IlyaShpitser

Karma: 7,126

IlyaShpitser 19 Aug 2009 9:13 UTC
2 points
in reply to: ChrisHibbert’s comment on: Ingredients of Timeless Decision Theory
Second Chris’ advice on reading Pearl.

If it helps, I am happy to help with the technical content of the book, or with general technical questions about causal inference (either over email or here).

IlyaShpitser 7 Sep 2009 19:13 UTC
3 points
on: Decision theory: Why Pearl helps reduce “could” and “would”, but still leaves us with at least three alternatives
The description of Pearl’s counterfactuals in this post isn’t entirely correct, in my opinion.

Your description of the “sunny day beachgoers” statement is describing an interventional distribution (in other words something of the form P(y|do(x))).

The key difference between interventions and counterfactuals is that the former are concerned with a single possible world (after an intervention), while the latter are concerned with multiple hypothetical worlds simultaneously.

Example of a statement about an intervention (causal effect): “I am about to take an aspirin. Will my headache go away?” This is just asking about P(headache|do(aspirin)).

Example of a counterfactual statement: “I took an aspirin an hour ago and my headache is gone. Would I still suffer a headache had I not taken an aspirin?” This is asking about a counterfactual distribution which is denoted in Pearl’s notation by P(headache (subscript) no aspirin | no headache, aspirin). Here the variable with a subscript (headache (subscript) no aspirin) corresponds to headache in the post-intervention world after do(no aspirin). This expression is a ‘true’ counterfactual since it contains a conflict between values observed in the ‘true world’ where events occurred (the values past the conditioning bar), and the ‘hypothetical world’ where we intervene.

IlyaShpitser 5 Feb 2010 6:36 UTC
2 points
in reply to: Eliezer Yudkowsky’s comment on: A problem with Timeless Decision Theory (TDT)
I don’t see how logical entailment acts as functional causal dependence in Pearl’s account of causation. Can you explain?

IlyaShpitser 6 Feb 2010 9:33 UTC
7 points
in reply to: Eliezer Yudkowsky’s comment on: A problem with Timeless Decision Theory (TDT)
Treating same inputs on duplicate functions also arises in the treatment of counterfactuals (since one duplicates the causal graph across worlds of interest). The treatment I am familiar with is systematic merges of portions of the counterfactual graph which can be proved to be the same. I don’t really understand why this issue is about logic (rather than about duplication).

What was confusing me, however, was the remark that it is possible to create causal graphs of mathematical facts (presumably with entailment functioning as a causal relationship between facts). I really don’t see how this can be done. In particular the result is highly cyclic, infinite for most interesting theories, and it is not clear how to define interventions on such graphs in a satisfactory way.

IlyaShpitser 28 Jul 2010 7:35 UTC
2 points
in reply to: cousin_it’s comment on: Metaphilosophical Mysteries
Kelly’s observation: inductive processes by necessity change their minds multiple times before arriving at the truth.

Kelly’s proposal: inductive processes ought to minimize how often they change their minds before truth is reached. (There are some subtle issues here—this proposal does not contradict “statistical efficiency” considerations, although it’s hard to see why at first glance).

IlyaShpitser 27 Aug 2010 22:51 UTC
3 points
in reply to: neq1’s comment on: Self-fulfilling correlations
Short nitpick—lots of assumptions other than ignorability can work for identifying direct effects (there is more to life than covariate adjustment). In particular, if we can agree on the causal diagram, then all sorts of crazy identification can become possible.

IlyaShpitser 21 Jan 2011 0:55 UTC
6 points
in reply to: metaweta’s comment on: Statistical Prediction Rules Out-Perform Expert Human Judgments
Bob should not become a Christian to become happier for the same reason that he should not stay away from hospitals if he’s sick (after all, sick people are a lot more likely to be in a hospital).

IlyaShpitser 18 Feb 2011 19:58 UTC
3 points
in reply to: Daniel_Burfoot’s comment on: Some Heuristics for Evaluating the Soundness of the Academic Mainstream in Unfamiliar Fields
Modern AI is an odd combination of statistics, applied math, discrete math/combinatorics and logic. My theory is the only reason AI is a subfield of Computer Science at all is founder bias (Turing).

IlyaShpitser 5 Mar 2011 19:04 UTC
6 points
in reply to: MartinB’s comment on: Rationality Quotes: March 2011
While more controlled experiments are undoubtedly a good thing, observational studies are often not useless, since one can often make a plausible argument for extracting causation from them. Sadly, the default state of causal analysis in medicine remains “use regression.”

IlyaShpitser 11 Apr 2011 1:35 UTC
14 points
in reply to: Tyrrell_McAllister’s comment on: Rationality Quotes: April 2011
Hi, you might want to consider this paper:

http://www.ssc.wisc.edu/soc/class/soc952/Wright/Wright_The%20Method%20of%20Path%20Coefficients.pdf

This paper is remarkable not only because it correctly formalizes causation in linear models using DAGs, but also that it gives a method for connecting causal and observational quantities in a way that’s still in use today. (The method itself was proposed in 1923, I believe). Edit: apparently in 1920-21, with earliest known reference apparently dating back to 1918.

Using DAGs for causality certainly predates Pearl. Identifying “randomization on X” with “dividing by P(x | pa(x))” might be implicit in fairly old papers also. Again, this idea predates Pearl.

There’s always more to the story than one insightful book.

IlyaShpitser 15 May 2011 5:56 UTC
2 points
in reply to: Vladimir_M’s comment on: Scholarship: How to Do It Efficiently
It is true that people in academia, as everywhere else, have a combination of low and high motives for what they do. The question is, on what grounds should one assume an average less-wronger is better in this respect than an average academic?

In my experience the biggest barrier in “mathy” areas is getting up to speed. Getting ideas accepted is a significantly bigger hurdle than getting ideas published. Most stuff that gets published is either not great or not useful (obscure) and often both.

edit: to clarify: I don’t see that the way one thinks about spreading lesswrong memes is different from the way one thinks about spreading “what I have published” memes. Both high and low motives are involved in both cases, and suboptimal equilibria are involved in both cases in the sense that one must think of ways of competing with other people’s memes.

IlyaShpitser 17 May 2011 13:25 UTC
4 points
in reply to: lukeprog’s comment on: Conceptual Analysis and Moral Theory
There is an entire chapter in Pearl’s Causality book devoted to the rabbit-hole of defining what ‘actual cause’ means. (Note: the definition given there doesn’t work, and there is a substantial literature discussing why and proposing fixes).

The counterargument to your post is that some seemingly fuzzy concepts actually have perfect intuitive consensus (e.g. almost everyone will classify any example as either concept X or not concept X the same way). This seems to be the case with ‘actual cause.’ As long as intuitive consensus continues to hold, the argument goes, there is hope of a concise logical description of it.

IlyaShpitser 17 May 2011 17:32 UTC
4 points
in reply to: lukeprog’s comment on: Conceptual Analysis and Moral Theory
Well, as I said, ‘actual cause’ appears to be one example. The literature is full of little causal stories where most people agree that something is an actual cause of something else in the story—or not. Concepts which have already been formalized include concepts which are both used colloquially in “everyday conversation” and precisely in physics (e.g. weight/mass).

One could argue that ‘actual cause’ is in some sense not a natural concept, but it’s still useful in the sense that formalizing the algorithm humans use to decide ‘actual cause’ problems can be useful for automating certain kinds of legal reasoning.

The Cyc project is a (probably doomed) example of a rabbit-hole project to construct an ontology of common sense. Lenat has been in that rabbit-hole for 27 years now.

IlyaShpitser 4 Jan 2012 3:40 UTC
4 points
in reply to: Bugmaster’s comment on: Rationality quotes January 2012
Note from a “sympathetic outsider”: I know you are joking, but the sorts of things like this subthread sometimes come across more creepy than funny.

IlyaShpitser 4 Feb 2012 6:15 UTC
4 points
in reply to: Stabilizer’s comment on: Rationality Quotes February 2012
With all due respect to Robin, this very thread supplies prior art for this idea :).

IlyaShpitser 24 Feb 2012 17:11 UTC
−2 points
in reply to: bungula’s comment on: Get Curious
Well, here’s a talk section of an article on a subject I know something about. This should give an idea of wikipedia’s process and what kind of content results from it:

http://en.wikipedia.org/wiki/Talk:Bayesian_network

Here’s another one:

http://en.wikipedia.org/wiki/Confounding

The very first sentence is wrong.

IlyaShpitser 11 Mar 2012 9:22 UTC
0 points
in reply to: khafra’s comment on: Causal diagrams and software engineering
In von Neumann’s defense, he may have said this before the strong negative results in computability theory. Machines cannot do lots of stuff.

IlyaShpitser 14 Mar 2012 20:57 UTC
3 points
on: The Singularity Institute’s Arrogance Problem
Hi Luke,

I think you are correct that SI has an image problem, and I agree that it’s at least partially due to academic norm violations (and partially due to the personalities involved). And partially due to the fact that out of possible social organizations, SI most readily maps to a kind of secular cult, where a charismatic leader extracts a living from his followers.

If above is seen as a problem in need of correcting then some possibilities for change include:

(a) Adopting mainstream academic norms strategically. (b) Competing in the “mainstream marketplace of ideas” by writing research grant proposals.

IlyaShpitser 15 Mar 2012 17:31 UTC
20 points
on: Cult impressions of Less Wrong/Singularity Institute
Here’s what an outsider might see:

“doomsday beliefs” (something “bad” may happen eschatologically, and we must work to prevent this): check

a gospel (The Sequences): check

vigorous assertions of untestable claims (Everett interpretation): check

a charismatic leader extracting a living from his followers: check

is sometimes called a cult: check

This is enough to make up a lot of minds, regardless of any additional distinctions you may want to make, sadly.

IlyaShpitser 18 Mar 2012 17:16 UTC
2 points
in reply to: buybuydandavis’s comment on: Causation, Probability and Objectivity
“Really? Has no one made any progress on this?”

(Interventionist) causality is not about probability, it is about responses to hypothetical interventions. Probability is just there to model uncertainty, it is not at all needed (in fact Pearl’s first definition of causal models is deterministic).

I think it is also a fair claim that “causality is in the mind,” since there does not seem to be any causality in quantum mechanics.