Designing Infrastructural–Embodied Coupling for Sustainable Human-AI Interaction

6000

AI feels effortless. A prompt goes in, an answer comes out. However, behind this seemingly weightless interaction lies a vast physical infrastructure and significant planetary resource consumption. Recent benchmarking shows that even a single short GPT-4o query, such as a casual extra prompt like “thank you,” consumes about 0.42 Wh of electricity; when scaled across hundreds of millions of daily requests, such ordinary user habits aggregate into annual electricity use comparable to 35,000 homes and water consumption equivalent to the annual drinking needs of 1.2 million people.

This SURF project proposes to design, prototype, and evaluate a new series of personal AI devices, including wearable, portable, and desktop AI interfaces, that experientially present infrastructural costs through embodied feedback. Students will first use system mapping to reveal how everyday AI behaviors trigger hidden infrastructural and planetary consequences, then translate those dynamics into embodied interaction concepts, prototype coupling mechanisms, and integrate them into coherent physical AI devices for more sustainable human-AI interaction.

This project investigates the following main question:
How might AI infrastructure dynamics and planetary consequences be translated into embodied interaction design that meaningfully influences user behavior across different scales of everyday AI use?
Which can be divided into three sub-questions through the development of the work:
1. How can planetary-scale AI infrastructure be modeled and operationalized as dynamic interaction variables?
2. What forms of embodied interaction can effectively translate infrastructural dynamics into meaningful behavioral modulation?
3. When embodied coupling mechanisms are materialized, how do they enact, modify, or complicate the intended bodily experience and behavioral response?

The project is expected to generate outcomes at three levels.
First, it will produce a set of personal AI devices/interface designs in wearable, portable, and desktop formats that embody hidden infrastructural costs through interaction design.
Second, it will produce companion design outputs showing how planetary-scale AI infrastructure can be translated into dynamic interaction variables. These include system maps, behavior-infrastructure-planet diagrams, and typologies of the embodied coupling concept.
Third, it will produce evaluative knowledge about embodied interaction. By comparing intended and observed bodily experience and behavioral response, the project will show which coupling mechanisms are effective, where they become ambiguous or complicated, and how device scale and context shape their impact. In other words, the design and making of the artifacts investigates how embodied coupling actually works in practice.
The work should result in portfolio-quality design outputs for exhibition as well as a research paper within the field of sustainable HCI.