Slashdot stories can be listened to in audio form via an RSS feed, as read by our own robotic overlord.

 



Forgot your password?
typodupeerror
AI

Judea Pearl Wins Turing Award 38

Posted by samzenpus
from the a-hearty-handshake dept.
alphadogg writes "Judea Pearl, a longtime UCLA professor whose work on artificial intelligence laid the foundation for such inventions as the iPhone's Siri speech recognition technology and Google's driverless cars, has been named the 2011 ACM Turing Award winner. The annual Association for Computing Machinery A.M. Turing Award, sometimes called the 'Nobel Prize in Computing,' recognizes Pearl for his advances in probabilistic and causal reasoning. His work has enabled creation of thinking machines that can cope with uncertainty, making decisions even when answers aren't black or white."
This discussion has been archived. No new comments can be posted.

Judea Pearl Wins Turing Award

Comments Filter:
  • by epine (68316) on Thursday March 15, 2012 @07:10PM (#39372195)

    I spent a day poking into his work around xmas time. I recall a long document with artistic illustrations on the nature of causality. One of those documents where every step seems almost too trivial to notice, until you discover you've reached the conclusion and haven't gaining any understanding. Well, some of those small points must be hiding more than they first appear. When the penny did finally drop, I felt his presentation was perhaps obscured by contrived simplicity.

    In my own thinking since, I've realized that causality is not what we think it is. In a sufficiently complex system, causality as we wish to know it ceases to manifest itself. Stephen Jay Gould tried to get this point across in The Mismeasure of Man [wikipedia.org] without entirely realizing it. You come away from that book mainly with a profound sense of how much he hates the Ascent of Man iconography at more of a gut level than a cerebral one.

    The second fallacy is "ranking", which is the "propensity for ordering complex variation as a gradual ascending scale."

    Implied in that ascending order is the human conceit of because-ness. When we tire of because-ness on the grand scale, we flit to worrying about the wings of butterflies. Both at the same time? Wow, your mind is more flexible than mine.

    What I realized in my subsequent thinking (after my Pearl diving) is that randomness and causality share similar housing. Pseudo-random numbers are useful precisely because they are decorrelated (to at least modest algorithmic depth) with any unwitting sequence you are likely to stumble over (but not ones maliciously prepared; to combat that you need true randomness).

    If you have a correlation between smoking and lung cancer, how do you show that smoking is causative? You need to perform some intervention which does not by itself explain the answer. Otherwise you might just as well conclude that the smoking of special "control" cigarettes prepared by men in white coats is a lung cancer cure, and apply to the FDA for treatment status. There is no such thing as a pure control (like pure randomness). All controls are pseudo-controls. Like pseudo-random numbers, some pseudo-controls are damn good enough for certain purposes.

    In a really big system, such as the evolutionary history of life on earth, you'll never get a clean separation you entirely believe a priori. You'll always wonder if your exploratory manipulation itself is the smoking gun. How to prove otherwise? Well, the problem regresses. With a billion evolutionary simulators (with roughly the complexity of the local solar system) run a billion times on each of a trillion microscopic hypotheses, you might reach some stunning conclusions—if your methodology section doesn't trigger black hole formation. It's not so much that causality ceases to exist, it's more the case that you'll just never get there in reductive purity.

    Recently there is the Taubes position on fructose (consumed in excess quantity) being a principle vector in metabolic syndrome. How do you prove this? Test diets with and without? Do they taste exactly the same? Are the digestive mechanics exactly the same? Can you slip the change in the lives of your subjects without them any the wiser? How did select a test group and maintain contact with it while none of them were any the wiser? Didn't half of them drift off to lives in new cities? And none of them heard about the fructose hypothesis on the radio and made subconscious life changes.

    This is where gene knock-out experiments in mice are the bomb. We presume that mice really don't know about experimental protocol. Douglas Adams wants to know why. The Ascent of Man is always hiding in causality arguments somewhere, as any acute satirist realizes.

    [snit]I'm pretty sure I spelled Pearl correctly (not Perle). If I succeeded, was the cause of this the Slashdot design where not even the title of the article I'm comment

  • Re:Fuzzy logic? (Score:5, Informative)

    by bmacs27 (1314285) on Thursday March 15, 2012 @07:15PM (#39372229)

    Pearl was responsible for the development of the concept of Bayesian Networks. The concept was popularized and developed in Probabilistic Reasoning in Intelligent Systems in the late eighties. In 2001, he followed it up with Causality to respond to justified critiques from the field of epistemology. That field (which produced many of the fuzzy logic motivation) has historically been reluctant to accept Bayesian interpretations of probability, although I always thought that there was very little "Bayesian" about Bayesian Networks beyond the name.

    In summary, the networks are Markovian Directed Acyclic Graphs with probabilistic weights between the nodes. He further formalized methods of "Inferred Causation" from data which infers such useful concepts as "latent common causes," particularly in the second book. Both books are good reads. I highly recommend them.

Power corrupts. And atomic power corrupts atomically.

Working...