Newspapers / Daily Tar Heel (Chapel … / Sept. 28, 1988, edition 1 / Page 19
Part of Daily Tar Heel (Chapel Hill, N.C.) / About this page
This page has errors
The date, title, or page description is wrong
This page has harmful content
This page contains sensitive or offensive material
10CompuFest '88Wednesday, September 28, 1988 Artificial intelligence: Is it an impossible dream? By ANDREW THOMPSON Science Writer Most students' only direct contact with computers is to use them as glorified typewriters. But how many students have entertained the possi bility that,, one day, the computers will be writing their term papers for them? Since the 1950s, when the mythical term Artificial Intelligence (AI) was first coined, the possibility of creating thinking machines has fascinated researchers and the public alike. "Things like common sense, intuition, reasoning and general problem solving are hard nuts to crack " Anil Nair i: It also raised questions as to humans? Machines? Computo, ergo whether computers would ever be sum. able to feel emotions and be creative. . Three decades and loads of money And if computers ever possessed these (mostly from the Pentagon) later, the traits, what would that make . same questions remain unresolved. fnPK rnn I'll - f 1 If ran nn w It's the freedom every user deserves. And the flexibility every MISDP.manager wants. Presenting Control Data's Transparent Computing Environment (TCE). On a desktop, TCE looks like one powerful, logical computer. Users don't need to know where their applications are actually running, or where their information is stored. The TCE seamless user interface makes controlling your systems less complicated, so users can get more work done. Behind the scenes, the Transparent Computing Environment accommodates all of your computing resources, both from 1988 Control Data Corporation. Transparent Computing Environment is a trademark of Control Data Corporation. Control Data and other vendors. So you can finally achieve what other mainframe manu facturers only wave their hands at. Like integrated information management. Or application software that runs on different processors. The Transparent Computing Environ ment is the open systems architecture for the 1990s and beyond. One that respects the rights of both users and managers. Once you've seen the Transparent Computing Environ ment, there's really nothing else worth looking at. CONTRpL DATA Although many computer programs are good at carrying out specialized tasks, they still cannot think in a human sense. Perhaps the AI researchers have not had enough time to vindicate their more optimistic claims. Or it may be that trying to build a thinking machine is theoretically impossible. One of the more successful AI programs, called MYCIN, was deve loped at Stanford University by Edward Shortcliffe. This program has the ability to diagnose bacterial blood infections after details of the patient's case history have been typed in. It works using heuristic if-then rules educated hunches. An exam ple of such a rule would be, "If organisms were not seen in the test culture and if the patient was badly burned then the infection may be P. aeruginosa." MYCIN can suggest medical tests to fine-tune the patient's diagnosis, as well as recommend a course of treatment. Its predictions have been shown to be on a par with those of a medical -practitioner, and the program has been marketed success fully as a reference tool. MYCIN is known as an expert system. Other expert systems have been developed to simulate weather, systems, design cars, play chess (to a very high standard) and predict the outcome of wars. They all operate very well within their own spheres of knowledge, but they have limitations that deny them any claim to real human intelligence. The most important limitation is that these programs are unable to adapt themselves to different condi tions; they are "brittle." One of the main reasons for that is the difficulty in creating a know ledge base large enough to deal with aspects outside the program's own narrowly defined area. All of us carry in our minds huge amounts of general knowledge and common sense, which we take for granted. This base allows us to get around on a day-to-day basis and to :ope vin any unexpected situations. Programming even a fraction of this knowledge into a computer is a very difficult task, and that explains why expert systems have remained idiot savants. Apait from the practical limita tions of these programs, AI has also been under philosophical attack. John Searle, a philosopher at Ber keley and a leading expert in this area, supports the claim that AI could be used as a tool to simulate aspects of mental processing. But he does not believe that AI could ever think intelligently in a human fashion. Searle argues that a program which uses formal rules to manipulate symbols can never think or be aware Creative because the symbols do not mean anything to the computer. : Herbert Dreyfuss, a colleague of Searle, also denies that AI could ever attain a deep level of thinking. He argues that perception, understand-' ing and learning are not just a matter of rules to be entered into a program. Rather, they are holistic processes, impossible to pin down with formal axioms. Recent research of the 1980s may yet prove the armchair philosophers wrong. At present, the new fields of "parallel computers" and "neuronal networks" are blossoming. Parallel computers are designed so the many components that carry out the information processing can oper- : ii l.. ti : : to most of today's computers, which only operate sequentially, one task after another. The improvement is equivalent to having an extra nine workers on a job, rather than making one worker labor 10 times more quickly to get the same job done. Just as a worker reaches a limit in productivity, so a serial computer reaches a limit in speed of processing. Parallel computers will be faster and more efficient. " The rationale behind the neuronal network approach is that if the brain of a living organism is good at thinking, why not model the hard ware of a computer on the structure of the nervous system? One UNC professor of philosophy, Stan Munsat, is especially optimistic about the potential of neural net works. "None of the objections of Searle and Dreyfuss apply to neural networks," he said. "They don't apply because neuronal systems aren't rule followers." The most promising aspect of the neuronal networks is their ability to learn, albeit at a rudimentary level, from their past experience. Because they lack a rigid structure of connec tions, they have a "slippery" quality that enables them to cope with new information and . unforeseen situations. But the ultimate success of these new developments remains to be seen. Anil Nair, a UNC computer science graduate who now works in Silicon Valley, Calif., is not confident about the more ambitious claims of AI. "Things like common sense, intuition, reasoning and general problem solv ing are hard nuts to crack," Nair said. "And I dont think it's worthwhile to pursue solutions for these problems." Will computers be replacing stu dents in their scholastic research? We need not worry for the time being. It will be many years, if ever, before the artificial intelligence of computers replaces the natural intelligence of humans. from page 8 manuscript paper method. "In a way, it's made quite a contribution to musical illiteracy, and It brings out the best in all of usT made inferior music instantly avail able in the performance arena." In addition, although computers are helpful in score-writing, Hannay said, "by Composing' right into the computer, the mystical quality of individuality in notation is lost. From this point on, libraries are going to be collecting disks, and you won't find the infinitely variable calligraphic idiosyncrasies of a Stravinsky manu script or a Brahms manuscript. Everyone's manuscript will look exactly the same. It is similar to the effect on writers when the typewriter came into general use, I think. It resulted in a definite change in writing styles."
Daily Tar Heel (Chapel Hill, N.C.)
Standardized title groups preceding, succeeding, and alternate titles together.
Sept. 28, 1988, edition 1
19
Click "Submit" to request a review of this page. NCDHC staff will check .
0 / 75