[FRIAM] Google self-evolving AlphaZero artificial intelligence program mastered chess from scratch in 4 hours: Rich Murray 2017.12.10

Mon Dec 11 23:41:20 EST 2017

John, 

Is one of these "exciting" games played through and commented on any engine
that mere mortals can access?  

n

Nicholas S. Thompson

Emeritus Professor of Psychology and Biology

Clark University

 <http://home.earthlink.net/~nickthompson/naturaldesigns/>
http://home.earthlink.net/~nickthompson/naturaldesigns/

From: Friam [mailto:friam-bounces at redfish.com] On Behalf Of John Kennison
Sent: Monday, December 11, 2017 2:23 PM
To: The Friday Morning Applied Complexity Coffee Group <friam at redfish.com>
Subject: Re: [FRIAM] Google self-evolving AlphaZero artificial intelligence
program mastered chess from scratch in 4 hours: Rich Murray 2017.12.10

I agree that it's not clear that AlphaZero would excel at the supergame.I
described. Still human intuition is probably not the indispensable
ingredient that it once might have seemed to be.

On the other hand, many chess commentators think that AlphaZero has a
playing style that is "more human" than the styles of other chess computers.
When humans play chess, we can often discern themes --maybe one player is
trying to breakthrough in the center while the other  is trying to
breakthrough on a flank. In contrast, most computers are simply trying to
maximize a position evaluation function. This only leads to a successful
breakthrough if the computer can see in advance that the breakthrough will
lead, in a relatively short time, to a measurable advantage, such as the
forced win of a pawn. 

Humans sometimes say that they need a plan -even a bad plan is said to be
better than playing without a plan. AlphaZero's games against the computer
Stockfish seem to pursue clear-cut plans (at least the games that have been
made available). It may be the case that having a plan leads to better play.
The point is that the plan changes the evaluation function --if you want to
breakthrough in the center, you try to post your pieces differently than if
you are planning to breakthrough in a flank. Having a plan, even a bad plan,
may lead to better coordination of your pieces --even for a computer.

Magnus Carlsen, the current human world chess champion, said that if you
play against a top computer you will surely lose but you will also be bored.
I think that you find a game you are watching interesting when you can sense
competing plans behind the moves. AlphaZero's games are quite exciting.

  _____  

From: Friam <friam-bounces at redfish.com <mailto:friam-bounces at redfish.com> >
on behalf of Russ Abbott <russ.abbott at gmail.com
<mailto:russ.abbott at gmail.com> >
Sent: Monday, December 11, 2017 12:36:53 PM
To: The Friday Morning Applied Complexity Coffee Group
Subject: Re: [FRIAM] Google self-evolving AlphaZero artificial intelligence
program mastered chess from scratch in 4 hours: Rich Murray 2017.12.10 

Not clear that AlphaZero would do well on John's SuperGame.  It won on chess
(and Go) by playing against itself in advance. If it doesn't have the
opportunity to do that it won't have that advantage. It's strategy would
have to be something like on-the-fly playing the selected game against
itself in the background at the same time as it is playing the human
opponent. The question then is how fast it can teach itself the new game.
It's strategy would have to be to slow down the game against the opponent as
much as possible to give itself time to learn the new game.  So it becomes a
matter of computer speed (for learning the new game) and the extent to which
the real game can be delayed as it is in progress. 

On Mon, Dec 11, 2017 at 8:58 AM Steven A Smith <sasmyth at swcp.com
<mailto:sasmyth at swcp.com> > wrote:

Marcus wrote:

Is a strategy anything more than a coarse-grained tactic?   And is intuition
anything more than an associative memory that connects coarse- and fine-
grained information? 

Is it any more?  Or any less?

Learning is an iterated game that operates at many scales and on many
dimensions... 

<TL;Don'tRead>

The Biological Evolution record shows myriad explosions in quantity and
diversity  that resulted from a (small? but) significant innovation (e.g.
multicellular organisms, Eukaryotes, photosynthesis, oxygen metabolisms,
vertebrae,  warm blooded metabolism, live birth, etc).   Punctuated
equilibrium?

There seem to be similar inflection points in the "learning" implied in
human social/technological/economic evolution and we may be in on the
shoulder of "yet another" which gestures in the direction of the von
Nuemann/Vinge/Kurzweillian "technological singularity".

I'm not much of a chess expert, myself, playing only *barely* competitively
in my late teens (as Spassky and Fischer were dukingit out), and revisiting
it in the pre-ALife era of "evolution, games, and learning" in the late 80s,
along with GO.  Chess itself, as a "playing field" for learning strategy is
a microcosm to observe the general idea of "learning".   The history of
chess is fascinating.  In the current context, it is fascinating that out of
about 1500 years of existence (in proto-forms), for a little over 500 of it,
the rules have settled on what we use today, but the tactics and strategies
developed *on top* of those has continued to  both *evolve* and *reflect*
society at large.  Most notably, perhaps, the "Romantic Period" where one of
the dominant ideas was that personal genius *and* style mattered more than
theory or logic or even board positions.   This somewhat reflected the
military and political style of that period.  During the "age of
Enlightenment" it also had a moral embedding...  The "modern" era emerged
with the industrial revolution and more importantly perhaps, the
mechanization of war where chess strategy, now somewhat more "scientific"
began to eventually give rise to "hypermodernism" which focus more on
controlling the center of the board from afar (a parallel to mechanized
warfare where power could be projected over a great distance in a short
amount of time).   Algorithmic play and mathematical analysis has been
considered since the late Romanitc period but didn't come into it's own
until the modern digital computer, with Claude Shannon taking an early swipe
at the problem as early as 1950!   The fact that it took more than 50 years
to get to Deep Blue's thin victory over Kasparov is more a testimony to how
subtle and hard Chess is than how intelligent humans are, etc.

"Deep Learning" itself seems like nothing more (and nothing less) than the
latest innovation in machine learning (game theory, neural nets, cellular
automata, genetic algorithms, learning classifiers, etc.) which *could* very
well portend the breakaway point of the AI-driven technological singularity.
I'm not THAT up on "Deep Learning" but things like Generative Antagonistic
Networks (and other unsupervised machine learning) seem to have the key
quality of not needing supervision by humans to learn...  there may be one
more level of indirection to be had before things go ape-shit (exponentially
speaking)...   

 I personally don't imagine that a *single* AI will be the source of this,
but rather a Cambrian-explosion-like plethora of AI's, though they may be so
pervasive and promiscuous as to cross-fertilize so thoroughly that they will
be a single "organism" for all practical purposes.  

</TL;DR>

- Steve

From: Friam [mailto:friam-bounces at redfish.com] On Behalf Of John Kennison
Sent: Monday, December 11, 2017 7:17 AM
To: The Friday Morning Applied Complexity Coffee Group
<mailto:friam at redfish.com> <friam at redfish.com>
Subject: Re: [FRIAM] Google self-evolving AlphaZero artificial intelligence
program mastered chess from scratch in 4 hours: Rich Murray 2017.12.10

I once thought I had a sure-fire way to make games between humans and
computers fairer. Start with a large set of chess-like games that use
different boards, different pieces, different rules. Enumerate the games so
that each one corresponds to a n-digit binary numeral (for large n). Then
make a "super game" in which the players start by creating a n digit binary
numeral by taking turns in which they can specify one of the n binary
digits. The super game would continue by playing the chess-like game that
corresponds to the created number. 

In a super game between a human and a computer, the computer would not have
access to all the insights into the nature of chess that humans have
established over hundreds of years of playing chess and which chess playing
computers use to defeat humans.  Of course, the human player would also be
deprived of all the years of research into chess, but humans can use their
marvelous intuition to figure out a reasonable set of strategies even for a
game they haven't studied before. The computer, without a reasonable set of
strategies, would (I assumed) find little benefit from  its massive
computing power. 

The new AlphaZero game playing computer refutes my idea.

  _____  

From: Friam <friam-bounces at redfish.com <mailto:friam-bounces at redfish.com> >
on behalf of Rich Murray <rmforall at gmail.com <mailto:rmforall at gmail.com> >
Sent: Monday, December 11, 2017 12:16:26 AM
To: Rich Murray
Subject: [FRIAM] Google self-evolving AlphaZero artificial intelligence
program mastered chess from scratch in 4 hours: Rich Murray 2017.12.10 

https://futurism.com/4-hours-googles-ai-mastered-chess-knowledge-history/

Chess isn't an easy game, by human standards. But for an artificial
intelligence powered by a formidable, almost alien mindset, the trivial
diversion can be mastered in a few spare hours.

In a new paper, Google researchers detail how their latest AI evolution,
AlphaZero, developed "superhuman performance" in chess, taking just four
hours to learn the rules before obliterating the world champion chess
program, Stockfish.

In other words, all of humanity's chess knowledge - and beyond - was
absorbed and surpassed by an AI in about as long as it takes to drive from
New York City to Washington, DC.

After being programmed with only the rules of chess (no strategies), in just
four hours AlphaZero had mastered the game to the extent it was able to best
the highest-rated chess-playing program Stockfish.

In a series of 100 games against Stockfish, AlphaZero won 25 games while
playing as white (with first mover advantage), and picked up three games
playing as black. 

The rest of the contests were draws, with Stockfish recording no wins and
AlphaZero no losses.

<https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campai
gn=sig-email&utm_content=webmail&utm_term=icon> 

Virus-free.
<https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campai
gn=sig-email&utm_content=webmail&utm_term=link> www.avast.com 

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
to unsubscribe http://redfish.com/mailman/listinfo/friam_redfish.com
FRIAM-COMIC http://friam-comic.blogspot.com/ by Dr. Strangelove

-- 

Russ Abbott 

Professor, Computer Science

California State University, Los Angeles

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20171211/b2e1903c/attachment.html>