Generative AI & Embodiment

Role

Undergraduate Research

Made with

Prof. Jennifer Turns Ph.D.

Timeline

8 weeks

Project

Generative AI & Embodiment

Project Overview

Over six weeks I ran five self-observational experiments on how AI shapes reflection and what happens to a person’s sense of autonomy when AI leads. Each experiment gave AI a different kind of control. In A, Claude led my body and my thinking through a 25-minute yoga session. In B, ChatGPT led a spoken reflection. In C, I changed my own behavior (agreeing, resisting, or answering at random) with the same neutral AI. In D, I compared reflecting with AI and with a friend. In E, I stepped out of the conversation and let AI and a friend talk without me. As a group we also compared our experiments through affinity mapping.

Challenge

How much control do we truly have in AI-assisted reflection? The question started in Experiment A, where I noticed how easily I surrendered to the AI’s guidance during yoga. I used research through design: each week I designed a procedure, tested it, and documented it with notes, photos, and transcripts, then used what I learned to design the next one. The work is exploratory and has a single participant, so it generates questions and does not test hypotheses.

Generative AI & Embodiment
Generative AI & Embodiment
Generative AI & Embodiment

Results


  • Guidance can steer without force (A). The AI freed me from planning but subtly set my reflection focus, and I adopted its framing even when it didn’t fully fit. Pausing to read its prompts also broke the flow.

  • Compliance is effortless, and resistance stays inside the AI’s structure (C). Agreeing took no effort. Defiance still kept me within the AI’s frame, only in opposition, and even random answers were absorbed as the AI searched for meaning.

  • Removing myself removed control (E). The AI’s framing shaped the discussion, and the reflection became static and impersonal.

  • Emotional reflection looked different with a person than with AI (D). With AI I could steer the conversation back to critical reflection. With a friend, validation took over and made steering harder. This is the finding I would present most carefully, because it has one participant and private content.

5

iterative experiments

6

weeks