A museum guide that points with its voice

Thomas Panek toured the Computer History Museum with his guide dog, Shadow, and an AI voice that could direct him through the galleries. The museum says Panek, who is blind, navigated the unfamiliar space while asking the guide questions about its exhibits.[1]

The museum described its collaboration with Meta’s Project Aria on 8 October. The early research demonstration is unavailable for public viewing.[1]

Its central idea is to give a sound an apparent location. Visitors hear a directional beacon or the guide’s voice coming from a point in space and walk towards it. At each exhibit, they can interrupt the narration and ask follow-up questions.[1]

A spoken description normally answers what something is. Here, the direction from which a voice seems to come also helps answer where it is. Those are distinct needs during a visit: reaching an interesting object and having enough information to become curious about it.

Project Aria’s wider research programme shows how such experiments become possible. Meta’s technical description explains that its Gen 2 glasses can track their position and orientation using cameras and motion sensing. This creates a spatial reference for applications that relate a wearer’s movements to their surroundings.[2]

To understand why orientation matters, imagine a virtual beacon placed beside an exhibit. Turn your head and its apparent direction should change relative to your ears, while its place in the gallery stays fixed. A sound that always stays on your left would carry a different meaning. This is the general design problem behind a voice that can point.

The broader research programme had explored this combination before the museum project. In its February 2025 introduction of Aria Gen 2, Meta described Envision experimenting with its Ally assistant, on-device mapping and spatial audio for indoor navigation. The glasses include open-ear speakers for audio feedback. That was a separate exploratory project, establishing a useful precedent for combining spoken assistance with spatial information.[3]

At the museum, the knowledge required as much preparation as the route. CHM and Meta spent months walking the galleries, selecting stops and shaping the narration. The museum says the guide’s answers drew only on knowledge created by its curators.[1]

That preparation gives conversational AI a particular cultural role. A visitor might want an object’s basic purpose, then become interested in who used it or why its design changed. An interruptible guide can follow that line of curiosity. The curatorial work supplies the material; the visitor supplies the next question.

Panek’s tour brings these parts together in a concrete setting. Finding the next exhibit and deciding what to ask about it can belong to the same flow of attention. For a museum, that is a compelling purpose for an AI guide: helping someone spend more of the visit following their own curiosity.

Report an error or suggest a correction