Simply in terms of nostalgia, this 1985 video called “Knowledge Engineering: Artificial Intelligence Research at the Stanford Heuristic Programming Project” from the Stanford archives is charming right down to its Tron-like digital soundtrack.

But it’s also really interesting if you care about the way we’ve thought about knowledge. The Stanford Heuristic Programming Project under Edward Feigenbaum did groundbreaking work in how computers represent knowledge, emphasizing the content and not just the rules. (Here is a 1980 article about the Project and its projects.)

And then at the 8:50 mark, it expresses optimism that an expert system would be able to represent not only every atom of proteins but how they fold.

Little could it have been predicted that protein folding even 30 years later would be better recognized by the human brain than by computers, and that humans playing a game — Fold.It — would produce useful results.

It’s certainly the case that we have expert systems all over the place now, from Google Maps to the Nest thermostat. But we also see another type of expert system that was essentially unpredictable in 1985. One might think that the domain of computer programming would be susceptible to being represented in an expert system because it is governed by a finite set of perfectly knowable rules, unlike the fields the Stanford project was investigating. And there are of course expert systems for programming. But where do the experts actually go when they have a problem? To StackOverflow where other human beings can make suggestions and iterate on their solutions. One could argue that at this point StackOverflow is the most successful “expert system” for computer programming in that it is the computer-based place most likely to give you an answer to a question. But it does not look much like what the Stanford project had in mind, for how could even Edward Feigenbaum have predicted what human beings can and would do if connected at scale?

(Here’s an excellent interview with Feigenbaum.)

Categories: science, too big to know dw