Posted by
David Eric Smith on
Nov 13, 2007; 11:39pm
URL: http://friam.383.s1.nabble.com/FRIAM-and-causality-tp525252p525261.html
Thank you Roger and Frank and Tom,
I hadn't looked up any of this material on Pearl, and really need to
learn it. Indeed, I think it was at Nihat's talk that I saw it the
first time, as you suggested, Frank.
Nick, two things, which because I can't do them justice may be too
much like stoning, but at least won't be like being ignored:
I would really push to change your mind about probabilities. This
may annoy you, but I think it is right to say that a probability is a
number that, within the context of an appropriate set of other
numbers, obeys the axioms of probability theory. Here I am again
advocating the development in the first couple chapters of Jaynes's
Probability Theory: the logic of science. These numbers need not
refer to frequencies, past, present, or otherwise. They could be
components in a gambler's algorithm for taking future deterministic
actions which are functions of evidence. The gambler need not be
competent, or even sane, but I am assuming for the purpose of this
example that he is capable of producing or in some other manner
acquiring numbers that do satisfy the axioms of probability theory.
They can be representations of "plausibility" as Jaynes uses the
term, which again are more in the nature of components in a cognitive
algorithm than representations of sample outcomes. I push this point
for what I believe to be the same reason as you bring probabilities
into the discussion of causality. There are lots of kinds of logical
constructions which can profitably be treated as "things" in a
syntactic sense, and probabilities can appear among the list of
"things" so treated. The relation of algorithms to experience is a
somewhat different matter.
A second point, to which I will do even less justice. You brush up
against a question that has been of interest to several of us, and
which I do not think has been well treated. In talking about the
causer of an event as the one who you can blame for it, you bring in
the role of narrative, as a particular kind of context-setting
framework. It is an intriguing question in all those areas that go
away from easy decomposability toward more context dependence and
potentially more historical contingency, what is the right model for
the source of confidence. For historians, in their own words, that
source is narrative. I think the question of how narrative as a
source for confidence relates to logic as a source begs for a
Bayesian treatment, in which we ask structural questions about the
nature of forward models in relation to priors.
All best,
Eric
On Nov 12, 2007, at 9:10 PM, Nicholas Thompson wrote:
> Friends,
>
> Darn it! I cant get anybody to tangle with the fundamental thing I am
> saying here. Anytime we embody something that is true of the
> aggregate of
> observables in a single unobservable case, we are committing a
> fallacy.
> The locus classicus of this fallacy is mental causation, where we
> hypostize
> our awareness of a pattern of a person's behavior and lodge it in an
> unobservable event with in his "mind" (or brain, it really doesn't
> matter).
> Here the problem is at its most obscene, but it lurks elsewhere.
>
> I am puzzling here how to put the point in the MOST ANNOYING WAY
> POSSIBLE,
> so that SOMEBODY will feel obligated to address it.
>
> Let's try this: To say that a probability attaches to an event at an
> instant is to commit this fallacy. What we know is a past relative
> frequency of relevant conditions and relevant consequences.
> Instantaneous
> probability is a fiction.
>
> OK. so perhaps it's a heuristic fiction. Well, not if it directs
> attention away from the evaluation of our knowledge, concerning the
> relative frequency of events. Perhaps another way make this
> point is
> that "cause" is an emergent.
>
> "Cause" is just another one of those misattributions. We saw the
> hammer
> hit the nail, but to say that the Hammer caused the nail to
> penetrate the
> wood is to invent an unobservable, an instantaneous "cause".
>
> Like most people, I would prefer to be stoned to death than be
> ignored.
>
> Nick
>
>
>> [Original Message]
>> From: Eric Smith <desmith at santafe.edu>
>> To: <nickthompson at earthlink.net>; The Friday Morning Applied
>> Complexity
> Coffee Group <friam at redfish.com>
>> Date: 11/12/2007 1:29:08 PM
>> Subject: Re: [FRIAM] FRIAM and causality
>>
>> Hi Nick,
>>
>> I assume you already know about the work Judea Pearl did to define a
>> notion of causality in the context of inference on Boolean networks?
>> I don't have citations on this, because I only learned about it
>> recently in someone's talk, but I gather it is fairly widely known.
>> Happily it doesn't claim to address all questions in which a given
>> kind of word appears, so it probably contributed something
>> concrete to
>> answering a single class of them.
>>
>> What is that old folk saying, said with a sigh?
>> "Always a physicist, never a philosopher."
>>
>> Best,
>>
>> Eric
>
>
>
> ============================================================
> FRIAM Applied Complexity Group listserv
> Meets Fridays 9a-11:30 at cafe at St. John's College
> lectures, archives, unsubscribe, maps at
http://www.friam.org