Our paper, open access via the ACM Library (click to visit)
Co-authored with Søren Pold and presented in August at Computing [X] Crisis in Aarhus, our new paper “Reading the Praise/Prompt Machine: An Interface Criticism Approach to ChatGPT” introduces novel methods for critiquing the linguistic interfaces of AI chatbots. We theorize one of ChatGPT’s central engagement mechanics as the “praise/prompt envelope”—a pattern of affirmation and nudging designed to keep users talking—and preview my forthcoming artworks that aim to expose and attune users to the engagement aesthetics embedded in ChatGPT’s interface.
The full abstract:
This paper critically examines ChatGPT through the lens of interface criticism. Our work develops new methodological approaches to AI critique and reveals how the platform’s core engagement mechanics operate via language rather than traditional interface elements. Through systematic three-way conversation experiments and critical prompting, we demonstrate how ChatGPT accommodates rather than challenges user perspectives, struggles to sustain disagreement or deliberation, and reinforces engagement through validation. We show that ChatGPT’s answers arrive wrapped in what we term the “praise/prompt envelope”—a carefully crafted package of validation and query designed to sustain user interaction. Two key artworks—Lux Affirma, a custom GPT that amplifies ChatGPT’s praise and affirmation to the point of absurdity, and the ChatGPT Demotivator, a browser extension that exposes the platform’s linguistic manipulation in real-time—make visible to users how ChatGPT shapes behavior through conversation itself. Our findings reveal that ChatGPT’s seemingly natural dialogic flow masks a carefully engineered system of linguistic engagement, designed not for deliberation but for continuation. This insight highlights the need for new critical approaches attuned to how the language-based interfaces of generative AI manipulate users.
In the paper, I outline three artworks currently in progress: Lux Affirma (try a prototype here), the ChatGPT Demotivator (see GIF below), and 100 Affirmations (a book series). The Demotivator is progressing well and I expect to release it soon.
ChatGPT Demotivator prototype running on chatgpt.com, 2x speed (watch full 3m video)
As of this writing, our paper is currently the top download from the proceedings. My sincere thanks to Søren Pold who is such a pleasure to think and write with, and to Lone Koefoed Hansen for her able co-stewardship of the conference. Further thanks to the Aarhus University Research Foundation, the Independent Research Foundation Denmark, and the Center for Digital Narrative at the University of Bergen for their support of this work.