Humanist Discussion Group, Vol. 38, No. 201.
        Department of Digital Humanities, University of Cologne
                      Hosted by DH-Cologne
                       www.dhhumanist.org
                Submit to: humanist@dhhumanist.org




        Date: 2024-10-22 08:15:50+00:00
        From: Alan Liu <ayliu@english.ucsb.edu>
        Subject: AI for generating alt-text descriptions of images?

Does anyone have experience, recommendations, or warnings about using AI
“alt-text” generators such as https://futuristic-facteau-g1.zipwp.dev/,
https://alttext.ai/, etc. to draft alt-text for images?

I'm seeking advice especially from a scholarly digital community like
Humanist. I’m editing an essay collection that will also be published
online. This is a situation that is increasingly prevalent,
necessitating ever more work in creating alt-text for images, diagrams,
tables, etc. to make them accessible. The critical issue going forward is
not just one of labor but of the standards for "good" alt-text.

There are proliferating guides and resources for writing alt-text. But from
what I have so far learned (seeking more learning here) we're fairly early
in the evolution of theoretically and also historically well grounded
principles for alt-text forms, styles, and other conventions based on
adequately informed psychological, historical, sociocultural, cross-field,
and digitally adapted combinations of principles and practices--ones that
answer the question: what is a "good" textual description of an image that
communicates the informational, structural, affective, aesthetic, social,
and other "essential" quiddity of that image to someone who cannot perceive
it well sensorily or cognitively and needs assistive technologies for that
purpose. (The audience for alt-text is thus not just humans but a chorus of
evolving machinic technologies.)

AI alt-text generators are de facto presuming to be "good" at that mission,
but of course with unstated assumptions and parameters. So, the practical
question at the present time: does one as an editor advise contributing
authors to use the AI alt-text generators? Or is it better on functional or
ethical grounds to rely on the ad hoc, sometimes ill-trained and
underthought quick attempts of mainly non-image specialists to do the job?

P.S. In mentioning historical conventions that might bear on creating
alt-text, I am referring especially to the long tradition(s) of *ut pictura
poesis* and other visual/verbal conventions. (Kudos here to Ronald Paulson,
R.I.P. who mentored me on visual/verbal when he oversaw me in my first
faculty position at Yale in the English dept. with an appointment in the
then "British Studies Program" and an office in the Yale Center for British
Art.) It seems to me that those long humanistic traditions should come to
bear on the al-text problem.


_______________________________________________
Unsubscribe at: http://dhhumanist.org/Restricted
List posts to: humanist@dhhumanist.org
List info and archives at at: http://dhhumanist.org
Listmember interface at: http://dhhumanist.org/Restricted/
Subscribe at: http://dhhumanist.org/membership_form.php