Double Feature: Computer Chess vs. Game Over — Kasparov And The Machine
“If all computers can do is calculate, then what is artificial intelligence?”
This is a double movie review, Computer Chess vs. Game Over — Kasparov And The Machine. These are both films addressing the development of artificial intelligence technology in the context of the game of chess that h+ Magazine readers will enjoy. Both films are currently available for streaming on Netflix.
This review includes some spoilers.
Artificial Intelligence and Chess Programming — A Brief Introduction
Chess has a long and interesting relationship with the field of machine intelligence. Perhaps the earliest attempt to develop a chess playing machine was a fraud. Known as The Turk, which was invented by Wolfgang von Kempelen in 1769 and was brought to the U.S. in 1825 by Johann Nepomuk Mälzel after von Kempelen’s death. The Turk worked by employing a hidden human chess player who was seated inside of what appeared to be a mechanical device and it is an interesting story in that the emphasis on deception later embodied in Alan Turing’s famous Turing Test is already evident.
Edgar Allan Poe wrote an essay arguing the Turk was a fraud, titled Maelzel’s Chess Player in 1836. However his ideas about how the machine worked were largely erroneous. What is perhaps more interesting about Poe’s essay is how he goes about arguing that the Turk must include, however it is contrived, a human player or as Poe terms it a “mind”.
[su_quote]The moves of the Turk are not made at regular intervals of time, but accommodate themselves to the moves of the antagonist — although this point (of regularity) so important in all kinds of mechanical contrivance, might have been readily brought about by limiting the time allowed for the moves of the antagonist. For example, if this limit were three minutes, the moves of the Automaton might be made at any given intervals longer than three minutes. The fact then of irregularity, when regularity might have been so easily attained, goes to prove that regularity is unimportant to the action of the Automaton — in other words, that the Automaton is not a pure machine. [/su_quote]
Of course Poe has no understanding of the mechanical production of apparently random sequences known as pseudorandom number generators.
He observes that the shoulder of the automaton seems to respond to the human opponent in advance of their completing a move. He concludes that the motions “are regulated by mind — by some person who sees the board”. Interestingly he also asserts that a mechanism would be universally able to beat human players, “The Automaton does not invariably win the game. Were the machine a pure machine this would not be the case — it would always win.” which of course is true today regarding the best chess playing machines in the world but, in general, is not true of arbitrary mechanical chess players. He also argues that the automaton isn’t really human-like and therefore can’t be “thinking”, for example it doesn’t pause and consider its moves or make appropriate facial expressions while doing so.
But chess playing wasn’t doomed to be a fraud forever, because interestingly Poe was partially correct. A perfectly executed chess playing machine would indeed beat any human player. But it wasn’t until 1997 that a machine beat Gary Kasparov, then the world champion, in a tournament. However, the start of real chess programming was in a military research project at Los Alamos National Laboratory.
Following World War II, a group of scientists were gathered together at Los Alamos to develop advanced computing machines in what became known as the Metropolis Project, named after its manager Nicholas Metropolis. The Metropolis Project included a variety of rather famous individuals, Claude Shannon, John von Neumann, and Stanislaw Ulam for example, and notably it was while working on the project that John von Neumann reportedly coined the term “singularity”.
The Metropolis Project made one of the first attempts to explore use of computers to play chess. However, because the computer they had available to them, known as MANIAC, was little more than a hand calculator by modern standards, they developed a reduced game known as Los Alamos chess. Los Alamos chess is played on a smaller board with the bishops removed and this greatly reduces the scope of the game tree search required for a machine to play well.
During the 1970s and 1980s, chess playing machines progressed from toys to the grandmaster level. Interestingly, although chess playing was a staple of AI research from the start, successful chess machines at this time commonly did not use AI techniques. They were, for example, largely based around brute force and didn’t employ so called “AI languages” such as LISP or PROLOG. While most AI researchers of this era found themselves deep in the “AI WInter”, chess playing machines advanced. Despite this, chess playing provided an excellent platform for AI research, and was commonly presented and discussed in academic AI courses at the time.
This sets the stage for the first film to be reviewed, Computer Chess.
Review: Computer Chess
“Computer Chess” is a film about the early days of computer chess programming, nerd culture, and delves into the deeper territory of what exactly makes us human or machine.
The film is set in a roadside hotel where an eclectic group of computer programmers gather for a tournament where chess playing machines play against each other. Simultaneously at the same hotel, a human potential movement group, called The Seekers is meeting and exploring cosmic truths. The storyline explores the notions of “spiritual” exploration and the development of technology, an artificial intelligence, a theme and dichotomy which will be familiar to transhumanists, for example appearing in Zoltan Istvan’s The Transhumanist Wager via the character of Zoe Bach.
The story is centered on the junior programmer of an academic team, Peter, who suspects that their computer is able to detect the difference between a computer opponent and a human one, and thus is exhibiting elements of self-consciousness. He later learns from another member of his team that the computer thinks it’s alive.
The head researcher, ridicules Peter’s hypothesis while boasting the the program known as SAR is “self-correcting” which is to say self modifying and seemingly this self modification has led to it accidentally becoming conscious or self aware. The involvement of the U.S. military in funding the research is also intimated.
This movie is fictional, but includes some great period pieces ranging from hairstyles to the computers themselves. If you grew up during the era of TRS-80s and Commodore 64s, you are going to enjoy a lot of this period schtick. The various AI developers take very different approaches, and are themselves all rather idiosyncratic and interesting, a fact which is elucidated at the beginning of the film during a pot smoking session.
If you hang out in any online forums were AI is discussed, e.g. the Facebook Strong AI group, you may recognize some of the characters in this film such as the developer who is exploring the “feminine” or “intuitive” side of AI. I laughed out loud.
Through the film the programmers cross paths with The Seekers. The Seekers are exploring what it means to be really human, and seemingly indulging in a bit of wife swapping, and swinging encouters along the way. I grew up around people like this in Berkeley California. Hilarious.
After realizing what he has created, Peter leaves the machine near a window where it is destroyed.
In the final scene, easily missed, a prostitute apparently solicited by Peter, reveals herself to be a robot. Or is she? The movie leaves open the question of what makes us humans vs. machines. This film will make you laugh out loud and it has some deeper insights too. Highly recommended.
Review: Game Over — Kasparov And The Machine
Unlike Computer Chess, Game Over is a straight documentary reporting on Gary Kasparov’s loss to Deep Blue in 1997. One of the important ideas revealed in this film is that while Kasparov is struggling to beat Deep Blue, it is revealed that he isn’t just playing a machine but a team of chess experts and programmers who manually alter the program between the games.
As a documentary, the film isn’t groundbreaking but it does document this important reality of the story that is seemingly lost — Deep Blue wasn’t a “true” AI but a man-machine symbiotic system. In this sense, Deep Blue followed the tradition of The Turk appearing to be “just a machine” but in reality involving plenty of human reasoning and expertise behind the scenes.
The film explores Kasparov’s psychology and his reactions to learning who and what he was really up against. The film includes interviews with Kasparov, his manager, chess experts, and members of the IBM Deep Blue team, as well as original footage of the match itself. The idea that Deep Blue was “faked” was extremely controversial at the time and some have called Kasparov a “sore loser” for his behavior after losing.
I watched this film immediately after Computer Chess and found the juxtaposition interesting. In Computer Chess, we see humans, who are all too human, trying to build real chess playing machines and accidentally creating a sentient machine intelligence.
In Game Over, humans are seen to be just faking it.
Deep Blue isn’t just a machine, but a group of people and a machine. It is a futuristic version of The Turk.
And the same is seemingly true of many AIs today which require teams of humans to manage and operate them, for example IBM’s famous Jeopardy winning AI Watson.
Not everything is as it appears to be in AI research.
Of course since the development of Deep Blue chess playing programs have continued to advance.
The best chess playing programs have ELO ratings above 3300. Magnus Carlson, the current human world champion, has an ELO rating of 2862.
But these machines are really just encoding the knowledge of human players with a game tree search of some sort. They are software “turks” following in the tradition of Deep Blue.
However real machine chess players are finally on the horizon.
We have, for example, Joel Veness’ Meep program which taught itself to play at the master level. This is quite distinct from Deep Blue, since Meep learns to play by playing against itself rather than by having humans writing new code to help it play better or by playing a human teacher.
A recent result, Deep Pink, used deep learning methods to build a chess playing machine. While this program doesn’t play as well as some others, it does suggest that a pure machine learning approach could be successful at learning to play chess well, and without any human intervention.