Things Don't Make Sense Till They Make Sense to a Stupid Robot
December 22, 2013 4:59 PM Subscribe
In 2010, Judea Pearl was honored with a symposium on his work. He gave a talk on how to use causal models to evaluate counterfactuals. (He takes the first 14 minutes to thank people and reflect on his intellectual debts. Skip it if you really want to get to the meat of the talk, but it is well worth watching.)
Several other researchers gave talks at the symposium:
Thomas Dean: Graphical Models of the Primate Visual Cortex
Rina Dechter: On the Power of Belief Propagation
Hector Geffner: Judea's Approach to AI
Michael Gelfond: Combining Probabilistic and Logical Reasoning
Sander Greenland: Predictive diagrams: Backing off from Causality in DAGs
Christopher Hitchcock: Pearl's Contributions to Causality
Robert Holte: Mechanical Generation of Admissible Heuristics (26 years later)
Mike Jordan: Bayesian Nonparametric Learning
Rich Korf: Space Complexity of Combinatorial Search Algorithms
Dana Nau: Paranoia Versus Overconfidence in Imperfect-Information Games
David Poole: Probabilistic Programming Languages
Thomas Richardson: The Binary Instrumental Variable Model
Jamie Robins: Ontological Primacy of Causation vs Manipulation
Stuart Russell: The Glorious Revolution of 1988
Richard Scheines: Causal Model Search
Yoav Shoham: Cause for Celebration, Cause for Concern
Jin Tian: On Identifying Causal Effects
Ingrid Zukerman: Arguing with a Bayesian Intelligence
Several other researchers gave talks at the symposium:
Thomas Dean: Graphical Models of the Primate Visual Cortex
Rina Dechter: On the Power of Belief Propagation
Hector Geffner: Judea's Approach to AI
Michael Gelfond: Combining Probabilistic and Logical Reasoning
Sander Greenland: Predictive diagrams: Backing off from Causality in DAGs
Christopher Hitchcock: Pearl's Contributions to Causality
Robert Holte: Mechanical Generation of Admissible Heuristics (26 years later)
Mike Jordan: Bayesian Nonparametric Learning
Rich Korf: Space Complexity of Combinatorial Search Algorithms
Dana Nau: Paranoia Versus Overconfidence in Imperfect-Information Games
David Poole: Probabilistic Programming Languages
Thomas Richardson: The Binary Instrumental Variable Model
Jamie Robins: Ontological Primacy of Causation vs Manipulation
Stuart Russell: The Glorious Revolution of 1988
Richard Scheines: Causal Model Search
Yoav Shoham: Cause for Celebration, Cause for Concern
Jin Tian: On Identifying Causal Effects
Ingrid Zukerman: Arguing with a Bayesian Intelligence
Could someone gift Watson a MiFi account so we could have authoritative postings about the state of strong AI.
posted by sammyo at 5:31 PM on December 22, 2013 [2 favorites]
posted by sammyo at 5:31 PM on December 22, 2013 [2 favorites]
What's with all the STEM posts these last few days? Keep them coming.
posted by polymodus at 7:14 PM on December 22, 2013
posted by polymodus at 7:14 PM on December 22, 2013
What's with all the STEM posts these last few days? Keep them coming.
Oh, no. This was a philosophy post. ;)
posted by Jonathan Livengood at 7:22 PM on December 22, 2013 [8 favorites]
Oh, no. This was a philosophy post. ;)
posted by Jonathan Livengood at 7:22 PM on December 22, 2013 [8 favorites]
This is so cool, it's like applied philosophy. Kant spoke of a "third thing" that has to be there for knowledge of causation to be possible, something beyond just the power of logical reasoning and the material of experience, and these guys are more or less trying to build that thing.
posted by thelonius at 7:23 PM on December 22, 2013
posted by thelonius at 7:23 PM on December 22, 2013
Pearl really encapsulates the very very best of academic work, in my opinion. Not only does he do foundational work on the very toughest problems out there, working with a large group of students and collaborators, but he also puts in the effort to make difficult subjects available to wider audiences by summarizing and streamlining some seriously difficult literature into complete books.
Seriously, taking on causality from a mathematical perspective is something that few have even attempted, despite causality being foundational to nearly all human thought. And the few people that have tried, have gotten some things wrong. But I think that this is finally the right formulation, and a huge huge advance, though I'm biased because pretty much any solution I think of comes in the form of graphical models. In grad school before I read Causality I was working on a project where I was trying to model experimental setups, but without the disconnections in the graph that experimental interventions cause, which are quite unintuitive, but eventually second nature. A postdoc saw that I was struggling, never quite able to learn the proper models, double checked some the exact likelihoods that I was producing. He then sent me an email that was basically "you fool" and plopped the book on my desk. That book was an eye opener. At that stage I had been through more prob and stats classes than anyone I knew in my cohort, but this was some black magic. Causality was quite simply the boldest book that I've ever read, and it's also extremely difficult. The algorithms and the rules and the setups of variables just seem so foreign. Other subjects, like quantum mechanics, have the benefit of all the entire educational structure building the tools and foreshadowing the mathematical tools that will be used. Causality does not, and it came across to me as pure wizardry.
Pearl is not the only one who has worked on this, but by writing books for outsiders to the community, even if they are extremely difficult books, he has done as much as the people working out the most difficult theorems, because he simply told people about the amazing stuff that was going on. That doesn't happen enough.
posted by Llama-Lime at 7:38 PM on December 22, 2013 [6 favorites]
Seriously, taking on causality from a mathematical perspective is something that few have even attempted, despite causality being foundational to nearly all human thought. And the few people that have tried, have gotten some things wrong. But I think that this is finally the right formulation, and a huge huge advance, though I'm biased because pretty much any solution I think of comes in the form of graphical models. In grad school before I read Causality I was working on a project where I was trying to model experimental setups, but without the disconnections in the graph that experimental interventions cause, which are quite unintuitive, but eventually second nature. A postdoc saw that I was struggling, never quite able to learn the proper models, double checked some the exact likelihoods that I was producing. He then sent me an email that was basically "you fool" and plopped the book on my desk. That book was an eye opener. At that stage I had been through more prob and stats classes than anyone I knew in my cohort, but this was some black magic. Causality was quite simply the boldest book that I've ever read, and it's also extremely difficult. The algorithms and the rules and the setups of variables just seem so foreign. Other subjects, like quantum mechanics, have the benefit of all the entire educational structure building the tools and foreshadowing the mathematical tools that will be used. Causality does not, and it came across to me as pure wizardry.
Pearl is not the only one who has worked on this, but by writing books for outsiders to the community, even if they are extremely difficult books, he has done as much as the people working out the most difficult theorems, because he simply told people about the amazing stuff that was going on. That doesn't happen enough.
posted by Llama-Lime at 7:38 PM on December 22, 2013 [6 favorites]
Pearl, his wife and family also founded the Daniel Pearl Foundation, in memory and honor of their son, a Wall Street Journal reporter who was kidnapped and murdered by terrorists in Pakistan in 2002.
Wow
posted by Golden Eternity at 9:19 PM on December 22, 2013 [2 favorites]
Wow
posted by Golden Eternity at 9:19 PM on December 22, 2013 [2 favorites]
Oh, no. This was a philosophy post. ;)
Oh my, what possessive language, for a philosopher…
posted by polymodus at 9:57 PM on December 22, 2013
Oh my, what possessive language, for a philosopher…
posted by polymodus at 9:57 PM on December 22, 2013
« Older Parents recreate movie scenes with their baby and... | What to say at the holiday table Newer »
This thread has been archived and is closed to new comments
posted by ZenMasterThis at 5:25 PM on December 22, 2013