Key Moments
Stanford CS547 HCI Seminar | Spring 2026 | Toward Ontological Multiplicity in AI and Computing
Stanford OnlineWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
AI systems unknowingly encode ontological boundaries, limiting reality to what designers deem possible, but new methods like 'purple zone' and boundary negotiation aim to reveal and expand these limits.
Key Insights
Ontological boundaries determine what we perceive as possible and are often invisibly embedded in AI systems, illustrated by an LLM failing to depict trees with roots until prompted with "everything in the world is connected."
The 'purple zone' phenomenon, identified through EDA sensor data, highlights how AI systems define and exclude certain human experiences by setting fixed boundaries, such as the 'measurable human'.
Design can actively shape ontological boundaries, as seen in the 'event marker' and 'pattern finder' probes, which influenced participants' ability to define categories based on lived experience versus data signals.
Boundary negotiation within AI systems can occur in at least four ways: negotiating the phenomenon itself, negotiating the subject's relationality, distinguishing signal from noise, and questioning the objectivity of data.
The development pipeline for AI, including LLM outputs and agent architectures, often defaults to specific ontological orientations (e.g., human as a biological individual) even when diverse data exists, limiting multiplicity.
The speaker's work offers methods for dissolving and negotiating ontological boundaries to create conditions for ontological multiplicity, moving critique from reaction to prevention and expanding imaginative possibilities in AI and computing.
Imagining trees reveals embedded ontological assumptions
The lecture begins by illustrating how our understanding of even a simple concept like a 'tree' is shaped by our background and disciplinary lens. A botanist sees mineral nutrients, a spiritual healer sees whispering connections, and a computer scientist might envision a binary tree. These descriptions reveal fundamental assumptions about a tree's boundaries and its very nature. Ontology, the study of being, is central to this discussion, with the speaker defining it as the boundaries of what we allow ourselves to think about and how these boundaries shape our perception of possibility. When these ontological assumptions are encoded into AI systems, they risk becoming universally accepted realities. An example with an LLM image generator shows how a prompt for a tree without explicit mention of roots failed to produce them, even when specifying origin ('from Iran'), suggesting the AI imposes its own default boundary of what a tree is and how it should be represented. Only by prompting with a concept like "everything in the world is connected" did the AI produce a tree with roots, demonstrating that the AI prioritized a particular set of relations over others, highlighting the significant role of invisible ontological assumptions.
The 'purple zone' dissolves the boundary of the measurable human
A key example of dissolving ontological boundaries is the speaker's personal experience with Electrodermal Activity (EDA) sensors. These sensors measure skin conductance as a proxy for arousal, but EDA baselines vary significantly between individuals, leading to a common distinction between 'responders' and 'non-responders.' Non-responders, like the speaker, often have baselines too low to show meaningful activity and are typically excluded from studies, effectively rendering them invisible to the system. However, the speaker discovered that during a particularly emotional meeting, their EDA signal became measurable, crossing the previously fixed boundary. This revealed an 'ontological glitch' and led to the concept of the 'purple zone'—an ambiguously measurable state where a dissolved boundary creates a space of possibility. This experience challenged the assumed boundary of the 'measurable human' as a discrete, objectively quantifiable individual, aligning with critiques from decolonial scholarship and feminist theorists like Karen Barad, who argue that research apparatuses enact, rather than merely measure, reality.
Developing an embodied sensitivity to glitches and boundary crossings
Drawing inspiration from feminist and queer theories of the 'glitch,' the speaker developed an embodied sensitivity to these boundary-crossing moments through an autoethnographic study. For nine months, they wore an EDA sensor daily, meticulously documenting their experiences and correlating them with data. This process revealed that 'purple zone' events looked different for various individuals, necessitating a move from standard statistical analysis to a 'situated understanding of data.' This involved intimate engagement, visualization, and manual annotation, leading to the co-evolution of an algorithm that approximated 'purple zone.' A subsequent study with a real-time biofeedback system on Apple Watches showed that participants, like 'Nigel,' began to see purple zone as an 'anticipated guest' that reconstituted relational units, or like 'Elena,' experienced it as a puzzling bodily signal beyond conscious control. Another participant deliberately sought to create conditions for purple zone before brainstorming sessions, shifting from control to cultivating possibilities. These examples demonstrate how attuning to glitches can dissolve fixed boundaries and create openings for ontological multiplicity.
Shifting paradigms in technical and design practice
The exploration of EDA purple zones led to four significant shifts for technical and design practice: 1) A move from classification (me vs. other) to noticing relationality, where the boundaries of the self change in relation to others. 2) A shift from control to cultivation, by enacting change in one part of an assemblage and waiting for reciprocal change elsewhere. 3) A transition from accuracy to algorithmic precision through care, where the goal becomes creating ontological openings rather than representing an objective property. 4) A shift from invisibility to the 'algorithmic in-between,' recognizing that the political implications of boundary drawing are most experiencable not in visibility or invisibility, but in this interstitial space. While boundary crossing holds power, it also involves ongoing labor to remain legible across different sides.
Negotiating boundaries through open-ended probes
The next approach focuses on negotiating existing boundaries rather than dissolving them, particularly concerning the 'universal self' often assumed in AI systems. The speaker developed two open-ended probes, 'event marker' and 'pattern finder,' to simulate training personalized machine learning systems, using a Wizard of Oz approach to bypass current technical constraints. 'Event marker' starts from lived experience (e.g., a feeling), moving towards quantifiable data, while 'pattern finder' starts with data (e.g., heart rate) and moves towards meaning. During a one-week study with eight participants, these probes allowed users to create, edit, and delete their own categories, revealing how design shapes what boundaries become visible. For instance, participants using 'pattern finder,' which focused on data signals, were less comfortable defining categories with fuzzy edges compared to those using 'event marker.'
Instances of boundary negotiation offer new possibilities
The study on boundary negotiation revealed several key types of negotiation. First, participants negotiated the boundaries of phenomena themselves, splitting a category like 'walk' into 'search walk' and 'chill walk' based on embodied experience, or questioning whether exhaustion was emotional, mental, or physical. Second, they negotiated the subject itself as part of relations, realizing that tracking a run with a dog meant measuring the run for both, blurring the boundary of who was being measured. Third, the distinction between signal and noise became negotiable; marking 'absence' in heart rate data, perceived as noise by an affective computing researcher, was actually an audit of the algorithm's functioning. Finally, participants negotiated the objectivity of data, preferring to understand their activity in relation to others rather than absolute values, or even marking feelings they lacked words for, labeling it as 'blackbox data' that makes sense to them. This highlights a paradigm shift where self-authored systems don't require universally understandable categories.
A framework for surfacing and questioning taken-for-granted boundaries
The final part of the talk introduces an analytical framework for systematically surfacing taken-for-granted ontological boundaries in existing or new systems. This framework is developed from two decades of socio-technical system design research and draws on theories of ontology. It comprises four provisional analytical orientations: multiplicity, groundedness, liveliness, and enactments. Each orientation serves as a lens to ask where boundaries are being drawn and what is being excluded. The speaker illustrates 'multiplicity' with analyses of commercial LLM chatbots and an LLM-based agent system. For instance, while Bard and GPT-4 acknowledged diverse human definitions, they defaulted to an underlying assumption of humans as biological individuals, only offering more relational views when explicitly prompted. Similarly, an LLM agent architecture modeling humans as isolated individuals within their own minds overlooks symbiotic relationships, such as the human microbiome's influence on cognition. This framework reveals how even diverse LLMs can present a single dominant reality and offers a way to critique and expand these boundaries, opening up new potentialities for AI design based on assemblages and collective memory systems.
Ongoing and future work in ontological multiplicity
The speaker outlines ongoing and future research directions. This includes exploring ontological multiplicity within the AI development pipeline, particularly concerning data, with an interest in small models to understand how ontological workings can be revealed through care and glitch attunement. Future work also involves probing LLMs for latent relational ontological information and designing AI architectures that de-center the individual and center relations. A new category of work focuses on AI systems that generate other systems (code, architectures), examining how these downstream systems embed assumptions and how to de-center them. The speaker proposes studying how design shifts when designing systems that design systems, and investigating multiplicity at the level of systems mediating millions of interactions. The contributions are theoretical (articulating ontological boundaries as openings), methodological (developing practical approaches), and applied (evolving theory into grounded insights) to expand imaginative possibilities in AI and computing.
Mentioned in This Episode
●Products
●Software & Apps
●Organizations
●People Referenced
Navigating Ontological Boundaries and Multiplicity
Practical takeaways from this episode
Do This
Avoid This
Common Questions
Ontological multiplicity refers to the idea that reality is not singular but composed of multiple, equally valid ways of understanding and defining existence. In AI and computing, it means recognizing and designing systems that acknowledge and accommodate diverse ontological boundaries, rather than enforcing a single, dominant perspective.
Topics
Mentioned in this video
A language model used by the speaker to generate images based on prompts. The speaker illustrates how altering prompts can lead to different, sometimes unexpected, visual outputs.
A commercial LLM chatbot examined for its ontological assumptions regarding the definition of 'human'. Bard acknowledged diverse perspectives but defaulted to a biological individual assumption until prompted with non-Western philosophies.
A commercial LLM chatbot that, along with Bard, offered multiple perspectives on the definition of 'human' but shared a common underlying assumption of humans as biological individuals.
A feminist theorist and physicist who coined the term 'agential realism' and the concept of an 'essential cut' in relation to drawing boundaries around phenomena.
An artist whose song lyric, 'There is a crack in everything. That's how the light gets in,' is invoked by the speaker as a concluding metaphor for multiplicity and finding openings.
More from Stanford Online
View all 86 summaries
40 minStanford MS&E435 Economics of the AI Supercycle | Spring 2026 | Infrasctructure, Enterprise AI, SaaS
48 minStanford CS547 HCI Seminar | Spring 2026 | Just-in-Time Objectives for Specialized AI Interactions
97 minLive from Stanford Health AI Week
35 minStanford Robotics Seminar ENGR319 | Spring 2026 | Towards Trustworthy Autonomy
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free