From being kicked out of a patient's room to 98% acceptance: a working method for the human system that lives beyond your design system.
Beyond design systems: designing for robots and seniors ft. Anna Oh (Norbert Health) | Config 2026 · Anna Oh
19 min total·Actually worth watching closely: ~11 min·3 must-watch clips
- 0:11 – 3:36Skim
Why seniors, why now
Lays out the crisis of presence with numbers: by 2030 more than a billion people will be over 65, outnumbering children under 10 for the first time in human history, and in many facilities one nurse takes care of 40 patients every shift.
The people who need AI the most are often the least able to adapt to it — the starting point for every design decision in the talk.
The visuals are mostly demographic and aging-pace slides (two key charts at 126s and 156s); a glance at the numbers is enough, the argument carries on audio.▶ Jump to 0:11Speaker · Anna Oh - 3:36 – 6:06Skim
Norbert and the first failure
Introduces the four-sensor edge AI device that turns any off-the-shelf robot into a clinical robot, then the outright failure of the room-to-room plan: patients were scared, and someone kicked the robot out and closed their door.
A nursing home room is a home residents have lived in for years; technology that walks in without a relationship gets refused — neither form factor nor models open that door.
The hardware shot at 217s is worth a glance; the failure story is spoken narrative — skim the visuals, listen closely.▶ Jump to 3:36Speaker · Anna Oh - 6:17 – 9:40Watch
The teddy bear wins: three ingredients of trust
Six rounds of face testing with more than 40 patients: the teddy bear won every single round and every patient picked the 3D face over the flat one, yielding three ingredients of trust — familiar, clear, warm.
Depth made low-vision residents say "it looks like it can hear me very well" — clarity isn't screen resolution, it's legibility at the level of perception; and the cuteness was really about feeling safe.
Face and form comparisons and test results cluster at 348s, 378s and 441s — without the visuals you can't feel the differences between forms or why the teddy bear won.▶ Jump to 6:17Speaker · Anna Oh - 9:44 – 13:29Watch
Recognition and perception: building shared reality
The first two layers of the relationship model — recognition creates identity, perception creates shared reality — centered on how messy face recognition gets in a real high-stakes environment, and the humans-in-the-loop design where operators become Norbert's eyes through a console.
Most of the team's time went not into the model but into building tools that keep teaching it what isn't a patient's face — the real world and the demo are two different stories.
The misdetection examples at 703s (a face on a TV, a wipe, a chair read as patients) and the operator console at 784s are the core evidence here — you need to see them.▶ Jump to 9:44Speaker · Anna Oh - 13:32 – 16:48Watch
Expression and continuity: reaching 98%
The last two layers: expression builds emotional connection through smiles, waves and mirroring; continuity teaches the AI where it is in the relationship — introducing itself on day one, skipping the introduction by day five, apologizing on a third visit the same day. In the end, 98% of patients accept Norbert.
Continuity can be designed concretely: when the AI is the one initiating the interaction, it has to remember how far the relationship has come.
The eyebrow and cheek expression iterations at 813s and the live mirroring at 878s are a direct demo of the Disney-style small signal that says "I'm with you" — hearing it described is nothing like seeing it.▶ Jump to 13:32Speaker · Anna Oh - 16:51 – 18:56Listen
Closing: what is your human system?
The industry is optimizing for speed, but the people who need AI most need more time to trust. The core claim: the right abstraction for physical AI is not a design system but a human system — not a soft skill, but the foundation the technology depends on.
Without acceptance there is no data and no measurement; what a human system finally produces is a relationship someone wants to continue.
The ending is pure spoken argument and reflection with no key visuals — fine to listen to while doing something else.▶ Jump to 16:51Speaker · Anna Oh