Hi Gary,
This is a topic near and dear to me, and one I am very actively
investigating (and using) personally (mostly with ChatGPT 4-o1, but also
the latest version of Grok). My first observation, granted based on my
sample of one, is that abductive reasoning in a Peircean sense is
lacking with current LLMs (large language models), as is true for all
general ML or AI approaches. Machine learning and deep learning have
been mostly an inductive process IMO. A major gap I have seen for quite
some time has been the lack of abductive reasoning in most ML and AI
activities of recent vintage.
This assertion is most evident in the lack of "new" hypothesis
generation by these systems, the critical discriminator that you
correctly point out from Peirce. One can prompt these new chat AIs with
new hypotheses, and in that form, they are very helpful and useful. It
is for these reasons that I tend to treat current chat AIs as dedicated
research assistants: able to provide very useful background legwork,
including some answers that stimulate further questions and thoughts,
often in a rapid fire give-and-take manner, but ones that are not
creative in and of themselves aside from making some non-evident
connections.
I believe that better matching of current chat AIs with Peirce's
thinking (esp abductive reasoning as he defined) is a particularly rich
vein for next generation stuff. Lastly, my own personal view is that the
current state of the art is not "dangerous", but we are also seeing very
rapid increases of what Ilya Sutskever
<https://en.wikipedia.org/wiki/Ilya_Sutskever> calls
"superintelligence", the speed of which is pretty breathtaking. We may
be close to tapping out on this current phase with most Internet content
already captured for training, but like with LLMs, there are certainly
new innovations not yet foreseen that may continue to maintain this
Moore's law <https://en.wikipedia.org/wiki/Moore%27s_law>-like pace of
improvements.
Best, Mike
On 12/17/2024 6:00 PM, Gary Richmond wrote:
List,
In a brief article, "How Does A.I. Think? Here’s One Theory" in the
New York Times today, Peter Coy, after noting that "Computer
scientists are continually surprised by the creativity displayed by
new generations of A.I.," comments on one hypothesis that might help
explain that 'creativity', namely, that AI is using abduction in its
machine reasoning. He writes:
One hypothesis for how large language models such as o1 think is
that they use what logicians call abduction, or abductive
reasoning. Deduction is reasoning from general laws to specific
conclusions. Induction is the opposite, reasoning from the
specific to the general.
Abduction isn’t as well known, but it’s common in daily life, not
to mention possibly inside A.I. It’s inferring the most likely
explanation for a given observation. Unlike deduction, which is a
straightforward procedure, and induction, which can be purely
statistical, abduction requires creativity.
The planet Neptune was discovered through abductive reasoning,
when two astronomers independently hypothesized that its existence
was the most likely explanation for perturbations in the orbit of
its inner neighbor, Uranus. Abduction is also the thought process
jurors often use when they decide if a defendant is guilty beyond
a reasonable doubt.
Yet Peirce argues in the 1903 Lectures on Pragmatism that only
abduction "introduces any new idea" into a scientific inquiry:
" Abduction is the process of forming an explanatory hypothesis.
It is the only logical operation which introduces any new idea;
for induction does nothing but determine a value, and deduction
merely evolves the necessary consequences of a pure hypothesis."
I had always thought of abduction as the unique domain of the
individual scientist, the creative genius (say, Newton or Einstein)
who, fully versed in the most important relevant findings in his
field, retroductively connects those pieces of scientific information
to posit a testable hypothesis concerning an unresolved question in
science.
But it makes sense that an AI program employing large data bases might
indeed be able to 'scan' those huge, multitudinous bases, connect the
salient information, and posit an hypothesis (or some other abductive
idea).
Any thoughts on this? For example: Is it potentially a valuable
feature and power of AI and, thus, for us (the use of AI in
medical research would tend to support this view)? Is it a potential
danger to us (some AI programs have been seen to lie, to 'hide' some
findings, etc.; might this get out of control)? If AI can create
testable hypotheses, is the role of the 'creative' scientist jeopardized?
Best,'
Gary R
_ _ _ _ _ _ _ _ _ _
ARISBE: THE PEIRCE GATEWAY is now at
https://cspeirce.com and, just as well, at
https://www.cspeirce.com . It'll take a while to repair / update all the links!
► PEIRCE-L subscribers: Click on "Reply List" or "Reply All" to REPLY ON
PEIRCE-L to this message. PEIRCE-L posts should go [email protected] .
► To UNSUBSCRIBE, send a message NOT to PEIRCE-L but [email protected] with
UNSUBSCRIBE PEIRCE-L in the SUBJECT LINE of the message and nothing in the
body. More athttps://list.iupui.edu/sympa/help/user-signoff.html .
► PEIRCE-L is owned by THE PEIRCE GROUP; moderated by Gary Richmond; and
co-managed by him and Ben Udell.
--
__________________________________________
Michael K. Bergman
319.621.5225
http://mkbergman.com
http://www.linkedin.com/in/mkbergman
__________________________________________
_ _ _ _ _ _ _ _ _ _
ARISBE: THE PEIRCE GATEWAY is now at
https://cspeirce.com and, just as well, at
https://www.cspeirce.com . It'll take a while to repair / update all the links!
► 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 UNSUBSCRIBE PEIRCE-L in the SUBJECT LINE of the message and nothing in the
body. More at https://list.iupui.edu/sympa/help/user-signoff.html .
► PEIRCE-L is owned by THE PEIRCE GROUP; moderated by Gary Richmond; and
co-managed by him and Ben Udell.
