Re: Humans no match for Go bot overlords

- wrote:
<remailer@xxxxxxxxxxxx> writes:
[ ... ]
http://blog.wired.com/wiredscience/2009/03/gobrain.html

Nick Wedd <nick@xxxxxxxxxxxxx> wrote:
What a load of cobblers.


Is "cobblers" now some sort of arcane technical term ?

no, it's the same as saying -
it's a load of crap
it's a load of bull***
it's a load of twaddle




"artificial intelligence is asserting its domination over gray matter." It isn't.


Agreed. I thought we were talking about a Go Game,
not "gray matter." -( BBC _Space_Cops_ 1987 )-



"Artificial intelligence researchers routinely said that computers capable of beating our best were literally unthinkable."
I doubt any of them ever said anything so stupid.


So, according to the world of Nick Wedd, who are they?



"a program called MoGo beat two professionals."
Yes, on a seven-stone handicap.


Which is still better than an 8 or 9 stone handicap.



"People hoped that if we had a strong Go program, it would teach us how our minds work."
I doubt anyone ever hoped anything so implausible.


The article, which perhaps you did not read carefully, states "Bob Hearn, a Dartmouth College artificial intelligence programmer" said it (devotee of Minsky).
http://www.dartmouth.edu/%7Erah/
Says that he "violently disagrees" with Roger Penrose.



"the Monte Carlo method, named by its Manhattan Project pioneers for the casinos where they gambled."
I do not believe the Manhattan Project team members were travelling to Axis-controlled Monte Carlo to gamble there in 1943.


Good point. Missing from the discussion is the possibility
that discovery of good moves is also related to random walk.



"Hearn and others say that, having started to beat human professionals, Monte Carlo-based programs will only get better."
This gives the impression that programs can already beat human professionals. They can't, by a long way.


Agreed. The article can be terribly misleading, especially
for those who are easily misled. And "conquering" would also
be a misleading term, because Kasparov traded wins & losses
with Deepest Purple, a trend even more evident perhaps in Go.
The win percentage over multiple games must be at least 50% even to qualify for consideration, yet the term "conquer" seems to imply a much higher win reliability than that.

( referenced at
http://www.newscientist.com/blogs/shortsharpscience/2009/02/computers-get-go.html

)
This one also stated:

"Because of the asymmetries in information, the game may be what mathematicians call an undecidable game, unsolvable by any algorithm," Hearn said.

"There is no computer, in principle, that can even tell us from a given position who will win," Hearn said.


Which is false, because there are -some- given positions
which can be more or less scored by computer, assuming
that the players play optimally.

Perhaps someday the computers will write better articles that
persuade Nick Wedd of their good intentions and accuracy.



- regards
- jb

--------------------------------------------------------
Computers Effective In Verifying Mathematical Proofs
http://www.sciencedaily.com/releases/2008/11/081106153638.htm

Quantum Computers May Be Easier To Build Than Predicted
http://www.sciencedaily.com/releases/2005/03/050309120855.htm

Problem of the Week ( Purdue University )
http://www.math.purdue.edu//pow/

5 dim A%(900):' Purdue Problem of the Week 1/n = 1/x + 1/y pow.ub
10 for N=2 to 2000:K=0
20 for I=1 to N*N
30 K+=(den((N*I)//(I+N))=1):' fast count of integer results
40 next: ' here could just print "k" for OEIS # A063647
50 inc A%(K):for I=1 to 30:print str(A%(I));:next:print:' distribution
60 next:' http://www.research.att.com/~njas/sequences/A063647

http://en.wikipedia.org/wiki/Riemann_Hypothesis
http://en.wikipedia.org/wiki/Talk:Riemann_hypothesis
--------------------------------------------------------

.

Loading