Dear Judea,
One further note -- to avert some possible confusion. In my original
message I did make a supposition that was plainly false -- and that I only
later realized was false. For some reason I assumed that both relative
frequency statements and statements "having the structure of relative
frequency statements" cannot reflect the sort of principles one finds in
expressions such as F = MA. I was clearly wrong about that; I now think --
and I should have thought all along that -- both subjective probabilities
and statements about the expected relative frequency of something or other
can be the product of principles such as F = MA. So my objection to
"Pearl-style" inferential reasoning has nothing to do with its supposedly
simple or simplistic character. My objection rests on other grounds. But I
hasten to add: Pearl-style causal reasoning is a powerful tool! What makes
me uneasy about some of the arguments of yours that I have seen is that I
think I see in them an attempt to "squash" _all_ valid inferential
reasoning into "Pearl-style causal patterns." For example, am I correct in
suspecting that you would reject the validity or value of "ancillary
inference netwworks" of the sort that I was trying to describe?
Best regards,
Peter T.
P.S. Starting tomorrow I will be incommunicado for about a week.
Judea Pearl wrote:
> Dear Peter,
>
> I read your mesg of last Monday and, unfortunately,
> I could not understand most of what you were trying to
> express. I must report, however, that if I understood correctly
> what you said about "Pearl-style diagrams", then
> your statement is regretfully false.
>
> Here is what you wrote:
> 3. The questions in par. 2 above bring me to an alternative to the sort
> of picture of inference that I have sketched above: Pearl-style causal
> reasoning. Isn't it the case that Pearl-style causal reasoning --
> despite being avowedly _"causal"_ -- isn't it the case that Pearl-style
> causal reasoning fails to "capture" the complexity of the sort of
> (jumbled) causal reasoning that is involved in my last example? Pearl's
> diagrams look "causal" to me _only_ in the sense that they posit that
> all events have an antecedent cause or causes. In every other respect,
> his diagrams make the links between events or propositions look like
> relative frequencies rather than like the sort of complex relationships
> that one finds in even simple "real" causal hypotheses such as F = MA.
>
> My response:
> I am not sure what need be "captured" in what you call
> "the complexity of the sort of (jumbled) causal reasoning
> that is involved in my last example? "
> but I am very very sure that the diagrams I am using
> are causal in every respect, that the
> the links between events or propositions have little to do
> with relative frequencies, and that
> your "real" causal hypotheses such as F = MA
> can be incorporated properly and formally into those
> diagrams.
>
> I would encourage you to read some of the latest literature
> on causal diagrams, eg., Proceedings of UAI 1990-1998,
> Biometrika, Dec. 1995, AIJ 1997, etc. or have a glimpse at my
> web site http://bayes.cs.ucla.edu/jp_home.html
> You will be pleased to discover there methods of
> formalizing, processing and learning many features
> of causal relationships , including effects of intervetions,
> counterfactuals, confounding, causal explanations, etc.
> including even measures of legal responsibility and
> "actual causation" which are of interest to
> legal scholars.
>
> I hope you take the time to read this literature,
> Enjoy,
> =========Judea
-- *********************************************************
Tillers' Dynamic Evidence Site
--with "frames":
http://www.tiac.net/users/tillers/index.html
--sans "frames":
http://www.tiac.net/users/tillers/home.html
**********************************************************
Peter Tillers, Professor of Law
Benjamin N. Cardozo School of Law, Yeshiva University
55 Fifth Avenue, New York, New York 10003, U.S.A.
(212) 790-0334; FAX (212) 790-0205
--------------F47634720C43D4E9491C995E
Content-Type: text/html; charset=us-ascii
Content-Transfer-Encoding: 7bit
Dear Judea,
One further note -- to avert some possible confusion. In my original message I did make a supposition that was plainly false -- and that I only later realized was false. For some reason I assumed that both relative frequency statements and statements "having the structure of relative frequency statements" cannot reflect the sort of principles one finds in expressions such as F = MA. I was clearly wrong about that; I now think -- and I should have thought all along that -- both subjective probabilities and statements about the expected relative frequency of something or other can be the product of principles such as F = MA. So my objection to "Pearl-style" inferential reasoning has nothing to do with its supposedly simple or simplistic character. My objection rests on other grounds. But I hasten to add: Pearl-style causal reasoning is a powerful tool! What makes me uneasy about some of the arguments of yours that I have seen is that I think I see in them an attempt to "squash" _all_ valid inferential reasoning into "Pearl-style causal patterns." For example, am I correct in suspecting that you would reject the validity or value of "ancillary inference netwworks" of the sort that I was trying to describe?
Best regards,
Peter T.P.S. Starting tomorrow I will be incommunicado for about a week.
Judea Pearl wrote:
Dear Peter,-- *********************************************************I read your mesg of last Monday and, unfortunately,
I could not understand most of what you were trying to
express. I must report, however, that if I understood correctly
what you said about "Pearl-style diagrams", then
your statement is regretfully false.Here is what you wrote:
3. The questions in par. 2 above bring me to an alternative to the sort
of picture of inference that I have sketched above: Pearl-style causal
reasoning. Isn't it the case that Pearl-style causal reasoning --
despite being avowedly _"causal"_ -- isn't it the case that Pearl-style
causal reasoning fails to "capture" the complexity of the sort of
(jumbled) causal reasoning that is involved in my last example? Pearl's
diagrams look "causal" to me _only_ in the sense that they posit that
all events have an antecedent cause or causes. In every other respect,
his diagrams make the links between events or propositions look like
relative frequencies rather than like the sort of complex relationships
that one finds in even simple "real" causal hypotheses such as F = MA.My response:
I am not sure what need be "captured" in what you call
"the complexity of the sort of (jumbled) causal reasoning
that is involved in my last example? "
but I am very very sure that the diagrams I am using
are causal in every respect, that the
the links between events or propositions have little to do
with relative frequencies, and that
your "real" causal hypotheses such as F = MA
can be incorporated properly and formally into those
diagrams.I would encourage you to read some of the latest literature
on causal diagrams, eg., Proceedings of UAI 1990-1998,
Biometrika, Dec. 1995, AIJ 1997, etc. or have a glimpse at my
web site http://bayes.cs.ucla.edu/jp_home.html
You will be pleased to discover there methods of
formalizing, processing and learning many features
of causal relationships , including effects of intervetions,
counterfactuals, confounding, causal explanations, etc.
including even measures of legal responsibility and
"actual causation" which are of interest to
legal scholars.I hope you take the time to read this literature,
Enjoy,
=========Judea
Tillers' Dynamic Evidence Site**********************************************************--with "frames": http://www.tiac.net/users/tillers/index.html
--sans "frames": http://www.tiac.net/users/tillers/home.html