Currently reading: Corbin et al. (2026) “On AI glasses and wearable AI in assessment”
One of the seemingly most promising responses to GenAI being misused in assessment situations has been to (try to) physically exclude technology by creating assessment situations where the assessor observes what the student is doing (like in face-to-face oral or proctored assessments). But what if accessing GenAI does not require a computer or phone, but devices are tiny and practically invisible?
Corbin et al. (2026) talk about “dual transparency“, meaning that GenAI use is both transparent to users (because text is projected on lenses of a pair of glasses teleprompter-style so that it can be read while looking through it at whatever else someone was looking at, e.g. an assessor) and transparent to observers (because in order to use it, the user does not have to shift their gaze or do any other tell-tale moves). Devices can take the form of smart glasses, ear pieces, rings, or probably any number of small objects, yet they might be able to listen in and film, and output in ways that are hard to detect. Plus seeing someone wear smart glasses does not mean that they are using them in their smart function, they might just be their normal glasses… So what to do?
Like so many problems in education, GenAI and assessment is also a wicked problem, which by definition does not have one best solution, but always involves balancing multiple trade-offs. Corbin et al. (2026) discuss, for example, how physically separating students from their devices in face-to-face oral examinations introduces biases related to how well students perform orally and/or under pressure. The pedagogical and equity costs associated with that might be acceptable or not, depending on a bunch of different considerations and values, and also on whether it is actually working. And that depends on separability (actually being able to binarily separate GenAI use from non-use) and observability (that we can actually clearly tell whether GenAI is being used).
What separability means in practice
AI wearables are potentially great as accessibility devices — that its practically impossible to observe that they are being used, so people do not have to disclose disabilities or support with foreign languages or navigating a new-to-them city — and they can become part of what people perceive as themselves. Like my glasses — they are not an object to me, they are how I see the world. If me seeing the world, for example, always had a subtitle translating Swedish conversations around me into English or German, that would very quickly become something I don’t notice any more, it would just be how I see the world. The smart function of my glasses would have become transparent to me, “withdraw[n] from the wearer’s awareness“, in ways that accessing translations on my laptop or phone never can (since they require the device that isn’t always in my field of vision anyway, a separate interface, …).
What observability means in practice
Something having become transparent to the wearer does not mean that it is transparent to everybody else. Corbing et al. (2026) use the example of a blind person’s cane, which they might not be aware of using (similar to me usually not being aware of wearing glasses until I am not wearing them, someone knocks into them, or they are fogging up), but to many other people, it is obvious I am using them. Whether something was projected on the lenses, however, would be not at all obvious to others, whereas it would be possible to observe if I was accessing translations on my computer or phone.
Dual transparency and the challenge it poses
Corbin et al. (2026) define dual transparency as “[t]he condition in which AI is simultaneously incorporated into embodied activity and not readily detectable to observers“, and predict a second wave of GenAI challenges to assessment. The first one, they argue, was about redesigning tasks and rewriting policies to deal with students using an external device. This second one, now, is different in three ways. Wearable AI
- “does not function as a tool the user consults but as a capability the user inhabits” (like my glasses — most of the time I am not aware that I cannot actually see as well myself as I can with my glasses)
- “will be worn routinely, by many students, for reasons that have nothing to do with assessment” (for example accessibility)
- “has the structural conditions for cognitive integration: hybrid practices in which bodily action, representational systems, and learned norms of use become part of how tasks are understood and accomplished” (we don’t know yet how that will play out, but it seems likely that if people are used to the support of their wearable GenAI things the way that I am used to wearing my glasses, they just become part of how they learn)
So in a nutshell, Corbin et al. (2026) write “[t]he first wave [of challenges caused by GenAI] asked whether students were delegating tasks to AI. The second wave asks whether students are developing capacities that are hybrids of themselves and AI, capacities that cannot be cleanly attributed to either. If they are, then assessment tasks aimed at separating them fail to make ongoing sense.”
What if we prohibit wearable AI, or even enforce that it be removed?
Prohibiting wearable AI is the first reaction in many places as a way to protect existing assessment formats through putting a new rule in place. But of course that only works if we can actually detect non-compliance (if we cannot actually detect non-compliance, some students might comply and others might not, and I would strongly suspect that there is a bias in who complies and who does not).
If we then wanted to enforce the rule, we would have to limit which “bodily attachments may enter the room“, which poses the problem of how we identify what we allow and what we don’t. This only works if the assessor (or someone) can identify possible AI-capable devices and then also whether they are AI-enabled, and neither is easy at all, especially when a device has accessibility or therapeutic function in addition to whatever use of GenAI the university wants to prohibit. And things become really problematic if we require disclosure of disabilities or all kinds of personal data to distinguish between what we think wearable AI can be legitimately used for and what not (who are we, and is anyone, to judge?). Corbin et al. (2026) warn of a regim of “bodily adjudication“, which changes what is being scrutinized (an artifact or someone’s body), whose authenticity is being questioned (that of a piece of work or of a person’s performance), and who carries the burden (“those students whose relationship to body-worn objects is least optional“). So ultimately, physical exclusion of GenAI-capable devices is not a viable solution any longer (if it ever was).
So what now?
Yeah, I wish I knew. To me, the most important take-away from this article is how AI wearables can become part of what people perceive as their body and their own capabilities, and that even if we could reliably make people take things off for examination situations, that would come at a huge cost for those who rely on those devices in their daily lives, whether as assistive technology, for therapeutic purposes, or just because it has become part of how they perceive the world and interact with it. How ubiquitous will AI wearables become? Probably more than we can currently imagine (although I shudder to think what consequences that has when, for any reason, the underlying GenAI then suddenly isn’t available, or hacked). If I relied on reading translations on my smartglasses in conversations with my partner, and suddenly those were gone, what then? We live in interesting times…
Corbin, T., Sharpe, S., & Dawson, P. (2026). On AI glasses and wearable AI in assessment. Assessment & Evaluation in Higher Education, 1–17. https://doi.org/10.1080/02602938.2026.2661367
Last night’s walk…
