Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
Apple Research on Personal Sensing Systems and Knowledge Extraction - OpenSmartRoute
Apple Machine Learning Research. Image: Apple machine learning research (original)
Apple researchers published a paper on October 5, 2026. They focused on personal sensing systems and user control. The team includes Nava Haghighi, Danielle Olson, and others. Their work appears in the Apple machine learning research collection. This paper explores how people define what machines should sense. Designed artifacts shape what is possible for users. These tools can limit imagination if not careful. Giving people power over design helps fix this.
Scholars have studied systems that allow authorship for years. Most evaluations check usability, usefulness, or technical feasibility. Fewer studies look at ontological boundary negotiation. This term means negotiating the limits of what exists in a system. The researchers designed two open-ended probes for their study. They used a Wizard of Oz technique to test these tools. Participants trained personalized machine learning systems on self-defined phenomena.
The study lasted one week in everyday life settings. Researchers observed how users interacted with the probes daily. They found four specific sites where boundaries were negotiated. These sites involve phenomena, relations, signal, and noise. The team also examined the objectivity of data collection. Their findings offer starting points for better design support. Open-ended probes serve as methods for ontological design work.
The Concept of Ontological Boundaries in Design
Designed artifacts are not neutral objects sitting on a table. They actively shape what becomes possible or imaginable for humans. When a tool limits certain actions, it creates an ontological boundary. This concept comes from philosophy and design theory fields. It asks how our creations define reality itself. Without this awareness, systems might exclude important human experiences.
For example, a voice assistant might ignore a specific type of command. The user feels the system does not understand them. This is a failure in negotiating ontological boundaries properly. The system assumes it knows what counts as valid input. But the user defines their own reality and needs. Giving people power over design mitigates these foreclosures. It allows users to define the limits of their tools.
Jennifer Neville discusses how standard benchmarks miss real-world AI failures in a new Microsoft Research Podcast episode.
The researchers used a specific method called the Wizard of Oz technique. This is not a standard machine learning training process. Instead, human operators simulate system responses during the study. Participants believed they were training a real personalized machine learning system. In reality, the "system" was guided by researchers behind the scenes. This approach enabled the experience of training on user-defined phenomena.
Participants used one of two probes throughout their everyday lives. The probes acted as interfaces for defining what matters to them. They could identify specific events or patterns they wanted tracked. This method allowed researchers to observe how people conceptualize data. It moved beyond standard datasets that lack personal context. The study focused on user-authored definitions rather than pre-existing categories.
Four Sites of Boundary Negotiation Identified
The team identified four distinct sites where boundaries were negotiated during the study. First, participants defined the boundaries of a phenomena they wanted to track. They decided what counts as a specific event in their lives. Second, they determined how subjects exist within various relations to others. A person's identity is often shaped by these social connections.
Third, users negotiated what constitutes signal versus noise in their data streams. Not every background sound or movement is worth recording. Participants chose which inputs mattered for their personal system. Fourth, they questioned the objectivity of the data being collected. Who decides what facts are true enough to store? The researchers found these four areas critical for design.
ODKE+ System for Open-Domain Knowledge Extraction
Apple also released information about a new tool called ODKE+. This stands for Ontology-Guided Open-Domain Knowledge Extraction with LLMs. It was published on October 27, 2025. ODKE+ is designed to extract and ingest millions of open-domain facts. These facts come from various web sources automatically. The system aims for high precision in its extraction process.
Knowledge graphs (KGs) are foundational to many AI applications today. However, maintaining their freshness and completeness remains costly for companies. Traditional methods struggle to keep up with the volume of new data. ODKE+ offers a production-grade solution to this specific problem. It automates the ingestion of facts from the open web.
How ODKE+ Combines Modular Components
The ODKE+ system combines several modular components into a scalable pipeline. The first component is the Extraction Initiator. This part detects missing or stale facts within the knowledge graph. It knows when information needs updating based on its ontology. The second component is the Evidence Retriever. It collects supporting evidence for each extracted fact from web sources.
The system uses LLMs to guide this entire process effectively. Large language models (LLMs) provide the reasoning needed for extraction. They interpret user-defined ontologies to find relevant facts. This combination allows the system to handle millions of facts efficiently. The modular design makes it easier to scale and maintain. Engineers can update components without breaking the whole pipeline.
Why These Tools Matter for Safety and Control
These tools matter because they address safety and control in AI systems. Personal sensing systems allow users to define their own data boundaries. This prevents unwanted surveillance or data collection on sensitive topics. ODKE+ ensures that knowledge extraction follows specific ontological guidelines. It reduces the risk of hallucinations or incorrect fact ingestion.
Both approaches give humans more agency over their digital environments. Safety comes from knowing exactly what data is collected and why. Control means users can stop or modify the system easily. These capabilities are crucial for high-stakes settings like insurance or procurement. Autonomous negotiation agents operate in areas where mistakes have real consequences. Formalizing inference attacks helps mitigate these risks effectively.
What Engineers Can Do With This Information
Engineers can use these findings to build better user-controlled systems. Start by defining clear ontological boundaries with your users before coding. Ask them what phenomena they want the system to track. Use open-ended probes to test how users define their data needs. For ODKE+, consider integrating it into your knowledge base pipelines. Check if your current KGs need automated fact ingestion updates.
Review your privacy policies against the four sites of boundary negotiation. Ensure you are not assuming too much about user intent or data nature. Test your systems with Wizard of Oz techniques to simulate user definitions. Look for opportunities where LLMs can guide open-domain knowledge extraction. Always prioritize user-defined limits over system convenience.
Apple Research Paper on Personal Sensing Systems
Apple researchers published a paper about personal sensing systems in October 2026. The team led by Nava Haghighi and James Landay focused on human-computer interaction. They wrote the article titled "Negotiating Ontological Boundaries in User-Authored Personal Sensing Systems". This research area explores how people define their own data within AI systems. Designed artifacts shape what becomes possible or imaginable for users. These designs can limit what humans can imagine or achieve. The authors argue that such limitations are often called foreclosures. Foreclosures mean the system stops allowing new possibilities. One path to fix this is giving people power over design. Users need control over how systems are built and trained.
Despite decades of scholarship, existing systems lack ontological boundary negotiation. Scholars have studied systems that enable user authorship for years. Yet, these systems are often evaluated only on usability or feasibility. Usability means the system is easy to use. Feasibility means the technology works technically. Questions about ontological boundaries remain unexamined in most studies. Ontological boundaries refer to the limits of what data counts as real. The paper identifies four specific sites where these boundaries get negotiated. These sites include the phenomena being tracked by the system. It also includes the subject as part of their relations to others. Another site is defining what is signal and what is noise. Finally, the objectivity of the data itself is a key negotiation point.
The authors designed two open-ended probes for this study. An open-ended probe allows users to define phenomena themselves. These probes utilize a Wizard of Oz technique. In this method, researchers simulate user input without real human interaction. This enables the experience of training personalized machine learning systems. Participants used one of the two probes during their everyday lives. The study was a week-long exploratory experiment. Researchers observed how participants interacted with the sensing systems. They tracked where users negotiated the boundaries of their data.
The study identified four specific sites for boundary negotiation. First, users define the phenomena they want to track. Second, they define themselves as part of relational contexts. Third, they decide what counts as signal versus noise. Fourth, they question the objectivity of the collected data. These findings show that users actively shape their digital realities. They do not just passively accept system definitions. The paper offers starting points for supporting this negotiation through design. Designers can build tools that facilitate these discussions. Open-ended probes serve as a method for ontological design. This approach shifts power from engineers to users.
The research area is Human-Computer Interaction. This field studies how people use technology together. Content type of the paper is academic research. It was published in October 2026 by Apple Research. The authors include Nava Haghighi, Danielle Olson, Halden Lin, Erdrin Azemi, Gierad Laput, Kayur Patel, and James Landay. Some names have asterisks indicating equal contribution or specific roles. The paper challenges the traditional view of AI data collection. Traditionally, systems collect data based on predefined categories. This research suggests users should define those categories themselves. It highlights the importance of user-authored personal sensing systems. These systems allow individuals to own their data definitions.
The concept of ontological boundaries is central to this work. Ontology refers to the study of being or existence. In AI, it means defining what exists in the data stream. Signal and noise are classic concepts in data science. Signal represents useful information for the model. Noise represents irrelevant or distracting information. Users must decide where to draw these lines. Objectivity is another contested concept in modern data science. Engineers often assume data reflects objective reality. Users may see their data as subjective experiences. Negotiating these boundaries ensures systems align with human intent.
The paper emphasizes the need for user power in system design. Giving people control mitigates foreclosures effectively. It prevents systems from limiting imagination unnecessarily. This approach respects the diversity of human experience. Different users have different definitions of truth and reality. A one-size-fits-all data model fails to capture this nuance. The study used a week-long duration for its exploratory phase. This length allowed participants to engage deeply with the probes. Short studies might miss long-term boundary negotiations. The Wizard of Oz technique provided a safe testing environment. It protected user privacy while gathering valuable insights.
The four sites of negotiation are critical for future system design. Phenomena boundaries define what the system tracks. Subject relations define how people connect in data. Signal and noise definitions determine data quality. Objectivity questions challenge the nature of the data itself. Designers must address all four areas to be effective. Ignoring any site leads to incomplete or biased systems. The paper suggests open-ended probes as a practical solution. These tools help users articulate their data needs clearly. They make the invisible process of data definition visible. This visibility is key to building trust between users and AI.
The authors conclude that ontological design requires user involvement. Traditional engineering approaches often exclude users from this phase. Engineers might prioritize technical constraints over human definitions. This creates friction between system capabilities and user intent. The paper advocates for a shift in this dynamic. Designers should support boundary negotiation actively. They can build interfaces that prompt these discussions. They can provide tools for users to refine their data definitions. Such tools empower users to shape their digital environments better.
The implications of this research extend beyond Apple products. Any system collecting personal data faces similar challenges. Social media platforms, health apps, and smart home devices all collect data. Users in these contexts need more control over their data definitions. The paper provides a framework for addressing these issues. It moves the conversation from usability to ontology. Usability focuses on ease of use. Ontology focuses on the meaning of the data itself. This shift is necessary for truly user-centric AI systems.
The study highlights the complexity of human-AI interaction. Humans do not provide data in simple, static categories. Their experiences are fluid and context-dependent. Systems that force rigid definitions fail to capture this reality. The paper argues for flexible, user-defined boundaries instead. This flexibility allows systems to adapt to changing contexts. It supports a more natural flow of human-computer interaction. The week-long study period was crucial for observing these dynamics. Short-term interactions might not reveal deep negotiation patterns. Long-term engagement shows how boundaries evolve over time.
The authors identify specific sites where negotiation occurs. These sites are not just technical features but social processes. They involve users interpreting their own lives through data lenses. Designers must recognize these sites as opportunities for engagement. They should create spaces for users to define their reality. This approach transforms the user from a passive data source to an active co-creator. Co-creation implies shared responsibility and ownership of the system's output. It aligns with ethical principles of autonomy and consent.
The paper also touches on the role of designers in this process. Designers are not just builders but facilitators of meaning. They help users articulate their ontological boundaries clearly. This requires empathy and deep understanding of user needs. Engineers need to work closely with researchers and designers. Technical teams must support these human-centric goals. The integration of technical constraints with human definitions is key. Balancing these two forces creates robust and ethical systems.
The research area of Human-Computer Interaction benefits greatly from this perspective. It moves beyond simple interaction design to ontological design. This broader scope includes the meaning and purpose of data. It considers how data shapes human understanding and behavior. The paper contributes significantly to this evolving field. It offers concrete methods for implementing user control. Open-ended probes serve as a tangible example of these methods.
The findings suggest that current AI systems are often misaligned with user intent. Misalignment occurs when system logic contradicts user definitions. This leads to unexpected behaviors and potential harms. Users may feel their data is being misinterpreted or misused. Addressing ontological boundaries reduces this risk significantly. It ensures systems work as users expect them to. Trust in AI depends on this alignment between human intent and machine action.
The paper concludes with a call for new design paradigms. Traditional paradigms assume fixed categories of data. New paradigms must accommodate fluid, user-defined categories. This requires significant changes in how we build and evaluate systems. Evaluation metrics must include measures of boundary negotiation success. Usability alone is no longer sufficient for judging these systems. Quality now includes how well the system respects user definitions.
The research by Apple's team sets a new standard for AI development. It prioritizes human agency over technical convenience. This priority should guide future innovations in machine learning. Engineers and managers alike need to adopt this mindset. Managers can decide on projects that support user control. Engineers can build features that facilitate boundary negotiation. Both roles are essential for creating responsible AI systems.
The paper's focus on personal sensing systems is particularly relevant today. Personal sensors collect vast amounts of intimate data. Users have little control over how this data is defined and used. This research offers a path toward reclaiming that control. It provides tools and methods for achieving this goal. The week-long study demonstrated the feasibility of these approaches. Real-world testing confirmed the value of user-defined boundaries.
The four sites of negotiation remain key takeaways from the paper. They provide a checklist for designers to consider. Phenomena, relations, signal/noise, and objectivity are all critical areas. Designers should audit their systems against these four criteria. They should ask how users define each of these elements. This audit process helps identify gaps in user control. It reveals opportunities for improving the system's alignment with human intent.
The paper also emphasizes the importance of open-ended probes. These probes allow users to explore their data definitions freely. They do not constrain users to pre-set categories. This freedom is essential for authentic boundary negotiation. Probes serve as a bridge between user intent and system implementation. They make the negotiation process visible and manageable.
In summary, the Apple research paper on personal sensing systems offers a fresh perspective on AI design. It challenges engineers to rethink how they define data boundaries. It advocates for user power in shaping these definitions. The methods proposed are practical and grounded in real-world testing. The findings have broad implications for the future of human-AI interaction. They call for a shift from technical optimization to human-centric design. This shift is necessary for building systems that truly serve people.
Google released a new model that turns search trends and weather data into location fingerprints. These inputs improve disease forecasting without needing custom training.
