01Ubicomp, twenty years on · Part one09 Jun 2025 · 6 min read
What wearables proved, and what they missed
Ray-Ban Meta, the Humane AI Pin and Vision Pro are the first honest test bench for a framework written in 2006. Two of its three premises hold. The third one — that invisibility is a design outcome — does not.
My thesis argued that computing would become environmental: sensed, contextual, and answerable to the person standing in front of it. Twenty years later the hardware exists. It arrived not because anyone followed the argument, but because transistor density, cheap sensor fusion and on-device inference converged at the same time.
The engine, and the point where density stopped being the metric
Moore's 1965 prediction bought the miniaturisation these devices run on. What matters now is what replaced it. Density gains slowed; packaging and architecture took over — 3D stacking, gate-all-around transistors, heterogeneous integration. The industry stopped optimising transistors per die and started optimising computation per joule.
That shift is the whole reason a wearable is possible. Desktop-class inference inside a sub-10-watt thermal envelope is a power-budget achievement, not a lithography one. Vision Pro runs vision models locally in a few hundred milliwatts; Ray-Ban Meta fuses a camera and spatial audio in real time; each mobile silicon generation has cut power per operation again. None of that was in reach in 2006, and none of it came from the axis everyone was watching.
Ray-Ban Meta: the contextual promise, and the bystander bill
Ray-Ban Meta — a camera, microphones and spatial audio in a frame nobody reads as a computer.
The glasses are the closest thing yet to the environmental memory I described. Ask about what is in front of you and the answer arrives with the surroundings already in the prompt. That is anticipation, and it works.
The cost lands on someone who never agreed to it. The recording indicator is a small LED that reviewers routinely describe as easy to miss, and voice-data retention policy stays with the manufacturer. My own 2005 Wi-Fi fieldwork found the same asymmetry in a smaller form: users consistently underestimated how uncomfortable their instrumentation made other people. Short-range channels that only work on skin contact — body-coupled transmission at a fraction of Wi-Fi's power — are one of the few technical answers that reduce the blast radius instead of asking for more consent screens.
Humane AI Pin: privacy by design, inequity by accident
Humane AI Pin — no screen; a laser projects onto the palm, and a hardware light says when sensors are live.
The Pin got the privacy architecture right. A laser projects a display onto the palm rather than lighting a screen for the room; a hardware light indicates when sensors are live. That is a dynamic, legible permission model built into the object.
It also priced itself at $699 plus a subscription, which is the part my framework never modelled. A technology described as ubiquitous that costs a month of median rent is not ubiquitous; it is a segment. Cheaper analog front-ends would cut the bill of materials substantially, and the research exists. Adoption does not follow, because nothing in the market rewards it.
Vision Pro: contextual permission, then physics
Apple Vision Pro — the permission model is exemplary; the 650 grams are the argument against it.
Vision Pro implements contextual permission almost to the letter: biometric templates stay in a secure enclave, and gaze data is used for interaction without being handed to the application. Twenty-three million pixels are driven in real time at an efficiency that would have read as fiction in 2006.
Then it weighs 650 grams and sits on your face. Weiser's premise was that the most profound technologies disappear. A device you are continuously aware of wearing has not disappeared, and no amount of process node fixes that. Ergonomics is a separate discipline with its own limits, and my framework treated it as a footnote.
What AI actually changed
Three software shifts did more for the 2006 model than any single sensor. Federated and on-device learning removed the assumption that personalisation requires a round trip to a server. Compressed vision and diffusion models made local inference cheap enough to run continuously. Event-driven, neuromorphic-style sensing made always-on plausible without always-on power.
Together they moved wearables from collection to participation: the device proposes rather than logs. That is the part of the thesis I would defend unchanged.
Two premises survive: context is the interface, and privacy has to be architectural rather than contractual. The one I would rewrite is invisibility. I treated it as something that follows from good design. It is a physical property — of weight, heat, battery and price — and it is where these devices are still losing.
02Ubicomp, twenty years on · Part two18 Jun 2025 · 7 min read
Ubicomp at planetary scale
Wearables tested whether the framework was human-centred. Smartphones and 5G tested whether it scaled. It scaled — and picked up two failure modes that were nowhere in the 2006 text: attention and geolocation.
The thesis was scoped to rooms, buildings and campuses, because that was the size of a deployment anyone could fund. The phone made the deployment planetary without asking, and every assumption about context had to be renegotiated at four orders of magnitude.
The phone as sensor hub
A current handset carries fifteen or more sensors — accelerometer, gyroscope, barometer, magnetometer, GPS, depth — which makes it the most widely deployed ubicomp platform ever built, by a margin nothing else approaches.
Dartmouth's StudentLife study is the clearest demonstration: passive accelerometer and location traces, correlated against sleep, sociability and academic outcomes across a term. It confirmed the contextual-memory argument at a granularity I had no way to picture in 2006, using no instrumentation beyond what the participants already carried.
Invisibility can also mean intrusion
The framework's blind spot is attention. Studies put the average smartphone user somewhere near seventy interruptions a day, and the mechanism producing them is the same context-awareness the thesis was arguing for: the system knows something, so it speaks.
Montag and Walla's work on digital distraction describes users toggling between a physical and a digital self, which erodes exactly the contextual integrity that was supposed to be the payoff. I optimised for seamlessness and assumed calm would follow. Seamless turned out to be a good delivery mechanism for interruption.
5G won latency and leaked location
Sub-millisecond latency and network slicing did what they promised: per-context quality of service, so a medical wearable or a vehicle can hold guaranteed bandwidth while everything else degrades. This is the first infrastructure that treats context as a first-class network parameter.
The same precision is the problem. The overwhelming majority of mobile apps collect location, mostly without granular consent, and millimetre-wave positioning narrows that to sub-metre. The New York Times' 2018 investigation remains the reference: dozens of companies receiving nominally anonymous traces from roughly two hundred million American phones, some apps sampling thousands of times a day, with reconstruction of a named individual's clinic and gym visits taking a reporter an afternoon. Senator Ron Wyden's summary was that location data can reveal the most intimate details of a person's life. Nothing in the standards prevents that; consent screens are not a control surface at this resolution.
Coverage is not access
Detroit's community air-quality network is the case I keep returning to. Sensors were installed across the city and published live readings — to residents who in many cases had no smartphone to read them on. The instrumentation was equitable; the interface was not.
My framework assumed the endpoint was already in the user's pocket. Where it isn't, ubiquity means the data exists about a neighbourhood rather than for it.
Edge computing, and the energy nobody budgets
Pushing inference to the device is the right architecture on privacy and latency grounds, and it is now the default in wearable silicon. It is not free. Dense small-cell deployment raises the carbon cost of the same coverage substantially compared with the previous generation, and platform-level mitigations — throttling background work under grid stress — sit in the operating system while the network layer has no equivalent.
Almost no operator runs sustainability-aligned slicing, though the mechanism to do it already exists. If context can prioritise an ambulance, it can deprioritise a background sync during a peak.
Hierarchies of context
Projects connecting autonomous vessels, drones and vehicles across continents need context models that resolve from a room to a continent in one hierarchy. The 2006 model has no vertical axis; it describes a single scale well.
Two phenomena also need naming, because they are structural rather than incidental: temporal fragmentation, where activity splinters into micro-interactions, and digital dualism, where the physical and digital self run in parallel and disagree. Weiser said the most profound technologies disappear. They disappear until they need a charger.
Scale confirmed the design principles and falsified the sociology. Context-aware, privacy-first design still holds up. The assumptions that everyone would have a device, and that a quiet system would produce a quiet life, do not.
03Ubicomp, twenty years on · Part three22 Jun 2025 · 6 min read
Privacy, equity, energy: the bills coming due
The technical problems in ubiquitous computing largely got solved. Three others were left on the table, and they are harder because none of them is only technical.
Photo by Lianhao Qu / Unsplash
Wearables and networks moved the 2006 framework from paper into use. What they exposed was not a missing capability but three unpaid accounts: privacy erosion, digital inequity, and the energy cost of ubiquity.
Privacy: from defence to discretion
Continuous collection is the premise of the field, not a side effect. Devices gather location, physiology and behaviour at high frequency, and analytics re-identify individuals from traces that were published as anonymous. A useful way to audit this is dimensional: identity, location, digital footprint, query and inference each leak differently, and a system can be sound on one and open on another.
The Berkeley formulation of why ubicomp is different still stands. In a conventional system the user has some trust relationship with whoever administers it. In an instrumented space they typically have none, and often do not know the space is instrumented. Access-control lists are the wrong instrument for that situation.
The alternative is data discretion: granular, revocable, machine-enforced control over what is collected and who may compute on it, backed cryptographically rather than by policy text. Framed that way privacy stops being defensive and becomes a negotiation between producer and consumer of data — which is also the only version that survives contact with a commercial product.
Equity: the divide did not close
Ubiquitous computing was pitched as democratising. Membership and participation data in the field's own professional bodies say otherwise: representation from Africa, the Middle East, Latin America and low-income countries is marginal, and women, racial minorities and disabled researchers remain underrepresented in the rooms where these systems are specified.
There is evidence that this responds to design. When UbiComp/ISWC moved to income-based registration with travel support and targeted programming, it produced its most demographically diverse cohort. That is a solved logistics problem being described as an intractable cultural one.
Sustainability: a linear lifecycle at planetary volume
Devices are still specified for peak functionality and a short life: non-renewable inputs, a couple of years of use, then e-waste. At the volumes ubiquity implies, embodied carbon dominates the footprint long before anyone argues about standby power.
Circular interactive materials are the most interesting response — substrates that sense and actuate but are designed to be recovered, recycled or to degrade. Biodegradable sensors and energy-harvesting materials are early, and they attack the part of the problem that efficiency gains cannot reach.
What responsible ubicomp would require
Three commitments, none of which is a research problem: give people real-time granular control over their own data; widen who participates in specifying these systems; and design for recovery rather than for function alone.
None of that is expensive relative to what the industry spends on silicon. It is unallocated, which is a different failure.
The 2006 thesis argued that ubiquitous computing needed both technical and social science to work. That reads as the least controversial claim in the document and it is the one the field has done least about.
