[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: computer-go: Most simple Go rules




David Fotland wrote:
> My personal preference would be to have all tournaments use AGA rules

Hear, hear!

Now I wonder why you prefer rules very close to Tromp-Taylor only
to raise objections as follows:

> My only real objection is that I want computer programs to play the same game
> that people play, and people don't use Tromp/Taylor rules.

People using Japanese scoring don't use the Nihon Kiin 1989
Rules because they do not understand them,
people using Japanese scoring don't use the WAGC 1979 Rules
because they do not understand them,
people using Japanese scoring don't use the Nihon Kiin 1949
Rules or KBA 1992 Rules because they do not understand them,
people using Chinese scoring don't use the Ing 1991 Rules 
because they do not understand them.
Especially this refers to people who are officially supposed
to use one of the aforementioned rulesets.

Thus your argument is weak.

> Also, I don't like superko.

You like AGA Rules with superko but you don't like superko!

> First, it is unusable in human games since it is so much
> more difficult than other ko rules for people.

This is the most unreasonable myth I have ever seen!

Never has any game been longer than 425 plays. People are
capable of remembering their game with 425 plays.

All standard ko cases are fought the same way as they would
be fought using Universalists Ko Rules (a basic ko rule plus
no result in case of long cycles; this is like Japanese ko
rules except that they have a third ko rule for the purpose
of defining life and death around the game end).

The standard ko cases treated as usual are:
- a basic ko: it is a fighting ko
- a double ko coexistence: it is a disturbing life even at
the game end
- a double ko death: it is a disturbing death and is dissolved
at the game end
- a perpetual ko: it is a disturbing death and is dissolved
at the game end while one involved basic ko might be a
fighting ko
- two basic kos: one of them is fought as a fighting ko or
none of them is fought at all
- three basic kos: one of them is fought as a fighting ko or
none of them is fought at all
- one basic ko and a double ko death: the basic ko is fought
while only the dead player in the double ko has ko threats
in it
- one basic ko and a perpetual ko: the basic ko is fought
while only the dead player in the perpetual ko has ko threats
in it
- a dead ko: it is a disturbing death and is dissolved at the
game end
- a sending-two-returning-one: it is a disturbing ko and the
disturber has little interest in disturbing
- bent-4-in-the-corner until shortly before the game end: it
provides ko threats
- 10000-year-ko: it provides ko threats
- three-points-without-capturing until shortly before the 
game end: it provides ko threats

The only exceptions that make a small practical difference:
- bent-4-in-the-corner around the game end: Chinese scoring
and a balance of uneliminable ko threats favouring the dead
player result in a ko fight in contrast to Japanese scoring
- a basic ko around the game end if the only available,
remaining threats are dame: due to theoretically proven 
parity characteristics in case of komi 5 or 7 this 
changes the winner only every 10,000th game when comparing 
Chinese scoring with Japanese scoring and can be avoided 
most of the time if the players are aware of that possibility 
before only dame and that basic ko remain (+0.5 komi make no
change while 6 komi make a different winner more likely)

The most frequent of rare cases that are already so rare that
I have never experienced any of them in one of my games:
- suicide as a ko threat (I could have used it once if allowed)
- suicide in a semeai
- three-points-without-capturing around the game end: a
certain balance of ko threats might lead to a ko fight,
which also depends on the scoring method
- one basic ko and ko threats in a double ko coexistence
- a triple ko
- an eternal life
- a round-robin ko
- triple ko stones

***

Which percentage of human players would even be able to
compile such a list, i.e. to understand why there are any
practical differences between superko and, e.g., 
universalists? Maybe above 5 kyu quite some could point
out bent-4-in-the-corner as an exception, however, how many
would be able to explain correctly that, why, and how the 
exact difference becomes relevant around the game end only 
and not earlier during a game? I should rather ask: How few
players? I, e.g., did not understand the difference properly 
before having studied Japanese rules for more than 100 hours.
Bent-4-in-the-corner is an exception that can be justified
only by an exceptional rule designed particularly for it. In 
the past this was a precedental rule, now it is the 
Pass-for-ko-rule, which only less than 1% of all players
could explain you correctly. Using a hardly understood,
exceptional rule for the only and furthermore only scarcely 
relevant exceptional shape as a justification for maintaining
the three Japanese ko rules (or an incomplete system of 
dozens of precedents) is weaker than weak.

You call superko "so much more difficult than other ko rules 
for people". I do not see even the slightest justification.
It may not be well known why superko is almost the same as
a system of Japanese ko rules in practice. However, this does
not show that superko would be any more difficult than a 
system of Japanese ko rules - it only shows that superko is
so similar in practice that all players can adopt it easily
in practice without even worrying about reasons for that 
similarity.

> Second, it makes some
> traditional, settled,
> local shapes into fights.

See above. This is bent-4-in-the-corner only around the game
end and some positions rarer by a factor of more than 1000.

> This second reason is why Ing switched from
> superko to his more complex ko rule.

To use his own words: "One should not treat a disease by 
using the wrong medicine for treating its symptoms but by 
eliminating the disease's causes." Those exceptional
positions are the symptoms. The causes were the traditionally
exceptional ko rules.

Ing wanted ko rules with which the players could fight kos
by watching only changes of available ko threats instead of
changes of whole board positions. He overlooked that with 
superko in practice (except in extremely rare cases like 
triple ko stones - never reported in any game so far - or 
molasses ko - reported in only one game so far -, where 
temporarily even the so called ko stones disapppear) it is 
always sufficient for the players to watch changes of 
available ko threats.

All known arguments against superko do not maintain any 
justification if considered carefully. The only exception is
preferring traditional precedental rulings for the sake of
adoring traditional precedental rulings as such. Clearly,
computer programs' AI and history are not sufficiently 
developed for such a traditional culture!

> keeps a reasonable ko rule.

A reasonable ko rule is a ko rule that can be stated by all
players who use it. This is not the case for the 
Pass-for-ko-rule. Your reference to human players fails
gloriously:)

--
robert jasiek