In May 1997, millions of people, despite likely never having played a game of chess themselves, gathered around their televisions to watch chess grandmaster Garry Kasparov play. His opponent, what appeared to be a pair of six-foot-tall cupboards enclosed in a black perforated metal grille, was IBM’s “Deep Blue” chess computer. Kasparov lost, and the event entered our cultural milieu as a “milestone” in artificial intelligence.
Stories of Kasparov’s defeat were soaked in the language of combat. “A technological knockout. Machine defeated man on Sunday”, wrote the Kentucky New Era. “a sensational rout … in the future we will be up against machines that are not only monstrously intelligent but utterly unfeeling”, prophesied the Pittsburgh Post-Gazette. “It crushed the world champion … ‘I was rightfully massacred’ said Kasparov”, the Guardian reported. One could be forgiven for thinking that IBM’s 1.4-ton machine chassis had leant over the table and bashed Kasparov on the head.
The media tends to romanticize the warlike aspects of artificial intelligence. News coverage often portrays AI as being in direct competition with humans. In this narrative, games are both metaphors for war and tests of intelligence.
Nearly twenty years later in 2016, the world champion Go player Lee Seedol found himself in a similar faceoff, this time against Google’s AlphaGo software. Deep Blue had beaten Kasparov by considering all possible chess moves and responses, selecting the move most likely to lead to victory. A Go player must consider a vastly greater number of moves in each turn than a chess player, leading to what researchers call a “combinatorial explosion” of possibilities. Considering all possible Go moves and responses, for even a few turns, was thought to require an infeasible amount of computational power. The matches were broadcast live to tens of millions of viewers, including 60 million Chinese-speaking viewers alone. Like Kasparov, Sedol lost. His defeat, like his predecessor’s, was hailed as a milestone for artificial intelligences and a tragedy for human minds.
Progress at the frontlines of artificial intelligence is jagged and uneven. Researchers can make great strides in a particular task, such as face recognition, while progress in others, such as text to speech transcription, might be stalled for years. The public does not often look with interest upon advancements, however significant, in narrow domains. It feels like a stretch to take any such task as representative of human intelligence. But this is not true of games.
We look to games as tests for artificial intelligence. The “imitation game”, proposed by Alan Turing in 1950 (and hence often called the Turing test), is the most famous. To win, an artificial intelligence must hold a conversation so convincing that a human cannot reliably detect that they are conversing with a computer, and not another person.
In ancient India, China, and Japan, games such as Chess, Go and Shogi were considered to be tests of intelligence and metaphors for war. A precursor to modern chess in 6th century India, the game chaturanga – meaning four parts – referred to the four divisions of the military. Elephants (which became the bishop), chariots (the rook), horsemen (knight), and footsoldiers (pawn), and their movements represented the military strategy of the time. Medieval Vikings played hnefatafl, a simulation of castle warfare. In the 19th century, the Prussian army developed Kriegsspiel, a family of highly realistic decision-making games to train war officers in battlefield tactics. Prussia’s victory in the Franco-Prussian War has been attributed, in part, to strategy developed through Kriegsspiel.
In the 18th century, a shared vision of ancient war games and artificial intelligence was manifested in a legendary machine known as the Mechanical Turk. The Turk was an anthropomorphic chess-playing robot that toured Europe and North America for nearly a century, defeating challengers including Benjamin Franklin and Napoleon. Though the Turk was destroyed in a fire in 1854 (its last owner believed he had heard it whisper its dying words through the flames: “Check! Check!”), drawings and engravings remain. A swarthy Turkish lord, bedecked in a regal turban and fur-lined robes, a long smoking pipe in one hand, ready to be slipped between mustached lips. He leans over a large wooden cabinet on top of which is a chess board. Its mere presence places the viewer in the position of the opponent. One cannot look at it without being drawn into the scene it creates.
Of course, the Turk was not intelligent, but a hoax: an ingenious construction which housed a secret compartment in which a human player could sit. An automaton hoax could have taken any form. Why did the designers choose to make it a chess-playing Turk? Possibly because, at the height of the Ottoman empire, their fearsome war prowess could be manifested in a sanitized manner for European nobles as chess skill. The Mechanical Turk was a fantasy of war as much as it was a fantasy of intelligence.
If games are a metaphor for war, and also tests of intelligence, then does our obsession with making AI play games stem from, or lead inevitably to, a view of AI as a tool of war? Probably not. The many peaceable and noncompetitive applications of AI provide enough counterexamples. No one is riding into battle equipped with email junk filters, autocorrect, or Amazon Alexa.
But war must be an aspect of how we relate to the idea of intelligence. Ethologists, who study animal behaviour, now acknowledge that each species carries its own type of intelligence, its own modes of cognition, emotion, and problem-solving. Like physical bodies, intelligence too is shaped by environmental and evolutionary circumstances. For homo sapiens, a species whose history is one of near-constant conflict, being successful at war must be a powerful selective force.
Under the surface of media excitement over AI gameplay may just lie an ancient preoccupation with military strategy. Are we smart? Are we smart enough?
Are we smart enough to defeat the enemy?
I received excellent guidance and feedback from Dr. Helen Scales, who runs the course “Popular Science Writing” at the University of Cambridge Institute for Continuing Education.
Notes and references
Chess as Indian military strategy:
- Kulke, Hermann; Dietmar Rothermund (2004). A History of India. Routledge. ISBN 0-415-32920-5.
- Meri, Josef W. (2005). Medieval Islamic Civilization: An Encyclopedia. Routledge. ISBN 0-415-96690-6.
Coverage of Deep Blue v. Kasparov:
- Kentucky New Era: https://news.google.com/newspapers?nid=266&dat=19970512&id=DfMrAAAAIBAJ&sjid=52wFAAAAIBAJ&pg=3496,887866
- Pittsburgh post-gazette: https://news.google.com/newspapers?nid=1129&dat=19970519&id=Vs5aAAAAIBAJ&sjid=vm8DAAAAIBAJ&pg=5029,3833524
- The Guardian: From the archive, 12 May 1997: Deep Blue win a giant step for computerkind | Chess | The Guardian
The last words of the Mechanical Turk: S[ilas] W[eir] Mitchell (January 1857). “The Last of a Veteran Chess Player”. The Chess Monthly. Vol. 1. pp. 3–7. hdl:2027/hvd.hn43vw; continued in February 1857. pp. 40–45. Reprinted: Levitt, 236–240, “Appendix L. Mitchell’s ‘The Last of a Veteran Chess Player’ (1857).” (via https://en.wikipedia.org/wiki/Mechanical_Turk#Final_years_and_beyond)
In the 21st century, the term “Mechanical Turk” is more commonly used to mean Amazon’s Mechanical Turk service, which allows employers to outsource small pieces of digital gig work to a global pool of casual labourers. The name is a direct reference to “human-powered” AI, but its meaning and significance is lost on Amazon.