Event Timeslots (1)

Thursday
6:00 pm - 8:00 pm
Large language models are usually judged by a standard they were never built to meet. We ask whether they know things, whether they reason, whether they tell the truth, and we call their characteristic failures hallucinations. This course starts elsewhere, from the suggestion that these systems are engines of confabulation, producing fluent and plausible narrative under the pressure of a prompt. That opens a question. How far is such generation constrained, by training, by feedback, by what a system can retrieve, and at what point might constrained confabulation look less like the opposite of knowledge than like one of the ways knowledge is arrived at, in us as much as in them? Read this way, the peculiar successes of these systems become as interesting as their failures, and a further question comes into view: what happens to us when we write, read and think alongside them. The seminar follows that shift outwards, into the practices where GenAI is now embedded. We begin with dialogical AI and the confabulation thesis as background, then work through four connected problems. First, authorship: what becomes of involvement, credit and oversight when substantial portions of a text are generated rather than written, and what kind of responsibility survives the delegation. Second, interpretation: the suggestion that these systems are better understood as hermeneutic instruments than as knowledge bases, and that using them well is a matter of interpretative skill rather than prompt technique. Third, cognitive ecology: the move from individual use to populations and practices, and the question of what a research culture, a classroom or a profession becomes once interpretative machines are ubiquitous within it. Fourth, the normative payoff, where we ask which uses close paths off, which amplify existing pathologies, and what human flourishing with such systems might actually require. Teaching is seminar-based and reading-led, drawing on current research papers alongside draft chapters from the lecturer's forthcoming book Brave New Minds: GenAI as Cognitive Ecology. The course also includes hands-on sessions with the Digital Andy System (DAS), a large language model derived from the work of the philosopher Andy Clark and partly designed by the lecturer, used here as an object of interpretative rather than merely technical study. No technical background is assumed. Students who wish to take both of this term's seminars will find them designed as a contrasting pair, this one looking outward to practices of making and reading, the other looking inward to agency, memory and the self. Introductory Reading Shanahan, M. (2024). Talking About Large Language Models. Communications of the ACM, 67(2), 68-79. Plato. Phaedrus, 274b-278b (the myth of Theuth and Thamus, on the invention of writing). Any edition; freely available online. Hutchins, E. (2010). Cognitive Ecology. Topics in Cognitive Science, 2(4), 705-715. Optional background: Frankish, K. (2024). What are Large Language Models Doing? In A. Strasser (Ed.), Anna's AI Anthology: How to Live with Smart Machines (pp. 55-78). Berlin: Xenomoi. Freely available at keithfrankish.github.io.