How About a Nice Game of Chess?

It's been 30 years since the release of WarGames, one of the most well-received non-Star Wars science fiction films of the 1980s. The plot goes something like this: a teenage computer-enthusiast-slacker named David - ably played by a pre-Ferris Bueller Matthew Broderick - tries to hack into a local software company to find a set of forthcoming games. He does find a set of games which progresses from basic strategy (tic-tac-toe, backgammon, chess) to bizarrely apocalyptic ("Theaterwide Biotoxic," "Global Thermonuclear War," and so on). Turns out that young David has accidentally gained backdoor access to a supercomputer called War Operation Plan Response (conveniently abbreviated as WOPR, pronounced like the greasy sandwich), which was developed as an automated missile launch solution to compensate for skittish missileers who are unwilling to turn the launch key. The idea is that, by supervising the computer's gameplay and allowing it to learn increasingly complex strategy, WOPR would eventually function as a seasoned, unemotional commander who thinks nothing of incinerating millions of humans. After David initiates a rousing "game" of Global Thermonuclear War, all hell breaks loose at NORAD and WOPR goes haywire, convincing both US and USSR personnel that actual warheads have been launched. The film's ending is particularly brilliant: after the typical thrilling cat-and-mouse game, David saves the day by instructing WOPR to play tic-tac-toe against itself; after scores of games end in a draw, WOPR applies principles from the simpler game to Global Thermonuclear War, learning the concept of mutually assured destruction (MAD) and defusing itself. (Caution: link contains mildly blasphemous language.)

The genius of WarGames is that it's just far-fetched enough to pique viewers' interest, but plausible enough to terrify them in the midst of Cold War tensions. While machine intelligence was still in an embryonic state in 1983, it's made stunning leaps and bounds since. While we may never see a WOPR, we can now (somewhat) freely converse with our handheld phones, play board games against a range of capable automated opponents, and allow search engines to finish our thoughts.

But whereas these advancements seem intelligent and might pass a poorly-administered Turing test, they by and large do not learn. Siri and other "intelligent" personal assistants can recognize voice patterns and accommodate requests but are generally not adaptable. Chess-playing automata "simply" identify possible moves, evaluating the value of board positions and future scenarios based on coded notation and algorithms. Data mining tools "intelligently" extract information based on complex pre-coordinated classes and framework.

IBM's Jeopardy!-dominating supercomputer Watson seems to buck this trend. At its core, Watson operates in a similar manner to the program described above: it deduces written questions based on rapid analysis of similar keywords and sentence fragments, then instantaneously executes thousands of language analysis algorithms to determine a probable answer. Watson's game-playing abilities are not revolutionary for new algorithms but rather for the machine's raw computing power and speed: it consists of ninety 3.5 GHz servers and can process 500 gigabytes per second. In order to find answers Watson drew upon four terabytes of data.

While Watson's innerworkings are understandably trade secrets, IBM staff have hinted that it has the ability to "learn"; that is, to avoid mistakes by analyzing previous incorrect answers on the fly. For example, in a preparation round preceding the February 14, 2011 Jeopardy! competition, Watson answered a question whose correct answer is "Coffee-Mate" with "milk," having incorrectly interpreted the word "dairy" within the question. It's likely that the correct answer would then be stored in Watson's database with a fairly heavy algorithmic weighting to avoid making the same mistake a second time around.

Thirty years is a long time when considering technological development, especially fictional development. (The Jetsons is set in 2062; 2004's I, Robot occurs in 2035; hey, we'll get there!) While some may contend that our modern programs are leaps and bounds ahead of thirty-year-old science fiction, I think that a closer analysis reveals that they're more like novelty items which give the illusion of intelligence. ("Hey, ma! I'm talkin' to my computer!") A 2013 WOPR would likely bypass the game-based strategy learning and just consume loads of data and historical situations and be expected to act accordingly, but what fun is that? Sounds like the makings of a box office bomb…

(Image credits: Changing Gears | Gadgets Magazine)