Ask HN: What's happened to fuzzy logic? 13 points by probably 2189 days ago | hide | past | web | 4 comments | favorite I've come across fuzzy logic mostly in pedagogical context these days. Of course there are still researchers and developers using fuzzy logic to develop algorithms, but has the mainstream future of AI been supplanted by machine learning, natural language processing, computer vision, robotics, etc.? Or fuzzy logic has become a part of each or some of these subdisciplines of AI?

 It's used somewhat in AI still, especially the communities going under the name "computational intelligence" (CI). In many communities it's been supplanted by various statistical approaches, though. Fuzzy logic starts from logic, and adds in non-binary distinctions. An alternate approach is to start from statistics, and add in structure: things like Markov logic networks, Bayesian graphical models, probabilistic planning, hierarchical reinforcement learning, and a dozen other such approaches. Both sets of approaches have the same high-level goal of bringing together symbolic reasoning with numerical inference, in some form.The second basket of approaches (statistics+structure) is now more popular overall than fuzzy logic I think, though it depends on what communities you're from; fuzzy logic is much more popular in engineering-flavored disciplines than in math-flavored disciplines.
 Great summary of the situation -- did not expect. Thanks!
 One technical problem with fuzzy logic, that is not suffered by statistics, is that there's no universally-accepted way of combining fuzzy logic truth values.The rules for manipulation of conditional probabilities, which statistics inherited from probability theory, allow building up complicated statistical models (i.e., complex enough to capture real-world applications) from pieces. This is what the comment by _delirium is saying.There's no such calculus for the truth values in fuzzy logic. The core problem is, what the concept of "truth value" refers to. In conventional Statistics, probabilities can be grounded in relative frequencies, and in principle measured in real experiments. The same can't be said of a fuzzy truth value.That much said, there are some people working on generalizations of probability theory to situations where relative frequencies don't make sense, and there's an overlap between the more sophisticated of the fuzzy logic theorists and this community. See, for example, http://www.sipta.org/isipta11/, or http://en.wikipedia.org/wiki/Imprecise_probability
 Thanks -- didn't realize this thread was still alive. So the showstopper is that there is no equivalent to Bayes theorem for fuzzy logic, you're saying. Great clarification.

Search: