Dear Jim,
Thanks for starting this stimulating discussion and everyone else for 
contributing.

In my experimentation with some models I have found - like yourself and other 
people replying - that these tools may be helpful for certain "non-critical" 
tasks to, indeed, increase the efficient use of human time (but, even then, 
there might be arguments to the contrary) but may also silently introduce 
problems which may be hard to spot.

Like you, I do see the potential for these tools and I am very open to the idea 
of them helping in my work.

Going to your more general initial points about the implications of AI and 
whether we'll all soon just write prompts to perform analyses (which is the 
most stimulating part of the conversation), I can think of at least three 
factors that make that problematic in the short run.

  1.
Most users who want to perform empirical analyses do not have the in-depth 
knowledge required for spotting problems in code, let alone in what's happening 
under the hood of a ready-made AI "app"
  2.
Often existing software relies both on peer-reviewed papers and on the domain 
expertise of scientist coders who developed the tools. Essentially, most users 
will trust software partly because of the academic accomplishments of the 
people who wrote it. It is an interesting topic and we may discuss about 
whether "it is right" but, to the point, at this stage it is unclear whether 
one can say these "AI models" have "expertise" and who is "accountable" (in a 
broad sense) for what they produce
  3.
There is a non-deterministic component in the behaviour of these models as they 
are today. For instance, providing a prompt worded differently may return 
different outputs. This raises all sort of issues in terms of reproducibility, 
trustworthiness and ability to actually describe what has been done (which is 
critical for things like drafting manuscripts and going through peer review as 
we know it today).

To my understanding of these tools - which is admittedly quite limited - some 
of these issues stem from the nature itself of these models (e.g., the language 
component, the fact that they do have a context window that gets filled at some 
point, and so on). So, while I think they are very useful tools, I don't think 
they will be - in the very near future (say, one or two years) to replace 
existing software. On the longer term, maybe, who knows!

Just my two cents,
Carmelo




Carmelo Fruciano
Department of Biological, Geological and Environmental Sciences
University of Catania
https://www.fruciano.org/

________________________________
From: '[email protected]' via Morphmet <[email protected]>
Sent: Wednesday, 25 March 2026 03:07
To: Joe Felsenstein <[email protected]>
Cc: Morphmet2 <[email protected]>
Subject: Re: [MORPHMET2] Thoughts on morphometric software and AI

Right, but the trend seems to be towards decreasing the efficiency of 
software/scripts/prompts execution and increasing the efficient use of human 
time. Human time may be more valuable (or just have a rapidly decreasing 
attention span).

Oh, I did not have the foresight to save any of those old IBM 650 manuals. Too 
busy learning new stuff to think about the day when those might be fond 
memories. Of course, they are now all online (IBM 650 
Manuals<https://piercefuller.com/collect/650man/index.html>) and available for 
the day I might be feeling nostalgic. Thanks for the reminder!

Jim

F. James Rohlf
Distinguished Professor, Emeritus and Research Professor
Depts: Anthropology and Ecology & Evolution
Stony Brook University

On 3/24/2026 3:10:21 PM, Joe Felsenstein <[email protected]> wrote:

This is the first time you received an email from this sender 
([email protected]). Exercise caution when clicking links, opening attachments 
or taking further action, before validating its authenticity.


Jim noted that AI was:

"Computationally less efficient (so much faster computers are required) but 
easier for humans to use with even less technical knowledge of how computers 
actually work."

It is computationally less efficient if one has to do the AI interaction for 
each
data set.  But if it is just making Python or R code and giving you that,
then not computationally much less efficient than just doing your own
R or Python.

(Reading Jim's list of successive stages, I was
reminded of much past pain.  I didn't quite start
with the IBM 650 (instead iof CDC 1604 in 1961),
but I have saved, from Jim Crow's lab, the user
manuals for the 650.)

Joe
----
Joe Felsenstein,   [email protected]<mailto:[email protected]>, 
[email protected]<mailto:[email protected]>
Department of Genome Sciences and Department of Biology,
University of Washington, Seattle
----
PS Please do not use  [email protected]<mailto:[email protected]>.  
It is an alias
and mail systems often recognize that and think it is spam.


--
You received this message because you are subscribed to the Google Groups 
"Morphmet" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to 
[email protected]<mailto:[email protected]>.
To view this discussion visit 
https://groups.google.com/d/msgid/morphmet2/CAE-xHnX1PmsEu3jXXimdmedQT-a5DYz1vMp%2BpwLTv8Lc%2BuGNzQ%40mail.gmail.com<https://protect.checkpoint.com/v2/r01/___https://groups.google.com/d/msgid/morphmet2/CAE-xHnX1PmsEu3jXXimdmedQT-a5DYz1vMp*2BpwLTv8Lc*2BuGNzQ*40mail.gmail.com?utm_medium=email&utm_source=footer___.YzJ1OnN0b255YnJvb2s6YzpnOjg4OGZjYjBiOTA5YzY2N2ZlZjg3NGNkMWFhMGNmYTMyOjc6ZmQwNDozZGRmNmQ1YzBkM2UyMDllZGZhNWI1NzQ2OTkxMDUwZTM5MzYyZDQ1YTA0YjhmYTRhMjE0N2RhOWYxNTE3ODBjOmg6VDpG>.

--
You received this message because you are subscribed to the Google Groups 
"Morphmet" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to 
[email protected]<mailto:[email protected]>.
To view this discussion visit 
https://groups.google.com/d/msgid/morphmet2/Mailbird-10068588-9685-4373-a12c-1ff508791779%40stonybrook.edu<https://groups.google.com/d/msgid/morphmet2/Mailbird-10068588-9685-4373-a12c-1ff508791779%40stonybrook.edu?utm_medium=email&utm_source=footer>.

-- 
You received this message because you are subscribed to the Google Groups 
"Morphmet" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion visit 
https://groups.google.com/d/msgid/morphmet2/AMBPR02MB117215EB7CFC25172C6B56449F249A%40AMBPR02MB11721.eurprd02.prod.outlook.com.

Reply via email to