Apple Study Examines How People Define Boundaries in Self-Trained Sensing Systems
Apple researchers have published a paper on what happens when people define the phenomena a personal machine learning system should sense. In a week-long exploratory study, they identified four sites where participants negotiated the boundaries of what the system meant.
- Apple researchers published a paper on how people negotiate boundaries in personal sensing systems they author themselves.
- The team built two open-ended probes using a Wizard of Oz technique to simulate training a personalized machine learning system.
- Participants used one of the probes in daily life during a week-long exploratory study.
- Four negotiation sites emerged: phenomenon boundaries, the subject in relations, signal versus noise, and the objectivity of data.
- The authors offer design starting points and discuss open-ended probes as a method for ontological design.
Why the researchers looked at ontological boundaries
The paper, titled "Negotiating Ontological Boundaries in User-Authored Personal Sensing Systems," comes from Apple's machine learning research pages in the Human-Computer Interaction area. It was published in October 2026. The authors are Nava Haghighi, Danielle Olson, Halden Lin, Erdrin Azemi, Gierad Laput, Kayur Patel and James Landay. Haghighi and Landay are listed with Stanford University, and several authors, including Haghighi, Lin and Patel, did the work while at Apple.
The starting point is the idea that designed artifacts are ontological. According to the paper, they shape, and sometimes limit, what becomes possible or imaginable. One way to reduce such limits, the authors write, is to give people power over how systems are designed and built.
The paper notes that decades of scholarship have looked at systems that enable this kind of authorship. Yet these systems are often evaluated on whether they are usable, useful or technically feasible. That leaves the question of ontological boundary negotiation unexamined, according to the authors.
How the study worked
The team designed two open-ended probes that use a Wizard of Oz technique. This approach lets participants experience training a personalized machine learning system on phenomena they define themselves, without the system needing to be fully built.
In a week-long exploratory study, participants used one of the two probes during their everyday lives. The setup was meant to capture how people shape and reshape the meaning of what they want a system to sense as they live with it.
Four sites of negotiation
The researchers identified four sites where ontological boundaries were negotiated:
- the boundaries of a phenomenon
- the subject as part of relations
- what is signal and what is noise
- the objectivity of data
The paper does not detail in its abstract how participants handled each of these, so the specifics sit in the full publication. The list does show where the authors found that people actively pushed back on or reworked a system's assumptions.
What it means for design
The authors say they offer starting points for supporting boundary negotiation through design. They also discuss open-ended probes as a method for ontological design.
For designers of personal sensing and machine learning tools, the work suggests that evaluating usability or technical feasibility alone may miss how people decide what counts as a phenomenon, what counts as noise, and how far data can be treated as objective. The study is exploratory, and the source does not describe any product plans tied to it.
