Source-linked AI summary
Negotiating Ontological Boundaries in User-Authored Personal Sensing Systems
Nava Haghighi, Danielle Olson, Halden Lin, Erdrin Azemi, Gierad Laput, Kayur Patel, James Landay
TL;DR
The paper examines ontological boundary negotiation in user-authored personal sensing systems, addressing a gap left by evaluations focused on usability, usefulness, and technical feasibility. It uses two open-ended probes in a week-long study and identifies four sites where participants negotiated such boundaries.
Problem
Existing user-authored systems are often evaluated for usability, usefulness, or technical feasibility, leaving ontological boundary negotiation unexamined.
Method
The authors design two open-ended probes using a Wizard of Oz technique to let participants train personalized machine learning systems on self-defined phenomena during everyday life.
Results
Participants negotiated boundaries around phenomena, relational subjectivity, signal versus noise, and data objectivity.
Takeaways & Limitations
The paper offers starting points for designing and supporting ontological boundary negotiation through open-ended probes.
Takeaways & Limitations
More agency does not automatically produce boundary negotiation; expanding beyond existing boundaries requires tools and practices.
Abstract
from arXiv · showhide
Designed artifacts are ontological, shaping, and at times limiting, what becomes possible or imaginable. One path toward mitigating such foreclosures is giving people power over how systems are designed and built. Despite decades of scholarship around systems that enable such authorship, these systems are often evaluated on whether or not they are usable, useful, or technically feasible, leaving questions of ontological boundary negotiation, unexamined. We design two open-ended probes that utilize a Wizard of Oz technique to enable the experience of training a personalized machine learning system on phenomena people define themselves. In a week-long exploratory study, participants use one of two probes in the course of their everyday lives. We identify four sites where ontological 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. We offer starting points for supporting boundary negotiation through design and discuss open-ended probes as a method for ontological design.
1 Introduction
The paper examines how user-authored personal sensing systems can support negotiation of ontological boundaries, rather than evaluating authorship only through usability, usefulness, or feasibility. Two open-ended probes and a week-long study identify four negotiation sites and propose design starting points.
- Designed artifacts shape what people can imagine and make possible, while potentially foreclosing alternatives.
- Personal sensing systems provide a productive site because they define boundaries around phenomena, persons, signal, noise, and data objectivity.
- Prior user-authored systems advance agency, but are often evaluated for usability, learnability, usefulness, or technical feasibility rather than boundary negotiation.
- The study uses two open-ended probes that simulate personalized machine-learning training on phenomena defined by participants, using Wizard of Oz feedback to separate authoring from technical constraints.
- Eight participants used one probe during everyday life for one week, defining categories, identifying phenomena, and imagining uses beyond technical-feasibility constraints.
- The analysis identifies four sites of negotiation: phenomena, relational subjectivity, signal versus noise, and data objectivity, offering design guidance and framing open-ended probes as ontological-design methods.
2 Related Work
The related work frames ontology as concerning what can be thought or imagined, and connects ontological design to critiques of assumptions embedded in AI and interactive systems. It positions personal sensing as a domain where these assumptions remain consequential.
- 2.1 Ontologies and Ontological Design: In philosophy, ontology concerns the nature of being; computing also uses the term for formal systems representing knowledge.
- 2.1 Ontologies and Ontological Design: The paper uses ontologies to mean what people allow themselves to talk or think about, including what they can imagine as possible.
- 2.1 Ontologies and Ontological Design: HCI scholarship argues that rationalistic computer-science traditions can overlook ontologies and that design embeds assumptions about social roles and practices.
- Critiques of AI have concentrated on values and bias while neglecting the ontological commitments underlying AI systems.
- The paper asks whether greater agency through user-authored systems can move beyond existing ontological limitations and what role design retains.
3 Participation in Design and User-authored Systems
Participation and user-authorship let people adapt, extend, interpret, and shape systems over time, but prior work has not examined what scaffolding supports expanded ontological possibilities. Personal sensing makes this gap visible because its systems often treat people as objectively measurable, controllable individuals.
- Participation in Design: Prior work studies appropriation, end-user development, ambiguous artifacts, infrastructuring, and systems that users extend after deployment.
- Participation in Design: Generative AI expands possibilities for participation in design, including visions of malleable software.
- User-authored Personal Sensing: Despite this work, the scaffolding needed to support expansion of ontological possibilities in user-authored systems remains unexamined.
- User-authored Personal Sensing: Personal sensing systems commonly assume humans are bounded individuals whose properties can be objectively measured and controlled through suitable interventions.
- User-authored Personal Sensing: Critical scholarship has found limited support for deeper transformative reflection in commercial apps and continuing reinforcement of these limitations by machine-learning models.
- User-authored Personal Sensing: The paper therefore examines how openness, authorship, and agency may create room for negotiating and expanding boundaries in personal sensing over time.
4 Two Open-ended Probes for Exploring Personalized AI-based Sensing Systems
The paper presents two open-ended probes that let participants engage with personal-sensing data at different abstraction levels while remaining flexible and revisable. A Wizard of Oz approach simulates personalized model training, enabling exploratory study of boundary negotiation without constraining the domain.
- Study Rationale: A week-long use period gives access to how engagement and existing boundaries change over time, without prescribed categories or fixed tasks.
- Probe Design: The probes were designed to resemble dominant personal-sensing approaches rather than serve as critical alternatives, while differing in where authoring begins.
- Probe Design: The probes connect low-level physiological signals, such as heart rate and accelerometer data, with higher-level constructs such as stress, fatigue, or focus.
- EventMarker: EventMarker starts from user-defined experiences and seeks sensor patterns associated with marked states or events.
- PatternFinder: PatternFinder starts from physiological patterns that interest or puzzle participants and translates them into discrete events for detection.
- Wizard of Oz: Wizard of Oz separates the authoring experience from present-day technical constraints by simulating training feedback while models are described as being trained after the study.
- Shared Features: Both probes let participants create, edit, and delete events, patterns, and categories, supporting revision throughout the study.
- Interfaces: EventMarker uses a timeline and optional smartwatch access for marking experiences, whereas PatternFinder presents Heart Rate, Active Energy, and overlaid physiological timelines for selecting patterns.
5 Exploratory Study
The study used two open-ended probes in an eight-day exploratory deployment to examine how participants negotiated ontological boundaries in everyday personal sensing. Analysis combined inductive coding with diffractive reading to identify boundary-negotiation moments rather than generalizable usage patterns.
- Study aim: The qualitative study aimed to explore potential sites of ontological-boundary negotiation and factors that limited or contributed to it, not produce population-generalizable findings.The authors frame situated insight as the study’s goal.
- Study design: Eight smartwatch-wearing adults used one assigned probe for eight days, with onboarding, daily surveys, a day-three check-in, and day-eight offboarding.Participants were recruited across gender, age, and personal-sensing experience.
- Probes: EventMarker asked participants to mark states or actions of interest, while PatternFinder asked them to label interesting patterns in physiological data.The probes differed in their level of abstraction.
- Data collection: Participants’ marks, categories, physiological signals, surveys, interviews, and check-ins were collected for analysis.The physiological data included heart rate, HRV, VO2Max, walking measures, calories, and exercise time.
- Analysis: The analysis used affinity diagramming, inductive open coding, and diffractive reading to foreground moments when participants questioned or redrew boundaries.These moments included deciding category membership and distinguishing signal from noise.
6 Insights
The insights section presents findings from a one-week exploratory study of two open-ended probes. It first uses participant vignettes, then identifies sites where ontological boundaries were negotiated.
- Findings overview: The findings draw on participant experiences with EventMarker and PatternFinder during the one-week exploratory study.Participants P1–P4 used EventMarker, while P5–P8 used PatternFinder.
- Findings overview: The section combines narrative vignettes with an analysis of sites where participants negotiated ontological boundaries.The vignettes provide context before the boundary-negotiation analysis.
6.1 Participant Narrative Vignettes
Participant vignettes show how the probes redirected attention toward personally meaningful experiences, patterns, and relationships in sensor data. Participants also encountered uncertainty, rigid categories, and practical limits during the week.
- Occasional users: Participants reported that marking and categorizing changed their attention, exposed distinctions between mental and physical exhaustion, and revealed contextual patterns such as social influences on binge eating.Some participants also stopped using categories when they could not explain or consistently interpret them.
- Frequent users: EventMarker enabled P3 to mark experiences such as eye strain and pulled muscle despite believing current sensors could not detect them.The probe shifted attention from what seemed technically trackable toward events of personal interest.
- Frequent users: PatternFinder led P7 to investigate mismatches between calorie burn and heart rate and whether repeated dog walks produced similar readings.These uses connected physiological patterns with bodily experience and recurring activities.
- Occasional users: P8 initially found physiological data boring and difficult to interpret, but began treating missing heart-rate measurements as a pattern.P8 also felt the one-week study was too short to become accustomed to finding patterns.
- Occasional users: P1 realized meaningful events could be tracked even if they would not make sense to developers or other people, while expecting automatic detection that did not occur.The experience also made P1 more deliberate about category naming and organization.
- Occasional users: P5 developed relational interpretations of physiological data, connecting signals with a partner, activities, and subjective perceptions, and found the progress wheel motivating.P5 established categories during an initial exploration and made no later changes.
6.2 Negotiating the Boundaries of a Phenomenon
Participants negotiated what phenomena belonged in categories, while the probes’ abstraction levels and progress-oriented features shaped whether boundaries remained open or became fixed. Prior experience, technical friction, and study duration also influenced these negotiations.
- Phenomenon boundaries: The probes invited participants to define what they wanted to track, drawing a boundary between what was and was not being sensed.Participants could revise their categories as their understanding changed.
- Phenomenon boundaries: P1 split “walking” into “chill walk” and “search walk” after recognizing that the activity contained distinct purposes.The change occurred on the third day.
- Phenomenon boundaries: Sustained marking led participants to recognize that “Feel exhausted” could combine distinct mental and physical states.P3 and P2 also encountered the difficulty of defining exhaustion and deciding whether to split the category.
- Probe effects: EventMarker created conditions for less clear boundaries, whereas PatternFinder reinforced clearer, more static boundaries and produced no participant categories for states or unintended events.The authors associate this difference with the probes’ abstraction levels and the greater definitional clarity of activities.
- Probe effects: PatternFinder’s guides encouraged consistency, leading P5 to formulate exact rules such as heart rate between 100 and 150 and medium intensity between 2 and 6.These rules pushed the category boundary toward a fixed form.
- Interface effects: The progress wheel made categories appear completable, and three participants felt negatively about failing to complete any category within one week.The authors suggest this desirable completion state may have foreclosed experimentation.
- Contextual factors: Prior tracking experience shaped negotiation: P8’s limited experience prompted questioning of what “interesting” meant, while technical slowness made category modification harder for P7.The study’s short duration also constrained P8’s opportunity to become accustomed to pattern discovery.
6.3 Negotiating the Subject (as Part of Relations)
Participants negotiated the subject of tracking as relational rather than singular, with phenomena emerging through relationships, co-occurring elements, and temporal sequences.
- Implications: Boundary negotiation exposed relational circumstances, other beings, and temporal relations as constitutive of what gets tracked.The subject boundary therefore remained open to entities and relations beyond the individual.
- Relational subjects: Participants questioned whether tracked phenomena belonged to an individual, instead locating them in relationships and circumstances.Binge eating was marked specifically in relation to being with a friend, while low moods were tied to circumstances that brought them into existence.
- Relational subjects: P7’s walks became a phenomenon shaped by both participant and dog, rather than a property of P7 alone.Route, pace, and energy expenditure varied with the dog’s behavior as well as P7’s own.
- Assemblages: P3’s three tics became an assemblage when their temporal co-occurrence revealed a changing phenomenon beyond isolated symptoms.The participant shifted from tracking individual tics to examining their combinations, correlations, and evolution across time.
- Assemblages: Feelings and actions were treated as sequences that constitute phenomena together, rather than as neatly separable states or events.P3 considered patterns such as sadness followed by running and peace, or feeling good followed by checking in on family.
6.4 Negotiating Signal vs. Noise
Participants negotiated whether patterns counted as signal or noise rather than accepting sensing systems’ classifications, including by treating divergences and absences as meaningful.
- Preclassification: Participants sometimes questioned signal and noise boundaries before seeing data because sensing systems had already decided what to record.The probes made this prior boundary visible as a design decision.
- Absence as signal: P8 treated missing heart-rate measurements as a pattern and thereby audited the sensor’s classification logic.They discovered the device did not monitor heart rate while they were sitting down.
- Reinterpreting patterns: P7 treated divergence between active energy and heart rate as a signal rather than an error or anomaly.The unexpected “high cal no hr” pattern challenged the assumption that both measures should indicate exertion together.
- Reinterpreting patterns: P6 wanted to decide whether a heart-rate spike and rapid decay was meaningful instead of allowing the system to determine its status.The pattern prompted reflection on what they were doing and feeling when it occurred.
- Fixed boundaries: Fixed thresholds could turn atypical bodily states into noise, producing repeated alerts and alert fatigue.P7’s chronically low heart rate triggered regular warnings that they wanted to override or teach the device was normal for them.
- Fixed boundaries: A classification that becomes too accurate or opaque can make its signal-noise boundary harder to question or perceive.P7 eventually stopped questioning predictions that matched their Garmin, while the system retained no trace of its classification history.
6.5 Negotiating the Objectivity of Data
Participants challenged the objectivity of sensing data by interpreting phenomena relationally and valuing classifications meaningful within their own interpretive frames.
- Interpretive frames: Personal sensing classifications need not have a generalizable meaning when people author systems for themselves.P1 described blackbox data that made sense to them even though it did not make sense to the system.
- Relational representation: Participants imagined understanding data relative to people, activities, times, or absences rather than as absolute truths.Relational representations changed how boundaries could be interpreted without necessarily changing the phenomena themselves.
- Relational representation: Comparing activities across routines or contexts allowed participants to interpret variation without treating measures as equivalent.Examples included comparing occasional activities with daily walking and outdoor runs with treadmill runs.
- Relational representation: P8 marked identical yoga sessions to detect anomalies against prior sessions rather than treating each session as an absolute measure.The intended comparison was longitudinal, across the routine’s previous instances.
- Relational representation: Meaning could reside in relationships between people’s patterns, such as P5’s calorie expenditure compared with their partner’s during shared activities.Relational translation can make classification less authoritative by foregrounding interpretive work.
- Political implications: Relational calibration is not neutral because it establishes who or what becomes the point of reference.The paper warns that relational understanding can forget its own politics.
7 Discussion
The study identifies four sites of ontological boundary negotiation and uses them to propose design strategies that keep classifications provisional while acknowledging their political and practical limits.
- Findings: The study found negotiations around phenomena, subjects, signal versus noise, and data objectivity, showing that participants could notice imposed boundaries.The authors use these sites to identify where open-ended sensing design can support further negotiation.
- Designing provisionality: Design can support provisional boundaries through flexibility, customization, ongoing revision, and visual communication of how classifications evolve.The authors emphasize making boundaries negotiable rather than permanently keeping every boundary open.
- Relational subjects: Design can represent the subject as relational, including phenomena co-constituted with other beings such as P7’s walks with their dog.This challenges the assumption of a singular or objective tracking subject.
- Designing provisionality: The representation and abstraction level of a probe shaped what participants could negotiate.EventMarker foregrounded moments and unnamed experience, whereas PatternFinder oriented participants toward certainty and objective truth; no PatternFinder participant created a fuzzy state, feeling, or event category.
- Algorithmic mediation: Algorithms and their visual communication can support negotiation by making category formation and boundary histories more visible.The paper connects algorithmic presentation, such as a progress wheel, with whether boundaries remain fixed or become negotiable.
- Politics and limits: Boundary negotiation can produce risks, including anxiety from constant monitoring, unequal labor for less experienced trackers, and the centering of one story over others.The authors call for scaffolding and support that centers marginalized experiences while preserving users’ ability to let boundaries become fixed when desired.
- Methodological contribution: The paper cautions that open-ended ontological design remains difficult because imagining alternatives to taken-for-granted boundaries is itself challenging.The method offers openings for examination but does not guarantee that another research team will interpret them as negotiations.
- Methodological contribution: Open-ended probes provide a cyclical method for identifying ontological openings, expanding them through design, and revisiting newly exposed boundaries.The authors frame the method as a way to attend to what authoring agency reveals.
8 Conclusion
The study identifies four sites of ontological boundary negotiation in user-authored personal sensing systems, while showing that these negotiations can remain partial and temporary. It proposes open-ended engagement with boundary-fixing technologies as a method for supporting expansion beyond existing boundaries.
- Two open-ended probes and a week-long study with eight participants generated empirical insights into boundary negotiation in user-authored personal sensing systems.
- Participants negotiated boundaries of phenomena, the subject as part of relations, signal versus noise, and data objectivity.
- These negotiations were sometimes partial and temporary, tending toward fixity rather than producing stable boundary expansion.
- More agency does not automatically produce ontological boundary negotiation.
- User-authored sensing systems require tools and practices that support expansion beyond existing ontological boundaries.
- Open-ended engagement with technologies that would otherwise draw and fix boundaries contributes a method for ontological design.