Volume 14, Number 3
PATTERN RECOGNITION LETTERS
Figure 4. The NASA/Goddard Space Flight Center Intelligent Data Management IIFS (courtesy of R.F. Cromp).
“Where's the AI?"
The issue of scalability is a key point of an interesting article by Roger Schank in the winter 1991 issue of AI Magazine (Schank (1991)). Key position statements of the article are (1) "The answer to where's the AI? is that it's in the size"; and (2) “AI means a machine that learns and from this view point there is no AI, at least not yet". Commenting on the view that, "if no machine did it before, it must be AI." Shank says, "Two important types of programs to discuss within this conception of Al are optical character readers and chess-playing programs. Are these programs AI? Today, most people would say that they are not. Years ago, they were. What happened? The answer is that they worked. They were Al as long as it was unclear how to make them work."
This is also not an uncommon experience concerning reactions to human intelligence and problem solving as has been commented on by a number of people. However, in many areas, including OCR and speech recognition, the statement that something "works" must be highly qualified. While we are much further along in OCR than the technology used in the postal readers when I was with Philco-Ford Corp., the latest report on the status of postal address reading is that the system being developed at the Center of Excellence for Document Analysis and Recognition at the State University of New York at Buffalo is able to read the Zip Code about 75% of the time and can read an address and Zip Code about 30% of the time. The current failure of hand-held pen computers to catch on in the marketplace is attributed to poor performance of the recognition software in reading handwritten input. Possibly in another two years much better performance will be achieved. Currently automation of handwriting recognition is far from the state of "working".
At a recent voice recognition show, the systems being exhibited performed equally poorly except in highly constrained situations. However, recent reports indicate progress. My former student, P.S. Gopalakrishnan (Gopal) works in Fred Jelinek's group at the IBM Watson Research Center. According to their publications, that group has been following an approach to speech recognition based on hidden Markov modeling and