A lovely article in The Economist this week: AI agents lie, cheat and steal. That is putting off users. The article makes a point that is perhaps obvious: AI needs to be reliable to be useful. E.g., omitting the second-highest result when asked for the top 10 — the example used in the article — is bad when answers matter. The article argues for the importance of the “trust layer” for the AI, to facilitate its adoption in business-critical use cases.
This argument translates directly to wearable AI, i.e., smart glasses, and augmented reality as a whole. We accept mistakes, like visuals that are incorrect or misplaced, in research prototypes and in games — and even there, only up to a point. No one enjoys being instructed, e.g., to reach for a virtual element that is out of bounds or to press a virtual button that is not getting pressed no matter how many times you poke at it. But wearable AI and AR are envisioned to do far more than games: support human decision-making, guide humans through complex physical-world procedures. A “trust layer” is essential for getting those types of applications off the ground.
Trust layer is essential in many day-to-day applications of AR; it is especially important when AR is used to mediate human-robot collaboration. Photo from an AR-mediated HRC interface we developed as part of our recent work, that is appearing in IEEE RO-MAN 2026.