AI systems such as automated vehicles and cleaning robots increasingly perform tasks on people’s behalf, often already outperforming them. Yet human-centered interaction design remains essential because these systems act for and around people. For instance, too much assistance or information can overwhelm users and foster over-reliance; too little can leave them unable to understand what a system perceives, intends, or can handle, reducing trust and willingness to use. Designing appropriate AI behavior for diverse users and contexts is a calibration problem.
In calibrated human-AI interaction, adaptive interfaces mediate AI behavior to keep human experience within individually appropriate ranges relative to the AI’s capabilities, reliability, and context. Planned changes in how AI behavior is mediated can challenge people’s current experience, supporting human learning and adaptation as they encounter increasingly capable and unfamiliar AI systems.
To realize this vision, user interfaces should follow two principles:
Homeostatic adaptation
In physiology, homeostasis is the process by which organisms maintain vital conditions within a stable range despite a changing environment.1
In interaction, homeostatic adaptation could maintain effective human-AI interaction as people and contexts change. Feedback such as ratings, physiological signals, and user behavior could indicate when human experience moves beyond its appropriate range, prompting the interface to adapt how AI behavior is mediated.
Hormetic doses
In toxicology, hormesis describes how controlled exposure to a stressor can stimulate adaptation, while excessive exposure can cause harm.2
In interaction, hormetic doses could deliberately expose AI behavior through planned changes in its mediation, supporting learning and adaptation. For example, an automated vehicle could gradually unveil uncertainty. Repeated doses might move human experience toward an appropriate range along dimensions such as trust as people adapt to AI capabilities.
I develop this vision through empirical and technical research across three areas:
Featured Work
Toward homeostatic adaptation Interface designs optimized per person from their own feedback.
