How long will it be until a computer system reach pro level play ?
(answering Bob Hearn
question)
Maybe that rather than taking the raw speed of hardware as a reference, we
could use the raw number of simulation (per second) as a base of speculation.
Assuming it's a fixed game time with the same setting than the 9 stone game
that was played.
+ So first question is : how much more 1.7 millions simulations can yield in
strength than what it does in today's mogo version ? Mogo has been in a very
intense state of development lately. I have no figures about the scaling of the
efficiency per simulation, but it has to be taken into account for trying to
guess what it will all look likes in ten years.
+ the second question is : how much more will the algorithm be tuned to the
hardware (or the hardware to the soft). I guess the theoretical throughput of
the system mogo was run onto is MUCH more than 1.7 millions simulations per
second (taking into account all the tree logic). Given the exact same property
of the simulation that were used, i'd think that in 10 years times, given
nothing more to do ... it could easily reach 17 millions simulations per second
on the same system.
+ the third question is : will moore law olds ?
So far moore law was linked to mono-threading paradigm. It may be true that
super-computer didn't improved by much (i don't know about that to much). But
there are also evidences that with the explosion and democratisation of
multiprocessors, the moore law will not held in it's current form. (GPU card
have had there efficiency increased but much more than x2 every two years)
+ the fourth question being : will there be much more efficient go-solving
method birthing in the next few years. And also, will not mogo get access to
more and more powerfull hardware as it'll close up to pro level ... (lately the
top raw computing power accessible to mogo program seems to have increased by a
lot ... was it running on a monocore 1ghz pentium 3 years ago ?)
It's true that ten years is a short span of time indeed. It may seems a bit
optimistic to hope for kim to fall in a fair even-game given 1day per move
seting in ten years for now. But i wouldn't call that totally irealistic
either. I guess i would easily put a bet on it, if the odds were about 1/20, me
getting 16 times the amount of money i have bet if a go-program reaches
pro-level on 19x19 toward the end of the year 2018. I'd probably go as far as
1/3 for the end of the year 2028.
-- Another interesting question is : how much time before mogo can tackle the
three-stone handicap game against a pro ? I have read somewhere that asked how
much stones he would need for a sure win against god himself if he's life was
at stake, the pro answered : three stones.
Bob Hearn wrote :
So -- quick back-of-the-envelope calculation, tell me where I am
wrong. 63% win rate = about half a stone advantage in go. So we need
4x processing power to increase by a stone. At the current rate of
Moore's law, that's about 4 years. Kim estimated that the game with
MoGo would be hard at 8 stones. That suggests that in 32 years a
supercomputer comparable to the one that played in this match would be
as strong as Kim.
This calculation is optimistic in assuming that you can meaningfully
scale the 63% win rate indefinitely, especially when measuring
strength against other opponents, and not a weaker version of itself.
It's also pessimistic in assuming there will be no improvement in the
Monte Carlo technique.
But still, 32 years seems like a surprisingly long time, much longer
than the 10 years that seems intuitively reasonable. Naively, it would
seem that improvements in the Monte Carlo algorithms could gain some
small number of stones in strength for fixed computation, but that
would just shrink the 32 years by maybe a decade.
How do others feel about this?
I guess I should also go on record as believing that if it really does
take 32 years, we *will* have general-purpose AI before then.
_________________________________________________________________
Pendant tout l'été, consultez vos emails Hotmail sur votre mobile !
http://www.messengersurvotremobile.com/?d=hotmail
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/