Thinking Machines: The True Story of Artificial Intelligence

Can machines think? In the 1950s and 60s, intelligent machines were thought to be only decades off. Today, almost fifty years later, we ask "Where are the thinking machines?" We'll take a look at the true story of why artificial intelligence hasn't lived up to the expectation of full human intelligence, whether we can expect to see the first truly intelligent machines any time soon, and what level of computer intelligence has already been achieved.

In 1950, just after the creation of the first modern computers, father of computer science Alan Turing described the "Turing test" for artificial intelligence, a conversation game which pits potential computer contenders against the wits of a real human opponent in a game of imitation. The computer wins if it convinces a human judge, though text conversation, that it is in fact the "real thing". According to Turing's estimations, by the year 2000 computers would be available that could fool human judges 30% of the time. As the year 2000 passed, his prediction couldn't have been more wrong - no computer has yet passed the Turing test. What went wrong? Have we made any progress towards the goal of true artificial intelligence?

Some claim that this goal is impossible to attain, that the essence of the human "soul" can't be copied in a computer. Others say that true artificial intelligence would pose a grave danger to humanity, and so shouldn't be developed. However, after over fifty years of attention from some of the best minds, many scientists and researchers would agree that creating a human-like intelligent computer is just a downright hard problem to solve. Early attempts, such as Joseph Weizenbaum's 1966 ELIZA chat program, fooled more than a few people who were all too eager to meet a real AI, but worked with little more than a few canned phrases and almost no real understanding of human conversation. Such approaches were at best a parody of the complex and subtle rules that define human language and social interaction.

Discouraged by the difficulty of simulating realistic conversation, some artificial intelligence researchers turned their attention to more limited aspects of human intelligence, which were believed to provide a more level playing ground in this battle of wits. For example, chess-playing computers have become increasingly successful over the past 50 years, with IBM's Deep Blue computer finally defeating Chess World Champian Garry Kasparov in 1997. With such success in chess, we might wonder why intelligent computers still can't even beat human toddlers in areas like language competence and problem solving.

As it turns out, full human-like artificial intelligence (known as "strong AI") is still considered by most AI researchers to be well beyond reach at present. This is why research turned to problems like chess-playing, called "weak AI", in which more limited aspects of intelligence are simulated. Weak AI does not attempt to recreate the complex and subtle human abilities of self-awareness and robust problem-solving; instead, it tackles narrower problems like identifying images, designing optimal schedules, and playing chess. This approach has so far been very successful and has produced results that many of us put to use every day.

Have you ever called a telephone support desk to be greeted by a computer that can understand your voice? Used a PDA or cell phone that can convert your handwriting to text? Played a video game with an "intelligent"-seeming computer opponent? Had a spam filter remove annoying junk e-mail? Used a search engine? These are all examples of the results of weak AI research in the past decades. And exciting new research is just around the corner: AI has recently been applied to problems like medical diagnosis, in which the computer suggests a diagnosis based on patterns in the patient's symptoms, with great success.

We might wonder whether such systems can be said to "think", in the sense that Turing was looking for in his test. Indeed, fully intelligent computers may still be a long way off, or the current line of research may never lead to this goal. But consider that with such practical, useful results, and new work constantly appearing in areas such as speech recognition and medical diagnosis, the "true story" of artificial intelligence has found unexpected success.