You're welcome, Danko.
I confess that I know little about cybernetics, Bayesian probability,
etc. - I don't have strong opinions about them. If I were to pick
something Bayesian in spirit that might help with abductive inference,
it would be some notion of subjective priors about feasibility and
optimality (reminiscent of instinctual facility), more than subjective
priors about probability (reminiscent of Peircean verisimilitude,
likeness of inductive conclusion to premissual samples). I don't know
whether the idea of subjective priors about feasibles and optima has
been worked out mathematically, but the field in which it would appear
would be that of 'inverse optimization' or 'inverse variations', more or
less the same thing under various names. Still, such a field seems
essentially inductive rather than abductive, since it involves adjusting
claims about constraints and optimization rules of a totality on the
basis of samples.
So in the end it's back to the idea of developing an instinct (or
'intuition' as we often say; Peirce denied the idea of intuition, but
that was in another sense, that of cognition based in no inference
conscious or unconscious). Peirce made an opposition between instinctual
plausibility and reasoned subjective probability, and one could make a
similar opposition between instinctual plausibility and reasoned
subjective feasibility/optimality. Well, I have to wave my hands here; I
don't know whether "reasoned subjective feasibility/optimality" been
worked out mathematically.
Peirce's emphasis on instinct (and maybe on infantile training, whereby
traditions are quasi-hardwired into the young), is rather different from
recent decades' free talk of 'developing an intuition' as if just
anybody from any educational background or sphere of experience could
equally well become a good guesser in just any field. I suspect that
Peirce was on to something there, only I would look for the effects of
infantile training before declaring that it all comes from inherited
instinct (note also that Peirce was Lamarckian, and in that viewpoint
the idea of inherited instinct was less remote from that of personal
learning than it came to seem during the 20th Century).
I have read your links, and I confess that it's mostly over my head.
Best, Ben
On 3/26/2015 6:07 PM, Danko Nikolic wrote:
Nice. Thanks Benjamin. This is very informative.
I also feel like it is better to look for "instinctual plausibility,
instinctual simplicity, naturalness, facility" than for probability.
One comment: Practopoiesis is supposed to give A.I. researchers at
least an idea in which direction to look for an artificial
intelligence design that can abduce well.
To this end, I wrote this (although I don't use the term abduction in
this text you can read "T3" as "abduction-capable"):
https://www.singularityweblog.com/practopoiesis/
Best,
Danko
On 26/03/15 22:21, Benjamin Udell wrote:
Danko, list,
To what Jon said, I would add that, as you may already know, Peirce
regarded Bayesian subjective probabilities and subjective priors as
treacherous. Peirce criticizes the idea of subjective probability in
a number of places. I don't know what researchers in abductive
inference think about it, but Peirce himself rejected the idea that
subjective probability is intrinsically valuable for an explanatory
hypothesis; he looked instead especially to instinctual plausibility,
instinctual simplicity, naturalness, facility. He also mentioned the
worth of objective probability of a hypothesis, though he thought
that the instinctual impression should avoid being influenced by
awareness of the hypothesis's objective probability.
Good texts by Peirce to read on abductive inference, besides his
discussion in his 1903 lectures on pragmatism, include "On the Logic
of Drawing Ancient History from Documents" (1901), /Essential Peirce/
v. 2, wherein Peirce gives one of his most systematic accounts of
abductive inference. He discusses instinctual plausibility,
instinctual naturalness (Galileo's "natural light of reason") of
abductive inference (or retroduction as he now called it) in "A
Neglected Argument for the Reality of God" (1908)
http://www.gnusystems.ca/CSPgod.htm#na0 in *Section III* and *Section
IV*.
In at least one writing, Peirce includes infantile training under
instinct. Still, the idea of instinct doesn't seem to tell the A.I.
researcher much about how to design an artificial intelligence that
can abduce well. I don't know how researchers have dealt with this.
Best, Ben
On 3/26/2015 4:24 PM, Jon Awbrey wrote:
Danko, List,
Bates' Rule is a mathematical theorem, that is, a deductive transformation that
can at best preserve the information in the data. Thus it is explicative where
abduction and induction are ampliative.
This is an old controversy that Peirce had with, I think it was the
Neyman–Pearson school of thought?
There used to be a lot of literature and some understanding of this point among
AI folk, but that may be forgotten now. That happens periodically …
Regards,
Jon
http://inquiryintoinquiry.com
On Mar 26, 2015, at 1:30 PM, Danko Nikolic<[email protected]> wrote:
Dear,
There was one more question that bugged me while writing the paper on
practopoiesis: There has been a lot of work on Bayesian inference in the brain.
So, my fear was that people who worked on Bayesian aspects of brain computation
would argue that all the issues regarding logical abduction have been addressed
through Bayesian-related work.
First, I have to say that my fear was not really grounded and for a strange
reason. It turned out that all the experts on Bayesian inference who I talked
to have never heard of logical abduction. That was kind of sad, but still did
not solve the problem.
My intuition is that abduction is much more than Bayesian inference, but I have
hard time defending this stance.
Can anybody tell me more about that relation? If one shows that a neural
circuit performs Bayesian inference, has it been automatically shown that the
circuit can perform logical abduction? I guess not. But I would like to know
more about that.
The way I treated the issue in the paper was that I discussed primarily abduction and
then briefly mentioned Bayes at the end as "related". I am not sure whether I
could have done a better job.
Thank you very much.
Best regards,
Danko
--
Prof. Dr. Danko Nikolić
Web:
http://www.danko-nikolic.com
Mail address 1:
Department of Neurophysiology
Max Planck Institut for Brain Research
Deutschordenstr. 46
60528 Frankfurt am Main
GERMANY
Mail address 2:
Frankfurt Institute for Advanced Studies
Wolfgang Goethe University
Ruth-Moufang-Str. 1
60433 Frankfurt am Main
GERMANY
----------------------------
Office: (..49-69) 96769-736
Lab: (..49-69) 96769-209
Fax: (..49-69) 96769-327
[email protected]
----------------------------
--
Prof. Dr. Danko Nikolić
Web:
http://www.danko-nikolic.com
Mail address 1:
Department of Neurophysiology
Max Planck Institut for Brain Research
Deutschordenstr. 46
60528 Frankfurt am Main
GERMANY
Mail address 2:
Frankfurt Institute for Advanced Studies
Wolfgang Goethe University
Ruth-Moufang-Str. 1
60433 Frankfurt am Main
GERMANY
----------------------------
Office: (..49-69) 96769-736
Lab: (..49-69) 96769-209
Fax: (..49-69) 96769-327
[email protected]
----------------------------
-----------------------------
PEIRCE-L subscribers: Click on "Reply List" or "Reply All" to REPLY ON PEIRCE-L
to this message. PEIRCE-L posts should go to [email protected] . To
UNSUBSCRIBE, send a message not to PEIRCE-L but to [email protected] with the
line "UNSubscribe PEIRCE-L" in the BODY of the message. More at
http://www.cspeirce.com/peirce-l/peirce-l.htm .
