Remembering Lotfi Zadeh, the Inventor of Fuzzy Logic

> Still, most of Zadeh’s colleagues in the West continued to express disdain. The electrical engineer Rudolph Kálmán called fuzzy logic “a kind of scientific permissiveness.” The mathematician William Kahan dismissed it as “the cocaine of science.” But Zadeh’s idea persisted. Since 1965, that inaugural paper has accumulated nearly ninety-three thousand academic citations, according to Google Scholar.

Reminds me of the saying that academic politics are so vicious precisely because the stakes are so low.

The main problem with fuzzy logic is that there is no link from the claimed fuzzy degree-of-membership back to the real world. This means that there is no clear-cut way to build hierarchies of fuzzy logic systems, i.e., to combine fuzzy values.

In contrast, mathematical probability has a link to the real world in terms of long-run averages. The constraint that "the average or relative frequency must work out correctly" yields a calculus for combining probabilities, that we call independence, conditional probability, and Bayes' rule.

In some ways, fuzzy logic is a less powerful solution to a less-well-characterized problem.

Fuzzy logic as such didn't really go anywhere, just as predicted by actual scientists and engineers. Zadeh was widely cited, but with little real-world "impact factor" outside the context of the journals themselves.

Still, I suspect that's because the core idea was ahead of its time. Fuzzy logic attacked problems that already had conventional solutions. You don't actually need fuzzy logic to minimize the time derivatives of position when starting or stopping an elevator, or to run an electric razor, or to do constraint minimization with linear programming, or in any number of other commonly-cited applications. But if you were to pin down a member of the AlphaGo team and ask him or her exactly how the machine beat Lee Sedol, the answer you'll get is going to sound a lot like "fuzzy logic," in the sense that our machines and the algorithms they implement are reaching a point where they're no longer entirely accountable to their human programmers, or understandable by them.

That makes me think that in another 50 years Zadeh's reputation may look somewhat different than it does today.

I get that this is the New Yorker, but I think it's a stretch to say he invented fuzzy logic. MV-algebras had been around for decades when he published his paper, based on Lukasiewicz's work on real valued logic. They come up quite naturally in functional analysis.