search space [for a rule set] is the
complete game tree of all legal games. OTOH, you correctly
refer to repetition of positions (although it need not
necessarily be positional repetition) and to the complete
move history. If you want to give estimates, then IMO
this should be for numbers of games rather than numbers of
positions.
- 2.4.3: It should become clear but does not yet that a)
it is unknown how bad top humans are compared to perfect
play, b) so far computer go (except for some maths
studies) aims to approach the quality of top human play
only rather than perfect play. [The difference is, to be
modest, gigantic.]
- 3: Here you can most easily shorten chapters, if
necessary for adding more information to the entire text.
- 6: Solve Go on Small Boards: W. Taylor deserves a
reference to his RGG message solving 2x2 only last year.
- References: Since you list several mathematical or
mailing list articles, you might also add [1] R.
Jasiek, et al. Various messages on go-rules mailing
list, 1997-2000. [2] Countless messages on
news:rec.games.go. [3] Various messages on
news:de.rec.spiele.brett+karten, 1999-2000.

> Other topics that should probably get more coverage than now are:
> - ko fights

- 2.4.2: If you want to draw a connection to ko fights
here, then you might add a reference to the (I think) 
EXPTIME proof relying on ko play. You might also add a
reference (and then maybe also in the References as [4]
R. Jasiek, New Approach to Cycles, Ko Strings, Ko
Threats. Private correspondence to W. Spight, 1999.) to
my general ko string definition showing that it has to
depend on perfect play [under given rules] to make some
sense [i.e. to avoid that any string is a ko string].
- 5: You might add a short section about ko fights (a la
Berlekamp, Spight, et al).
- 5: Adding a go rules theory related section about ko
fights seems to be an overkill for the paper, except for
the next point.
- 5: I would strongly recommend to add a short section
(which I might write, if necessary) about the Fixed Ko
Rule (see reference [1] above or provide [5] 
http://home.snafu.de/jasiek/rules.html, also see
superko.html). It greatly simplifes research because all
ko play is cold (disturbing) during the entire game. The
middle game has no ko fights. Although the game (esp.
life and death) is slightly different, the model is
extremely useful for theoretical study, even if almost
no research is done yet.

> - problems with end of games, rules

What would you like to see here?

> - how to count the score

Trivial given logical rules.

--
robert jasiek
http://home.snafu.de/jasiek/