The following conversation between K Allado-McDowell, Callum Dean, and myself occurred via Zoom on Wednesday, 14 July, 2021, at 9:00 New York/14:00 London. GPT-3 was subsequently invited, with Callum’s help, to respond at the end of the editing process to a series of prompts I selected from our conversation. Callum solicited multiple responses to each prompt, and I chose from among these the responses that I found most relevant and/or most surprising. In several instances, GPT-3’s entire, unedited response to a prompt has been included. In other cases, I have selected only a portion of an entire response, just as I would when editing a human-authored passage. However, I have not altered GPT-3’s vocabulary or grammar, nor suggested any other rewrites, as I often do when working with human-authored texts.

This is the fourth publication of Version Space, a series of pamphlets transcribing conversations among artists and graduate students in visual art regarding Artificial Intelligence and related topics, produced in collaboration with Library Stack and funded by the Artistic Research program of the Sandberg Instituut, where I was a research fellow in the Department of Fine Arts for the academic years 2018–19 and 2019–20. The conversations take place via videoconference, are recorded, transcribed by an AI, then edited for clarity by humans, including myself, the participants, and Library Stack. The series title is borrowed from a machine learning framework conceived in the 1980s as a technique for identifying a range of possibilities in the overlap between maximally general and maximally specific hypotheses.