Building a Glossary with GAI.
Ars Industrialis offers, on its site, a vocabulary. This lexical choice is interesting because vocabulary is the living language of a thought. It is the typical or atypical formulations of that language. What I browsed of this vocabulary made me want to do the same with the concepts I have mobilized, developed, sometimes invented in my notebook.
The idea matured for a while. However, I have not invented a new language, so the term “vocabulary” does not seem relevant to me at this stage of my reflections. After a year of writing, I had covered enough notions in my notebook dualite for a glossary to become not only useful but necessary. Especially since I have started experimenting with neologisms. Imperceptif, ablation attentionnelle, depossession operatoire: these terms do not appear in the dictionary. Although they are defined within the written notes, grouping these concepts is a way of guiding the reader who encounters them and not leaving them to their own interpretations.
Memory as Working Material
By using GAI to explore my notes, I used these tertiary retentions as material that I mobilized to initiate the construction of the glossary. The request was:
For my site and my notebook dualite, I want to write a glossary of terms, concepts, and neologisms that I use in my reflections. Write a list of these terms.
The result, though excellent, did not surprise me. The GAI had captured the conceptual trends of my writings, the terms that recur, by “catching” those I invest with a particular weight. I then asked it to explore my site to enrich this list.
Out of 21 proposed terms, only one was not relevant. The rest – structuring the whole thing with a table of contents and individual entries – was a quick affair for the GAI.
What This Exercise Reveals
What is interesting to share here is not the efficiency of the operation. It is what it illustrates about a practice of GAI.
I had started listing the terms that seemed relevant to me according to my own guidelines: produce a first version without AI, even if imperfect. I had listed the few terms I planned to include in this glossary, but the volume of work required to arrive at a “publishable” result was significant. As I had observed in the phenomenological experiment, the material was present in this history of published notes. I thought the machine was therefore well positioned to process it in a relevant and rapid manner.
This is not a cognitive delegation in the sense I intend in the note on operational dispossession. I did not abandon my thought to the machine. I entrusted the machine with a task of cataloging rather than conception. The distinction, I believe, is important. The GAI did not decide the meaning of my concepts. It identified their occurrences, proposed a list and a structure based on an idea I gave to the machine. Then it was I who evaluated, corrected, validated.
One could object that the procedure described here dangerously resembles the reverse of grammatization: rather than formalizing my methods before delegating, I here let the machine formalize the method. This objection is legitimate, and it allows me to specify that this grammatization by the machine is not without value. It makes it possible to observe what one has done, to name it, to give it a structure that can serve as a framework for future developments. It is a form of reflexive return on one’s own practice.
A Therapeutic Practice of the Pharmakon
I have set out, in several of my notes, the idea that GAI is a pharmakon: both remedy and poison, according to the formula taken from Derrida and Stiegler. This ambivalence is not a rhetorical metaphor. It describes a tension between what technology enables and what it does to us. The construction of this glossary is a good illustration.
The machine traversed a corpus that I would have reread with the fatigue of familiarity, a form of anesthesia of the gaze one casts on what one has written oneself. It identified recurrences and trends. In this sense, it played a role I willingly acknowledge: an amplifier of attention on material I risked no longer seeing.
But this is not enough to qualify this use as therapeutic. Therapeutic use is not about the efficiency of the result. It is about the posture in which the user should place themselves. What makes this use therapeutic is that the machine only took charge of cataloging. It used what I had already thought to reformulate the definitions.
Let us be honest, however: this line is not always easy to hold. The temptation is great to let oneself slide: from cataloging toward interpretation, to validate a formulation because it is convenient, to accept a structure because it is coherent, until the machine creates or produces for us an entire reasoning we would not have had. This is where the pharmakon reasserts its rights. The same operation, carried out with less attention, would produce the opposite effect: a dispossession disguised as efficiency.
What would make the difference is not the tool but the gesture or the attention of the one who uses it. This does not resolve the tension, but it allows, I believe, “domesticating” it. It is in this sense that I aim for a therapeutic practice: not a way of eliminating the poison but a way of not dying from it.
The Open Question
There remains a question I have not yet settled: how to maintain this glossary up to date?
Two options emerge. The first is manual: I update the glossary as I write, at the moment I forge or mobilize a concept. It is slow, it is demanding. The second is automated: I periodically perform a site scan by asking the GAI to identify new potential entries.
The first option is that of effort. It requires that I observe my own conceptual practice with sufficient attention to identify what deserves to be recorded. It is a form of observing, in the therapeutic sense I have sketched elsewhere.
The second is that of delegation. It is more efficient, but it deprives me of that moment of attention that constitutes choosing to extract a concept for inclusion in the glossary. By selecting a term, I decide that it deserves to exist in a more prominent way. I would therefore still need to keep my attention on what the machine proposes and to be able to diverge from it or make suggestions.
To be continued.