22 Being a Nine-Day Wonder
Back in the early 1980s, the field research for The Fifth Generation was eye-opening, but more so was the aftermath. The book became a best-seller in Japan and the United States and eventually, with authorized and unauthorized copies, sold about half a million copies around the world. Ed Feigenbaum and I enjoyed the giddy experience of being nine-day wonders. Yes, it’s fun to walk along Madison Avenue and see your own book in bookshop windows. Translations abounded, the phone rang constantly, and the publishers and we were happy at last. Congress held hearings: was Japan’s Fifth Generation a threat to national security? Should the National Science Foundation or DARPA invest more dollars in computer research? Snarky researchers claimed that Feigenbaum and I had written the book only to beef up his research budget. In that farfetched scenario, I was a willing and invisible tool. But I was to become invisible in other ways, too.
Invitations poured in to be interviewed, give talks, be lionized, be attacked. Among the odder lionizations for me was a phone call from Vogue magazine. They’d seen me interviewed on public television’s MacNeil-Lehrer NewsHour and wanted to conduct their own interview. Because they’d fingered me on a relatively serious TV news show, I thought their interview would be moderately serious, too.
But no. What did I think about clothes? (I loved them; I cared about self-presentation, as anybody who’s seen me knows.) What was my idea of a balanced day? (Being left alone to work.) Did I see enough of my husband? (Sure. Next question?) Work comes first? My goodness! And your husband knew that before you got married? Goodness! Finally, if I could throw it all over and escape to paradise, what would paradise look like? I thought I was already in paradise in New York City and harbored no urges to open a bed-and-breakfast in New Hampshire or a boutique in Mendocino.
A few months later, I picked up a copy of Vogue at the airport and saw I’d been placed in an article called “Life at the Top: Women Talk about Success, Time, Love.” I was amused, especially as I sounded like the compleat nerd. But I was sandwiched between Alice Waters and Mrs. Byron Janis, an apotheosis of some kind, I guess.
It pains to me to say that a best seller also brought out the worst in journalists. Not my first experience: The Newsweek cover story for June 30, 1980, was “Machines That Think.” The issue appeared just weeks after Machines Who Think was published but had no mention of my book or me. Possibly coincidence.
When The Fifth Generation was published, Newsweek made it a cover story, pictured my coauthor, mentioned the book—but never once mentioned me as coauthor. “Maybe the third time I hit,” I wrote in my journal, “I’ll even see my name attached to my ideas.” More egregious was Time magazine. Writing about Japan’s effort, the journalist carved out some of the best parts of the book and wrote as if he’d discovered them for himself. Whole quotes were lifted from the book and attributed solely to Feigenbaum, but neither the book nor I was mentioned. To this I did write a note of protest—polite, slightly humorous, saying if writers don’t respect each other as specialists, who will? The theft is more than petty.
The journalist called me immediately, deeply apologetic. He’d done it, knew he shouldn’t have, and was even more ashamed to receive such a nice note from me instead of the chewing out he richly deserved.”You were so quiet when you were in my office I figured you didn’t have much to do with the book.” To that I was nearly speechless. Did he think I was Ed’s NYC bimbo? No, what he saw was years of socialization, of being properly deferential to men. I told my sister the story, and she said: “I expect you’ve learned a lesson you’ll never forget.” I reported this in my journal and then went on to speculate:
June 28, 1983:
But this raises interesting questions about how writers are perceived, especially by other writers. I’m put in mind of the reporter who picked my brains for fifteen minutes then asked me for the name of an “expert” who could tell him all that. In some sense, it reflects American anti-intellectualism, which is to say that we honor doers, no matter how good or bad, but not those who think about things and try and make sense of them. But why should journalists be the worst at that? Interestingly, people in the field I write about don’t have that attitude. To them, I’m a professional—in a different profession. They appreciate how well I perform my craft, and scold me when I falter. I’m not interested in celebrity, but I am saddened by lack of public recognition, which is slightly different. And that I lay to the curious attitudes of members of the press. Speaking of which, one such called me a day or so ago to ask questions. He’d obviously read nothing by me (nor anyone else in the field). How can people waste your time like that?
May 21, 1984:
Joe hears from David Lee, his former graduate student, that David met an IBM researcher at Yorktown Heights who’s just come from Stanford. “How come,” says this young former Stanford student, “that Pamela McCorduck let Ed Feigenbaum put his name on that book when she could just as easily have done it herself? Everybody at Stanford knows Machines Who Think. Everybody knows that Feigenbaum not only didn’t write The Fifth Generation, but hardly knows what’s in it. Why’d she do it?”
The IBM researcher continued with further wicked graduate student folklore, each tale growing taller and more distant from anything resembling fact.
But I’d endured so much “only the ghostwriter, not coauthor,” even from the book’s original publisher, so much “we only want to interview the expert,” that I laughed. Hard.
Feigenbaum and I were invited back to Japan from time to time to celebrate Japan’s coming of age in computing. Tokyo, November 8, 1984:
That evening, the banquet. Go to pick up Ed, who sits on his chair, barefoot, and says, “I can’t go on. I’m so tired. I just can’t go on.” I say,“Cheer up, you’ll make it. Only two hours and then you can go to bed.” We go down and get into the reception line, where Ezra Vogel joins us. [Vogel a well-known Harvard expert on both Japan and China.] We three are chatting briskly when two kimono-clad ladies walk up to the men, bow, say “spesha guest” and lead each of them away, leaving me standing there looking more than a bit forlorn, I’m sure. Furokawa-san has been watching this, rushes over to me with a “spesha guest” chrysanthemum tag, for sure enough, I too am a “spesha guest,” but the ladies are so unused to a woman being such that they’ve overlooked me. An elegant French-style banquet, and I’m asked to say a few words. I talk about the momentous occasion not only for Japan, but for the human race. Had I known I was one of only two people to make such remarks, I’d have worked harder at it, but short and sweet, thinking I’m one of many. Afterwards, much to-ing and fro-ing, and Ezra, Ed and I retire to my room, where Ed, risen to life like the phoenix, holds forth until Ezra literally falls sideways off his chair in jetlagged fatigue, and I’m incoherent. This is the man who could hardly put his shoes and socks on when I went to get him.
November 11, 1984:
Ed takes me to Akihabara, where I see more electronics than I’ll ever want to see again, and demonstrates something called a karaoke machine, standing in the shop, singing “I Did it My Way,” for me, except he doesn’t know too many of the words. I didn’t know who I wished could see us then, but I certainly wished somebody we knew had ambled by.
In April 1983, just after The Fifth Generation was published, I was invited to a Washington meeting to help plan a film for the U.S. Pavilion at the upcoming World’s Fair in Tsukuba, Japan, to be produced by the United States Information Agency (USIA). Japan’s dramatic, and to Americans, shocking leap ahead in AI, must have a response from American AI, and a number of AI researchers were gathered, most of whom I knew.
In the air, I sniffed a strange piety. The USIA’s brochure for American businessmen had stressed competition, but my AI acquaintances were speaking somewhat sanctimoniously about cooperation instead. I laughed to myself, wondering how much “cooperation” existed in their home institutions. Did they share their DARPA grants with the starving English department?
The meeting continued over dinner. After a while, Roger Schank, then at Yale, began a vitriolic attack on Feigenbaum, not a surprise. But Harry Pople at the University of Pittsburgh began to attack Feigenbaum too, the gist being that he was overselling expert systems and raising expectations that could never be met.
Finally, I decided I’d better say a few words on behalf of my friend and co-author. This came as a surprise to Schank, who seemed embarrassed; a relief to David Hertz, sitting beside me, who said he wondered when I’d speak up. I’d been silent because I was amused, because I was loath to get into confrontations over dinner, because the spectacle was so interesting, and because I wanted to see how far they’d go.
I began by saying that Feigenbaum had a right to whatever claims he was making about expert systems, and furthermore, the expert system he praised most generously was Harry Pople’s, giving full credit to Pople everywhere he went. Pople looked a little abashed (but not as much as he should, I thought) and sputtered some inane defense. Then I turned to Schank. “You’re marketing these things. Are you marketing a fraudulent product?” “Of course not!” he cried.
I went on to say that in our book, Feigenbaum had been the cautious one, toning things down. “Ah,” said Pople, “perhaps there’s a difference between the public persona and the private persona.” It was useless to point out that nothing is more public than a published book, and I changed the subject. I happened to talk to Raj Reddy right afterwards and confided all this to him. “Professional jealousy, nothing more,” he said, confirming my own feelings.
This kind of sniping had many faces. Tony Ralston, a computer scientist at SUNY Buffalo and the editor of a short-lived journal called Abacus, wrote to ask if I’d do a profile of Feigenbaum. I declined because my friendship with Ed was so close and enduring that I couldn’t even pretend to objectivity. My views, I added, would have to wait until my old-age memoirs. “But they’ll get a hack,” I wrote in my journal, “who’ll miss the marvelous range and breadth, and listen to his jealous dwarf enemies.”
To tempt me, Ralston had sent me a review written by some nonentity who presumed to review the oeuvre of Donald Knuth, a giant of computer science, whose The Art of Computer Programming is one of the momentous scholarly achievements of the field, the multiple volumes known deferentially as Knuth: “Look it up in Knuth.” Although the review was mainly praise, the discrepancy between Knuth’s accomplishments and those of the reviewer was so wide as to be ludicrous. It made even praise of Knuth sound presumptuous. To add to my irritation, Ralston added I “ought to know” that this month’s Abacus had a somewhat harsh review of The Fifth Generation. Why ought I to know? Should I un-write the book?
Expert systems were in fact problematic. Ed Feigenbaum and his colleagues had upended AI research, pretty much devoted earlier to games, mathematics, puzzle-solving, and some attempts to model modest instances of human problem solving. They’d insisted on putting real-world knowledge into such a system, so that it made decisions the same way a real-world, intelligent human expert might. By taking such a step, Feigenbaum himself left the kind of AI he’d been trained in, cognitive psychology and the modeling of human memory, and moved to what would later be categorized as a knowledge-intensive rational agent. He was less interested in emulating human cognitive behavior than in producing a system that would perform as well as, even better than, human experts in the real world. He’d had some stunning successes and was plenty excited about this brainchild of his.
Expert systems were difficult to build and almost as difficult to maintain. Each application seemed to draw smart young graduate students into pushing the field of AI into one more app, instead of pushing it forward, beyond this relatively fragile early paradigm. It was a path of easy payoff and least resistance.
But as I heard Feigenbaum’s colleagues blame him for a waste of brainpower and other resources, I was puzzled. If a better path lay elsewhere, why weren’t they taking it? Or more important, guiding their graduate students to take it? Why weren’t they persuading the funding agencies that AI had a different, better future? Why blame businesses for taking up this technology, which seemed on its face useful to business? Why blame Feigenbaum and fail to take responsibility for moving the field forward themselves?
Eventually, the Japanese Fifth Generation did not reach the ambitious goals it originally set. Chapter 18 offers some possible reasons why. But the Fifth Generation’s major accompishment was not negligible: it trained a generation of young scientists in the field. Thus Japanese AI research continues apace. Yet an unfortunate chasm has appeared between Western and Japanese research. Ed Feigenbaum thinks the biggest wedge is the language barrier. Westerners fail to learn Japanese, and Japanese scientists are less and less interested in learning English. Perhaps neither side thinks it can learn much from the other. This is a great pity, and I hope the chasm closes.
Time passed. I wrote another book, The Universal Machine, and was mulling a new book on how people adapted to great changes in their lives. Feigenbaum stepped into this mulling process and suggested we collaborate again. This time he wanted to include Penny Nii because the book would be about expert systems in business, and she knew as much about them as anyone. By then, Joe and I had moved to Princeton and bought a drafty old house in desperate need of new windows on two floors. We faced an outlay of tens of thousands of dollars, and we were already financially stretched with the house itself. This new book offered me chance to see if I could write a good book about business and certainly a chance to earn some much-needed money to make the Princeton house habitable in the winter. Above all, because The Fifth Generation had been so much fun, I couldn’t see any problems.
Expert systems had burgeoned in universities and in firms. Hundreds of them now embodied the knowledge of experts in science, business, medicine, and many other fields. Each system’s knowledge was explicit and offered ways of communicating, exchanging, and improving on it. The disadvantages of that sudden growth were a premature commercialization of AI (fragile systems that were difficult to maintain) and, as some argued, a lateral instead of a forward push in research, as each little piece of expertise was shoehorned into an expert system and earned somebody, somewhere, a masters’ degree.
Feigenbaum, Nii, and I began work on the book with a marathon round of interviews in firms that were using expert systems. One interview took place in the Houston offices of Schlumberger, the oil field services firm.
March 12, 1987:
At a Schlumberger installation in Houston to interview. Praise the lord, Schlumberger is run by the French, so lunch is the first decent meal in Texas. Simple but excellent. Lunch is also info-gathering for me: I listen to the Houstonians (even as I once listened to the Japanese) gather pearls from Ed The Guru. The truth of the matter is, The Guru is pretty damn smart, so they’re correct to regard him with reverence. That combination of native intelligence and sheer decades of experience is priceless. I think of Ed as on top of things in every sense. Now, why can’t I get excited about expert systems? Partly they lack the plethora of grand ideas I’m used to. Partly it’s a writer’s problem, trying to make the stories appear different when they’re basically all alike. But that’s an old storyteller’s problem.
Buried in that book we eventually wrote, The Rise of the Expert Company, was a subtle prophecy. Ed Mahler, a senior chemist at DuPont, was high on expert systems and had formed a group to introduce them in DuPont’s various branches. But Mahler thought you didn’t need fancy programs, fancy machines, or knowledge engineers to turn it all into code. You gave a chemist (or another scientist) some means of interrogating himself, you gave him a laptop, and you made him his own knowledge engineer. That’s pretty much what happened, upending many a business plan that proposed to manufacture special-purpose machines for AI or to train knowledge engineers to go across disciplines, like business consultants, evoking knowledge from human heads to be cast into computer programs. Later, machines would begin to develop their own expertise, learning from the environment, whether they did it supervised by an expert or independently. The idea animating expert systems became fundamental, an implicit part of a much larger genre of programs called knowledge-intensive rational systems, whether that knowledge came from huge data sets (comprising aggregate human behavior such as your Facebook loyalties, your response to digital ads, and your smartphone conversations) or directly from human heads.
Ed and I kept talking to each other as always, but I began to take a rest from AI. The field seemed to have passed from revolutionary to normal science, in the sense Thomas Kuhn means in The Structure of Scientific Revolutions. For me it was no longer as interesting. Change would come again, but until then, I found other things to explore.