Section 13.9 Special Topic: Are Computers Intelligent?
Contemporary computer interfaces are largely visual and graphical, and many things we use a computer for, such as word processing, still require us to type. Will there come a day when instead of typing a letter or e-mail message, we'll be able to dictate it to our computer? Will computers eventually have the same kind of interface we have—that is, will we someday be able to carry on conversations with our computers? Clearly, a “conversational interface” would require substantial intelligence on the part of the computer. Do computers have any chance of acquiring such intelligence?
The question of machine intelligence or artificial intelligence (AI) has been the subject of controversy since the very first computers were developed. In 1950, in an article in the journal Mind, Alan Turing proposed the following test to settle the question of whether computers could be intelligent. Suppose you put a person and a computer in another room, and you let a human interrogate both with any kind of question whatsoever. The interrogator could ask them to parse a Shakespearian sonnet, or solve an arithmetic problem, or tell a joke. The computer's task would be to try to fool the interrogator into thinking that it was the human. And the (hidden) human's task would be to try to help the interrogator see that he or she was the human.
Turing argued that someday computers would be able to play this game so well that interrogators would have no better than a 50/50 chance of telling which was which. When that day came, he argued, we would have to conclude that computers were intelligent.
This so-called Turing test has been the subject of controversy ever since. Many of the founders of AI and many of its current practitioners believe that computation and human thinking are basically the same kind of process and that eventually computers will develop enough capability that we'll have to call them intelligent. Skeptics argue that even if computers could mimic our intelligence, there's no way they will be self-conscious and, therefore, they can never be truly intelligent. According to the skeptics, merely executing programs, no matter how clever the programs are, will never add up to intelligence.
Computers have made some dramatic strides lately. In 1997, an IBM computer named Deep Blue beat world chess champion Gary Kasparov in a seven-game chess match. In 1998, a computer at Los Alamos National Laboratory proved a mathematical theorem that some of the best mathematicians were unable to prove for the past 40 years.
However, despite these achievements, most observers would agree that computers are not yet capable of passing the Turing test. One area where computers fall short is in natural language understanding. Although computers are good at understanding Java and other computer languages, human languages are still too complex and require too much common sense knowledge for computers to understand them perfectly. Another area where computers still fall somewhat short is in speech recognition. However, an American company recently demonstrated a telephone that could translate between English and German (as well as some other languages) in real time. The device's only limitation was that its discourse was limited to the travel domain. As computer processing speeds improve, this limitation is expected to be only temporary. Thus, we may be closer than we think to having our “conversational user interface.”
Natural language understanding, speech recognition, learning, perception, chess playing, and problem solving are the kinds of problems addressed in AI, one of the major applied areas of computer science. Almost every major research group in AI has a Web site that describes its work. To find some of these, just do a search for “artificial intelligence” and then browse through the links that are returned.