[Computer-go] Thoughts about bugs and scalability

Don Dailey dailey.don at gmail.com
Mon Jun 20 08:27:35 PDT 2011


On Mon, Jun 20, 2011 at 10:09 AM, terry mcintyre <terrymcintyre at yahoo.com>wrote:

> Any particular instance of a program will probably fail to scale -
> especially against humans who share the lessons of experience.
>

That is complete nonsense.   How are you backing this up?   What proof do
you have that computers don't play better on better hardware?   Why are the
top programs being run on clusters and multi-core computers?   Are the
authors just complete idiots?

Every bit of evidence we have says they are scaling very well against
humans.     That has also been our experience in game after game,  not just
in Go.

I apologize for being so harsh on this,  but you are too smart to be saying
such dumb things.



> Among a bunch of high-dan players, several will become expert at defeating
> the current generation of programs.
>

That statement is out of context here.   Yes,  it's true but it doesn't have
anything to do with scalability.


> However, programs are not static; programmers also share the lessons of
> experience, and develop newer, smarter, and more scalable programs.
>

That is true,  but like your previous statement doesn't seem to make a point
in the context of a scalability discussion.   Are you just saying that if
the program is smarter it will scale better?    Yes,  I agree with that
completely.


>
> It's an arms race, and the race is not to those with the fastest CPUs, but
> to the smartest programmers, programs, and players.
>

Saying that is not about hardware but about smart software is a tired and
worn out cliche that has only has meaning when applied in the proper
context.    If you have a weak and stupid program your statement is
painfully obvious,  but in the real world of competitive programs it means
everything.

In chess it turns out that software and hardware have been equal
contributers to performance.    People used to believe that programs have
not advanced and that it's all about the incredible hardware advance of the
last 40 years but nothing could be farther from the truth.    Programs only
10 years old pretty much suck compared to our best programs of today,
regardless of hardware.    In 10 years we will laugh at todays go programs.
   They will be a lot smarter and they will run on much better hardware and
they will be many stones stronger than today.    I suspect that the software
will be a much bigger contributer to this than in chess but (assuming
"Moores law" continues) the hardware contribution will be impressive too.
  And I think there is synergy here too,   better software will greatly
improve the search,  making additional hardware look even better.

I think this is where everyone goes "stupid."   We should not be separating
hardware and software in our minds,  they are both critical and synergistic
(I hate that word but it works here.)      One of the big advances of
computer chess was realizing that you could simulate faster hardware by
making the software much more efficient.   So the search was engineered so
that a million stupid things that take time unnecessarily were removed from
the programs.    We got hash tables (it's stupid to do work when it was done
before),  null move pruning (it's stupid to search when your position is
hopeless),   and many various ways to (favorably) trade some risk for extra
depth.    And many other much smarter things too.     And then we figured
out what the search was not very good at and what it was good at and thus
improved the evaluation function in ways that mattered.   The result is that
a modern program can beat a 15 year old program with 100 to 1 hardware
advantage.      That is going to happen in GO too,  perhaps not quite as
dramatically but let's wait and see.

Perhaps we should have a match with the current best go program and the
first publicly available Mogo, a program that was then incredibly
impressive?     What do you think would happen?


Don






>
> Terry McIntyre <terrymcintyre at yahoo.com>
>
> Unix/Linux Systems Administration
> Taking time to do it right saves having to do it twice.
>
> _______________________________________________
> Computer-go mailing list
> Computer-go at dvandva.org
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20110620/5842cb52/attachment.html>


More information about the Computer-go mailing list