Why Neural Networks Look Set to Thrash the Best Human Go Players for the First Time

[Hans-Georg Michna]

One of the last bastions of human mastery over computers is
about to fall to the relentless onslaught of machine learning
algorithms.

Computers are rapidly beginning to outperform humans in more or
less every area of endeavor. For example, machine vision experts
recently unveiled an algorithm that outperforms humans in face
recognition. Similar algorithms are beginning to match humans at
object recognition too. And human chess players long ago gave up
the fight to beat computers.

But there is one area where humans still triumph. That is in
playing the ancient Chinese game of Go. Computers have never
mastered this game. The best algorithms only achieve the skill
level of a very strong amateur player which the best human
players easily outperform.

That looks set to change thanks to the work of Christopher Clark
and Amos Storkey at the University of Edinburgh in Scotland.
These guys have applied the same machine learning techniques
that have transformed face recognition algorithms to the problem
of finding the next move in a game of Go. And the results leave
little hope that humans will continue to dominate this game. ...

Read the complete article at:
http://www.technologyreview.com/view/533496/why-neural-networks-look-set-to-thrash-the-best-human-go-players-for-the-first-time/

Hans-Georg

[Detlef Müller]

Post by Hans-Georg Michna
One of the last bastions of human mastery over computers is
about to fall to the relentless onslaught of machine learning
algorithms.
[...]

sounds a bit exaggerated :)

Post by Hans-Georg Michna
http://www.technologyreview.com/view/533496/why-neural-networks-look-set-to-thrash-the-best-human-go-players-for-the-first-time/

I suggest looking under http://arxiv.org/abs/1412.3409
which is (closer to) the original work and further from sensational
press (but a longer text, I didn't read all of it carefully).

Quiet impressing that "braindead" picking the most profi-like looking
move for each occuring position (without any reading at all) leads to
winning games even against lower ranked go-programs.

I'd be surprised if this would not be combined with
MC-Algorithms pretty soon.

Detlef

[Hal Womack 3-dan]

otlichna Dec 16, 2014
Even by the usual low standards of this blog, this article is a disaster. As other commenters have pointed out, 1 kyu is not an expert go rank, it is a high beginner ranking. It is not at all uncommon for talented humans to reach a rank of 1 kyu in only a year or two of study. GnuGo is a significantly outdated and very weak program only used for training beginners, and Fuego 1.1 is hardly "state of the art," considering it was created in 2011 and is nowhere near the top of the Go computer rankings anymore. Cursory googling could have turned up all of this information.



Top level computer Go programs can achieve a rank of around 3 dan running on good hardware. This is strong enough to defeat Fuego 1.1 in 100% of games, even with a handicap, and far stronger than the absurd neural network program described in this article. However, 3 dan is only a strong amateur ranking. Even the weakest of human professionals could routinely crush such a program, even at substantial handicaps..



If it weren't about such a trivial topic, I'd say this article was a violation of basic journalistic ethics. Seriously people, go read a science blog that isn't full of BS.
------------------------------

I second the above comment. If M.I.T. hopes to get past the 10-kyu level in journalism, perhaps it would interview at length the owners of the strongest bots now playing witch AFAIK ~5-dan on KGS?

==================

Post by Hans-Georg Michna
One of the last bastions of human mastery over computers is
about to fall to the relentless onslaught of machine learning
algorithms.
Computers are rapidly beginning to outperform humans in more or
less every area of endeavor. For example, machine vision experts
recently unveiled an algorithm that outperforms humans in face
recognition. Similar algorithms are beginning to match humans at
object recognition too. And human chess players long ago gave up
the fight to beat computers.
But there is one area where humans still triumph. That is in
playing the ancient Chinese game of Go. Computers have never
mastered this game. The best algorithms only achieve the skill
level of a very strong amateur player which the best human
players easily outperform.
That looks set to change thanks to the work of Christopher Clark
and Amos Storkey at the University of Edinburgh in Scotland.
These guys have applied the same machine learning techniques
that have transformed face recognition algorithms to the problem
of finding the next move in a game of Go. And the results leave
little hope that humans will continue to dominate this game. ...
http://www.technologyreview.com/view/533496/why-neural-networks-look-set-to-thrash-the-best-human-go-players-for-the-first-time/
Hans-Georg