Artificial Intelligence - 1
In 1969 I was awarded an M. Sc. for developing a useful set of AI techniques to solve an interesting but obscure little problem in computer language development.
Back in 1969, the term "Artificial Intelligence" seemed about right. We were programming computers to solve problems that would ordinarily require human intelligence. However, what we were doing (and what we are still doing) is using powerful and sophisticated programming techniques to solve "hard" problems. The techniques have become more and more sophisticated, the hardware has become more and more powerful and the problems have become harder - leaving the idea of somehow mimicking human intelligence in the rear view mirror. "AI" is now an engineering discipline, routinely creating electronic systems that solve problems far beyond any conceivable human capabilities.
Back in 1969, AI techniques were already very powerful and useful when applied to the right problem. However, at the time, things like speech recognition were regarded as very hard, perhaps impossible. Since then, AI techniques (basically very complex and sophisticated computer programming techniques) have made very significant progress in areas that would have been unthinkable back in 1969, including:
Back in 1969, the term "Artificial Intelligence" seemed about right. We were programming computers to solve problems that would ordinarily require human intelligence. However, what we were doing (and what we are still doing) is using powerful and sophisticated programming techniques to solve "hard" problems. The techniques have become more and more sophisticated, the hardware has become more and more powerful and the problems have become harder - leaving the idea of somehow mimicking human intelligence in the rear view mirror. "AI" is now an engineering discipline, routinely creating electronic systems that solve problems far beyond any conceivable human capabilities.
Back in 1969, AI techniques were already very powerful and useful when applied to the right problem. However, at the time, things like speech recognition were regarded as very hard, perhaps impossible. Since then, AI techniques (basically very complex and sophisticated computer programming techniques) have made very significant progress in areas that would have been unthinkable back in 1969, including:
- Automatic translation
- Automatic speech recognition
- Self-driving cars
- Face recognition
- Chess
- Jeopardy
Minsky has noted that there is a tendency for AI to be regarded as "mere technology" once it solves any particular problem. This contributes to the public misunderstanding of what AI is all about. It seems that the average person doesn't think of speech recognition as an AI application, it's just a smart phone "feature". However, in the late 20th century, some of the leading experts in the field regarded speech recognition as impossible. Some of the leading experts in the field gave it up as a hopeless fantasy. The Google search engine, clearly initiated as a problem in AI, has come to be so familiar and so powerful that, paradoxically, it's not seen as "intelligent" any more.
So, the massive and impressive gains of AI technology take place "under the radar" and are seen by the general public as just more examples of the inevitable march of progress. What the public is left to expect from AI is thinking robots. When I mention AI to my non-technical friends, they ask questions like "Can a machine think" or "will a machine ever feel". Such questions assume that somewhere, someone is at work building an artificial human and such a project is "Artificial Intelligence". They assume that a "thinking machine" will inevitably emerge from the constant march of progress in technology.
In fact, such a project is still has hopeless as it was in 1969 for the simple reason that humans have the unique ability to select the problem out of the infinite set of possibilities. A computer can play a great game of chess but cannot decide to put the chess board away and work on, say, driving a car. Put another way, a computer will never know why to play chess, drive a car or translate Japanese into Russian. This may seem like a minor issue, but it turns out to be huge. Computers don't know why they do things any more than a hammer knows why it hits a nail or hits someone over the head.
In fact, such a project is still has hopeless as it was in 1969 for the simple reason that humans have the unique ability to select the problem out of the infinite set of possibilities. A computer can play a great game of chess but cannot decide to put the chess board away and work on, say, driving a car. Put another way, a computer will never know why to play chess, drive a car or translate Japanese into Russian. This may seem like a minor issue, but it turns out to be huge. Computers don't know why they do things any more than a hammer knows why it hits a nail or hits someone over the head.
In fact, since World War II, humans have been employing AI on all kids humanly-decided problems. AI is having a huge and decisive effect on humanity as it automates human decision making, taking human beings "out of the loop". Most notably, the crash of 2008 was the result of mis-directed, incompetent and irresponsible human application of automated decision making, but, because all the humans did was basically "write programs" or "push the button", nobody was held accountable. For example, it is best to think of Goldman Sacks as a machine, playing a game to maximize profits for its stake holders. If it plays badly, catastrophe in the "outside world" results, but it's nobody's fault. Even though corporations have the legal rights of persons under the law, we don't send machines to jail or submit them to capital punishment.
Comments
Post a Comment