Artificial Intelligence: Still Not So Bright?
The ability to analyze hundreds of news accounts rapidly is of keen interest to both government and corporate intelligence-gathering types. Now comes CIRCUS, artificial-intelligence software developed at the Natural Language Processing Lab at the University of Massachusetts at Amherst. The program, which tries to spot patterns from written reports and news articles, recently beat out 12 industry labs and two other universities in a test sponsored by the Naval Ocean Systems Center in San Diego to try to predict terrorist incidents.Prior to the test, participants were given 1,300 news stories to "teach" their computers. Competing systems were then scored on their ability to recall correctly and place in context 100 new articles fed into them during the test. The UMass program scored 51% on the recall aspect and 62% on comprehension--enough to garner the top recall score, as well as the best combined tally. That's no great shakes. But then, test architects say the point was to evaluate the state of the art in automated language processing, not to come up with a marketable system. Researchers are already scurrying to improve things. But for the foreseeable future, it looks as though intelligence agencies will still have to live up to their names.