[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [computer-go] A chess programmer at the Go-Olympiad
I think it's obvious the branching factor is not necessarily relevant when
it comes to outperforming humans at some game (or task). I don't know
Arimaa, so I can't comment on that game, but I can think of many examples
where this would be the case.
In Go the branching factor is only part of why it's so hard for computers,
but certainly not the main thing. I've been saying for over a decade that
the hardest thing for the computer is evaluating the position on the Go
board. Which is partly shown by the fact, as David pointed out earlier,
that a decent evaluation can't be made without doing local searches. But
those local searches need some form of evaluation too. Ideally, those
evaluations are of the same quality of the top level evaluation so that
you don't lose some important effect on the rest of the game, but
obviously this leads to a chicken-and-egg problem.
Hence my suggestion earlier of annotating the search result with as much
useful extra information you can gather during the search. And have the
top-level evaluation use the result and the extra information somehow.
There's also still a great deal to be gained in board-analysis. I haven't
been following this mailing-list until recently again, but it always
strikes me that not much attention is given to new ways of gathering
information about the position in a static way. A few years ago I posted
an idea that I worked out about the concept of 'inside influence' and
'outside influence' (also called 'territory' opposed to 'influence'). I
found a very elegant algorithm to support the idea too, but there seemed
to be zero interest for it. Where I think it was one of the most exciting
things because it suddenly makes visible the difference between real
territory and a 'moyo', or potential territory.
And it's not as if we don't have enough computing power to think about
some extensive analysis algorithms. Everybody seems to agree that so far
computers don't use the extra power given. So I'd say: we leave the search
algorithms to Chess programmers. They're with larger numbers and have more
to gain from them. We then thank them for the work and steal their ideas
and we (with 'we' I means Go programmers) spend a good part of our time on
new ideas. Really new ideas.
>
> It's interesting to me that Arimaa has a very much larger branching
> factor than GO and yet we have a small but not insignificant chance of
> beating the human champion and winning the $10,000. Branching factor
> is often cited as being one of the reasons go is hard, but there is
> clearly more to it than that. Of course the human world champion at
> such a new game cannot be compared to the world go champion.
>
> - Don
>
>
>
>
> X-Sender: fotland%smart-games.com@xxxxxxxxxxxxxxxxx
> Date: Sun, 07 Dec 2003 18:28:19 -0800
> From: David Fotland <fotland@xxxxxxxxxxxxxxxxx>
> Cc: computer-go@xxxxxxxxxxxxxxxxx, computer-go@xxxxxxxxxxxxxxxxx
> Content-Type: text/plain; charset="us-ascii"; format=flowed
> X-BeenThere: computer-go@xxxxxxxxxxxxxxxxx
> X-Mailman-Version: 2.1.2
> Precedence: list
> Reply-To: computer-go <computer-go@xxxxxxxxxxxxxxxxx>
> List-Id: computer-go <computer-go.computer-go.org>
> List-Unsubscribe:
> <http://computer-go.org/mailman/listinfo/computer-go>,
> <mailto:computer-go-request@xxxxxxxxxxxxxxxxx?subject=unsubscribe>
> List-Archive: <http://computer-go.org/pipermail/computer-go>
> List-Post: <mailto:computer-go@xxxxxxxxxxxxxxxxx>
> List-Help: <mailto:computer-go-request@xxxxxxxxxxxxxxxxx?subject=help>
> List-Subscribe: <http://computer-go.org/mailman/listinfo/computer-go>,
> <mailto:computer-go-request@xxxxxxxxxxxxxxxxx?subject=subscribe>
> Sender: computer-go-bounces@xxxxxxxxxxxxxxxxx
>
>
> I wasn't referring to you, Don :) I know you understand the difficulty
> of
> computer Go. I was responding to the string where someone said that
> Go programs don't search.
>
> If my Arimaa program is stronger than yours, it's just because I've put
> a lot
> more time into it. I meant no disrespect to your abilities :)
>
> Regards,
>
> David
>
> At 08:13 PM 12/7/2003 -0500, Don Dailey wrote:
>
> > > I've seen many comments from computer chess programmers that
> imply
> > > that go programmers are just clueless and haven't bothered to
> > > understand or apply the lessons from computer chess. This is
> > > simply not true :)
> >
> >The lessons from computer chess is that most of us are clueless. Even
> >the smartest people were saying the stupidest things and we can laugh
> >at them now. I don't think most of us have learned from this and I
> >include myself too. The lessons I talk about have nothing to do with
> >specific programming techniques. It has to do with glib statements
> >about how wonderful and deep and difficult something is and will
> >always be.
> >
> >Now if I thought you were really clueless I wouldn't tell yo
> u, I would
> >show you by writing a much better GO program. But as you know, I
> >haven't done this.
> >
> >
> >- Don
> >
> >
> _______________________________________________
> computer-go mailing list
> computer-go@xxxxxxxxxxxxxxxxx
> http://computer-go.org/mailman/listinfo/computer-go
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://computer-go.org/mailman/listinfo/computer-go