What happened
Researchers at Apple and Stanford ran a week-long exploratory study with two open-ended probes that use a Wizard of Oz technique, letting participants train a personalized machine learning system on phenomena they defined themselves, and identified four sites where ontological boundaries were negotiated.
Why it matters
Participants could shape how a personal sensing system is built, not just use it, so the study points to design approaches that give people more say over what such systems treat as meaningful, a question prior work left unexamined.
What to watch
The findings come from a small exploratory study of two probes, so whether they hold for real products hinges on further testing. The paper was accepted at the AI4TCI Workshop at ARES 2026.
WHO IT HITSProduct designers and researchers building personal sensing and personalized machine learning features, who may need to account for how users define their own data categories rather than only whether a system is usable or technically feasible.
Summaries like this, in your inbox every morning.
The paper starts from a familiar critique: the artifacts people use shape what they can imagine and do, and one way to loosen that grip is to let people author the systems themselves. Prior scholarship on user-authored systems, the authors write, has mostly asked whether those systems are usable, useful, or technically feasible, leaving the negotiation of ontological boundaries unexamined. To look at that gap directly, the team built two open-ended probes around a Wizard of Oz technique, in which participants could go through the experience of training a personalized machine learning system on phenomena they defined themselves, and then ran them in a week-long exploratory study in which people used one of the two probes in the course of everyday life. From that, they name four sites where boundaries were negotiated: the boundaries of a phenomena, the subject as part of relations, what is signal and what is noise, and the objectivity of data. They also propose open-ended probes as a method for ontological design, and the work was accepted at the AI4TCI Workshop at ARES 2026. Because the study is exploratory and built on two probes, how far these four sites generalize to shipping products is likely to depend on whether other teams test the same method at larger scale; for designers of personal sensing features, the useful question may be less whether a system is usable than who gets to decide what counts as meaningful input.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
