In article , Andrew
Walkingshaw writes

In article , Neil Fernandez wrote:
In article , JVT
writes

The best available software do
not perform any better (and often worse) on 9x9 than on 19x19 boards.

So is it basically the case that the best available go program, given
the most processing power it has ever run on, would lose every single
game against a weak club-level human opponent on a 9x9 board?

I'd expect to win in excess of 80% on 9x9 against any of the top PC
programs on 9x9, assuming reasonable komi[1].

Any komi issue could be got rid of if you counted games in pairs, one
with each colour, and took mean difference in scores.

For what it's worth, I'm a British 3 kyu, which isn't *weak* - in chess
terms, somewhere between 1500 and 1700 ELO, I'd speculate

20% against an opponent suggests a rating 240 points lower, i.e. between
1260 and 1460 ELO. I wonder how you'd do if a program was given a lot
more processing power? :-)

Neil

(posted to all three groups, the comparison being the topic of the
thread :-) )

--
Neil Fernandez

In article , Neil Fernandez
writes

[snip]

Here is an article that might be found of interest with regard to this:

http://www.guardian.co.uk/uk_news/st...987745,00.html

(Why they used Mockney as one of the accents/dialects, I don't know -
couldn't they find anyone who spoke real Cockney? :-) )


..================================================ ===========

Brain buzz that proves Chinese is harder to learn than English

Tim Radford, science editor
Monday June 30, 2003
The Guardian

It's official: Chinese is more difficult than English.
The Chinese need both sides of the brain to grapple with challenges of
Mandarin, but English speakers listen with only half their minds on the
job.

Sophie Scott, a psychologist at the Wellcome Trust, and colleagues from
hospitals in Oxford and London performed brain scans on volunteers as
they listened to their native languages.

When English speakers heard the sound of Mockney, Mersey or Geordie,
their left temporal lobes lit up on screen. When Mandarin Chinese
speakers heard their native tongue, there was a buzz of action in both
the right and left temporal lobes.

"We were very surprised to discover that people who speak different
sorts of languages use their brains to decode speech in different ways,
said Dr Scott. "It overturned some long-held theories."

The left temporal lobe is normally associated with piecing sounds
together into words; the right with processing melody and intonation.

In Mandarin, a different intonation delivers a different meaning: the
syllable "ma", for instance, can mean mother, scold, horse or hemp
according to its musical sound.

"Speech really is a complex sound," said Dr Scott. "As well as
understanding words, the brain uses the way in which words are spoken,
such as intonation and melody, to turn spoken language into meaning.
This system has to be robust and flexible enough to deal with variations
in speech sounds such as regional accents. We think Mandarin speakers
interpret intonation and melody in the right temporal lobe to give
correct meaning to the spoken words."

The research throws new light on how speech is understood, and could one
day lead to new treatments for people whose understanding of language
has been impaired by stroke.

Dr Scott and her colleagues will put their research on show at the Royal
Society summer science exhibition in London tomorrow, Wednesday and
Thursday. The study suggests that language itself might affect the way
the brain develops in a young child.

It could explain why native speakers of English find it so
extraordinarily hard to learn Mandarin. It might also help doctors
understand what happens when people have to learn speech comprehension
all over again, for instance after a stroke, or after being fitted with
a cochlear implant hearing aid.

..================================================ ===========

Neil

--
Neil Fernandez

Neil Fernandez writes:

In article gers.com,
james writes

For computer,

Grand master (chess), street-corner player (chinese chess, Shogi),

I'd be surprised (and very interested!) if that were the case for
Chinese chess, at least when the best programs are run with the same
processing power at which chess programs play at GM level.

I'm a lot weaker at Chinese chess than I am at chess (I've spent much
less time at it), but it seems to me to be a less complicated game than
chess. It's also more 'tactical' - pawns capture as they move and
therefore can't block each other, and the game is much more open.

beginner (Go).

Chinese Chess is on 9 by 10 board

True but apart from pawns, only 4 out of each player's 9 pieces can
cross the river. Of the other 5, 2 can reach only 7 squares, and 3 are
restricted to a 3x3 area, in which 2 of them can reach only 5 squares.

Minor correction ... you forgot about the cannons. Besides the pawns,
6 pieces can cross the river (2 each of chariots, horses, and cannons).

--Yu

james wrote:
For computer,

Grand master (chess), street-corner player (chinese chess, Shogi), beginner
(Go).

Chinese Chess is on 9 by 10 board and Shogi on 9 by 9 board. Both can be
solved the chess way - (I think)

Chinese Chess and Korean Chess can both be "solved" using the same
methods as Chess, but Shogi is a slightly different matter as I
understand. Because in Shogi you can place pieces on the board instead
of moving (I have never learned the game but this is my understanding)
the size of the tree increases dramatically as pieces become available.
The whole problem with Go is that the sheer size of the search tree
makes it near impossible to program strongly using the same methods.
That being the case, Shogi would have the same problem to a lesser degree.

The size of the board makes little difference, it is the size of the
search tree that matters. How many different moves can one make from a
given position? This is the question that's answer tells you how wide a
search tree gets. With Go the size of the board makes a difference
because the amount of available moves is based on the size of the board.
The reason why you might not see much difference though is because at
9x9 the exponent is already more than twice that as chess (81 vs. ~35).
Go engines have to use some methods to cut that down, I don't know
these methods but I bet they cut down to the same level no matter what
the size is and will have something to do with the patters on the board.

The only reason why Chinese Chess has not reached the strength of Chess
engines is that evaluation methods have not been so widely researched.
Chess has been a large research topic since the beginning of computers,
Chinese Chess is a relatively infantile subject. I myself have written,
and am currently working on, Chinese Chess engines and there is nothing
stopping them from getting as strong as Chess engines. Move generation
is slightly more complicated (or maybe just different) but doesn't add
that much overhead if done correctly. Evaluation function just haven't
evolved enough.

But that could just be my perspective. The strongest OS XiangQi engine
is written in Chinese C, I know a little Chinese but not enough to be
able to adiquately study the source code.

For those truely interested in this subject you can go to citeseer (
http://citeseer.nj.nec.com/cs ) and do some searches. All of the above
games are widely talked about in various papers at that site. Go and
Chess are by far the hottest topics but Shogi shows up a lot and Chinese
Chess a little. Since Chinese Chess can mostly use the same methods as
Chess it is talked about a lot less than Go and Shogi. Eventually my
engine project's website ( http://xiangqi-engine.sourceforge.net/ ) will
have links to helpful papers and sites.

NR

In rec.games.go Simon Waters wrote:
Humans don't like the idea of computers being "better" than they are, so
we tend to judge them by the things we are good at, playing Go,
translating poetry, and talking. Rather than the things they are good
at, like chess, maths, logic, data handling.

Really? I've never noticed that computers are any good at maths. I mean,
remind me again who came up with the proof for Fermat's Last Theorem?
Sure, computers are good at arithmetic, but you need a human
mathematician to tell it what to calculate...

I'd say probable, as far as we can ascertain massively parallel
computers have exceeded the processing and storage capacities of the
human brain, so all we need is the software for whatever task is at
hand, be it Go, or consciousness (whatever that may be).

How do you propose that we measure the processing and storage capacities
of the human brain, so that we can verify your claim?

I suspect the primary reason that computers beat the world at Chess and
not Go, are cultural, the west followed this route in AI and used chess
as its guinea pig, the east seemed more interested in robots and ended
up with large amounts of the worlds car manufacturing industry. If
Shannon and Turing had been more into Go than Chess, I wonder what
modern computers would be like?

I suspect that the quoted paragraph is bull****. But if you have some
facts to back up your claims, I'm sure lots of people reading this
newsgroup will be interested.

Regards,
Michael

--
You have had a long-term stimulation relative to business.

Sometimes I think we all forget that the only way a computer is better
than a human is that it is able to process the information faster .
It "the computer" is never smarter than the human !

On 11 Jul 2003 20:16:16 GMT, Michael Goetze
wrote:

In rec.games.go Simon Waters wrote:
Humans don't like the idea of computers being "better" than they are, so
we tend to judge them by the things we are good at, playing Go,
translating poetry, and talking. Rather than the things they are good
at, like chess, maths, logic, data handling.

Really? I've never noticed that computers are any good at maths. I mean,
remind me again who came up with the proof for Fermat's Last Theorem?
Sure, computers are good at arithmetic, but you need a human
mathematician to tell it what to calculate...

I'd say probable, as far as we can ascertain massively parallel
computers have exceeded the processing and storage capacities of the
human brain, so all we need is the software for whatever task is at
hand, be it Go, or consciousness (whatever that may be).

How do you propose that we measure the processing and storage capacities
of the human brain, so that we can verify your claim?

I suspect the primary reason that computers beat the world at Chess and
not Go, are cultural, the west followed this route in AI and used chess
as its guinea pig, the east seemed more interested in robots and ended
up with large amounts of the worlds car manufacturing industry. If
Shannon and Turing had been more into Go than Chess, I wonder what
modern computers would be like?

I suspect that the quoted paragraph is bull****. But if you have some
facts to back up your claims, I'm sure lots of people reading this
newsgroup will be interested.

Regards,
Michael

In article , Noah Roberts
writes

The size of the board makes little difference, it is the size of the
search tree that matters. How many different moves can one make from a
given position? This is the question that's answer tells you how wide a
search tree gets. With Go the size of the board makes a difference
because the amount of available moves is based on the size of the board.
The reason why you might not see much difference though is because at
9x9 the exponent is already more than twice that as chess (81 vs. ~35).

This idea could be tested, if someone were to transform go and/or chess
(better to use the international game if the Chinese game has had a lot
less research into evaluation functions), to make them 'equivalent' in
terms of tree width, then give the respective programs the same amount
of processing power and see how well they perform. One could use, for
example, a 10x10 version of Capablanca's chess variant.

Neil

--
Neil Fernandez

In article s.co.uk,
Chris Lawrence writes

On Fri, 11 Jul 2003, Neil Fernandez wrote:

In addition it would be wrong to assume that there is no game with a
clearly defined and finite state space at which no computer programs
will be _able_ to play as strongly as the strongest humans can, even
with lots of processing power and parallel processing.

I've never claimed it is not possible or that it will never be possible.
Merely that it is not possible today and processing power is not the
stumbling block.

My own view is that there are or will be games at which humans will
always be able to thrash computer programs (or would be able to do so if
such programs got written) - call me an optimist! :-)

When we play Go we are using a lot of gut-feeling, intuition, visual
effect, experience, reading, etc. Some of those things can be
represented or effected in today's computers. Others cannot.

We need to a) understand what happens when we think about things, b)
create a computer with a more suitable architecture on which to map
those processes and c) understand the thoughts which take place in Go
and map those onto b).

....if one shares the aim of programming computers to beat humans.

Neil

--
Neil Fernandez

In article , Simon Waters
writes

Justin O. Wyss-Gallifent wrote:
In rec.games.go Chris Lawrence wrote:

[snip]

It is of course an irrational response, as we are nothing like
computers, so trying to draw a comparison is pointless. A program either
is or isn't better at a specific task than a specific human, assuming
that that task requires

....or does not require...

special human like powers, unless you have
evidence to that effect, other than no one has figured out how best to
program it, is also irrational.

--
Neil Fernandez

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Michael Goetze wrote:
In rec.games.go Simon Waters wrote:

Humans don't like the idea of computers being "better" than they are, so
we tend to judge them by the things we are good at, playing Go,
translating poetry, and talking. Rather than the things they are good
at, like chess, maths, logic, data handling.


Really? I've never noticed that computers are any good at maths. I mean,
remind me again who came up with the proof for Fermat's Last Theorem?

Who proved "Robbins Conjecture"?

Sure, computers are good at arithmetic,

.... and symbolic computation including differentiation, integration,
simplification and reduction of algebraic expressions, propositional
logic ....

but you need a human
mathematician to tell it what to calculate...

I never said you didn't, although I'm not convinced on that point either
as I believe there are already examples of computer systems that look
for interesting results autonomously. I suspect the problem is not
coming up with proofs, but determining which are going to be interesting
to the humans.

I'd say probable, as far as we can ascertain massively parallel
computers have exceeded the processing and storage capacities of the
human brain, so all we need is the software for whatever task is at
hand, be it Go, or consciousness (whatever that may be).


How do you propose that we measure the processing and storage capacities
of the human brain, so that we can verify your claim?

"as far as we can ascertain"! We can certainly measure how well basic
factual information is stored and retrieved and how fast. Total capacity
is harder to assess but we can compare number of configurable elements,
rates of operation, and energy consumption, amongst other things.

Whilst I think the brain is an amazing organ when we get down to raw
computation, it costs, and those costs can be measured, and there is
only so much you can do on a bowl of breakfast cereal.

The assumption is usually worked from 10^13 neurons, thousands of
interconnects, with a frequency of 200Hz, and you end up with the
equivalent of a few hundred low end PC processors. However such
assumptions have a lot of hidden assumptions about how brains work, and
how much is available to handle a particular task. To even do ordinary
day to day stuff the brain must be making very efficient use of its
limited resources, however it doesn't alter the fact those resources are
severely constrained by having to be fitted into an oversized chimpanzee.

Don't know about you but I'd be very surprised if my brains storage
capacity exceeded that of Google, or even came close.
-----BEGIN PGP SIGNATURE-----
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQE/D1WgGFXfHI9FVgYRAhkwAKCB7nS+xfsb9f+aGuPzddaoLVeGBQ CeKXvT
XP9+zY9FYcA3v3u3KfXEgwo=
=YkIr
-----END PGP SIGNATURE-----