Hermeneutics is not the sort of word that slips comfortably into a computer column but, what the heck -- it's become the latest buzzword among the artificial intelligentsia.
Derived from Greek, hermeneutics is the study of the methodological principles of interpretation, and it's a conceptual approach that's ignited the most tumultuous debate in AI circles in a decade.
Pioneered and articulated by Fernando Flores and Terry Winograd in their book "Understanding Computers and Cognition: A New Foundation for Design," the argument is that artificial intelligence research as currently defined -- expert systems, natural language interfaces, machines that think, etc. -- is doomed to fail.
To grotesquely oversimplify, Winograd and Flores assert that the "rational" tradition of AI, the belief that we can use symbols and representations of things to imbue machines with humanlike knowledge, is flawed to the point of intellectual bankruptcy. The authors say that language, not decision-making or expertise, is the prime determinant of "intelligence" and that, inherently, ". . . computers cannot understand language." Among their assertions:
We do not relate to things primarily through having representations of them.
"Meaning is fundamentally social and cannot be reduced to the meaning-giving activities of individual subjects." In other words, AI takes too much of a solipsistic view of intelligence, rather than recognizing that most of what we know is a function of relationships with others.
Every representation of knowledge is an interpretation. Drawing upon Heidegger and other hermeneutically inclined philosophers, the book so powerfully attacks the cornerstone assumptions of existing AI that it has become an intellectual succes de scandale.
Indeed, the book merited not one but four reviews in a recent issue of Artificial Intelligence, the community's most prestigious journal. A couple of the reviews were nothing short of scathing: "Many of the statements in their book bring to mind a cartoon I once saw -- it features a sheep in wolf's clothing. Their conclusions promise to be devastating, but they are in fact quite meek."
But other reviewers were completely taken. Wrote William J. Clancy of the Stanford Knowledge Systems Laboratory: "After reading the book twice and much consideration, I believe that Winograd and Flores are mostly right. We have here the stuff of Copernicus, Darwin and Freud: At its heart, the human world is not what we thought."
Okay, okay, Bill, but what do you really think?
What gives this book a cachet that other critiques of AI lack is its coauthor, Terry Winograd. A Stanford computer science professor, Winograd was once a tough young turk of the artificial intelligentsia. His natural language program called SHRDLU made quite a splash in the 1970s.
Now, says Winograd, older and wiser, "I've renounced the great expectations of the early day . . . and the idea that the technology will be the answer" to the question of machine intelligence.
Instead, Winograd and Flores are focusing their efforts away from the mechanisms of "intelligence" and more toward the interaction of man and machine. In essence, Winograd sees the problem as a failure to communicate. He sees the key research questions as focusing on how meaning and relationships are shaped by language.
"Understanding Computers and Cognition" (Ablex, Norwood, N.J. $24.95) is as much manifesto as it is a tough-minded analysis of rationalistic AI assumptions. There's an agenda here -- a clear challenge and, yes, a tossing down of the gauntlet and a pistols-at-dawn attitude toward the status quo.
Sure, this is an intellectual duel -- but it's a duel whose results may reshape both expectations and research.
Those of you enthusiastic about expert systems and the potential of AI may be due for a powerful disappointment. Those concerned with using PCs as a complement to, rather than as a substitute for, human intelligence may find succor in the wealth of intriguing questions the book poses.
Believe me, this book will be the basis for a new generation of software that is neither AI nor the conventional spreadsheet/data base stuff. Indeed, Flores already has software out that explores some of the principles espoused in the book. It's called The Coordinator. More about it in a future column.