I don't know what its worth, but I bought the game and gave it a quick try
today.. Here're the thoughts..
- The music in the game repeats after 10-20 seconds. Don't know if there
are multiple sound tracks, but in a single game, only one looping music
sample is playing, which gets very repetitive starting already from the
first game. I felt very annoyed already after 5 minute and had to turn off
the sounds.
- The story mode felt interesting, giving tsumego puzzles in between
in-game opponents. The problem is that it starts in the easiest settings
mode (the very first tutorials are skippable, although when you do that,
you seem to lose a part of the story), and you can't choose a story mode
difficulty, which made it very easy and boring to play (my official level
is 9 kyu in the finnish OGP club). In the first game in story mode, the
opponent gives you 3 stones on 9x9 board as handicap, and the result was
81-0 for me (apparently chinese scoring). In that game, the opponent
passed in mid-game (I thought one of his two groups was livable if it had
played at that moment).
- The second game in story mode was 9x9 with 2 stones handicap, which was
enough to force me abandon the whole story mode altogether. Perhaps I'll
retry doing the story mode and just grind through those easy difficulty
levels to see if there are decently difficult tsumego problems coming up.
- After the second game, I started a new free game against the CPU on
19x19 at the most difficult level. I occupied one corner, white occupied a
second one, followed by me occupying a third one, after which the opponent
approached my corner without taking the fourth one (I guess nothing bad
there, a matter of style). We played the first corner according to a
standard komoku joseki, which was ok. In the second corner I made a
beginner's mistake (I thought I was playing a joseki), and the CPU
properly punished.
- The third corner however, was a catastrophy for the opponent. I ended
the joseki and tenukied in a situation where one of my stones was
catchable in a ladder, although the ladder favored me, since I had a hoshi
stone at the opposing corner. The bot didn't realize this, but we played
the whole board through the ladder, and only at the very end it realized
the capture is not going to happen (the ladder spawned the whole board
from bottom left to top right), and it tried to desperately back up its
moves. After shattering the opponent, I captured the AI stones for a
while, and eventually just closed the game, since it got boring. It seems
like the AI spends a constant amount of time per move (no time control,
and it never immediately makes a move e.g. even if it's part of an opening
book).
I oughta try it again on 9x9 even start (although I expect to completely
shatter it). It makes me wonder, why didn't David Fotland make the AI for
the game, doesn't he work for Microsoft after all? :)
To find something positive, I think the game will be a success in terms of
trying to get new amateur players into the game, but the AI is not
competitive by any means.
On Mon, 20 Dec 2010 04:17:31 +0200, [email protected]
<[email protected]> wrote:
not so hard to beat but feels quite natural.
it is quite good at finding good spot, abusing weakness and protecting
its own weakness.
but it does fatal mistakes.
on a 9x9 it has 5 level of difficulty we can choose from. only 3
levels are available for larger board.
chinese scoring is a pain IMHO
UI is nice
5$ is cheap
we shall make it play against many face or other strong AI to compare.
we could setup a game through KGS, I could manually play the moves
from the game.
another website you might want to take a look at is
http://research.microsoft.com/en-us/projects/go/
I would not be surprise that some of the AI researcher are following
this list
On Sun, Dec 19, 2010 at 5:17 PM, Petr Baudis <[email protected]> wrote:
Hi!
There's been some talk about the XBox Go game from Microsoft Research:
http://research.microsoft.com/en-us/news/features/thepathofgo-121510.aspx
Anyone has more details on how strong is its AI, etc.? The article above
implies it's a MCTS + patterns engine, perhaps using the pattern matcher
as described in "Bayesian pattern ranking for move prediction in the
game of Go"?
Petr "Pasky" Baudis
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
