Memory Contamination Exhibition — Immersive Audio Guide

Challenge

Audio guides have long served the same purpose at exhibitions: borrow a device, press a number in front of a work, listen to the commentary. The interaction ends there.

“Memory Contamination Exhibition” is built around the Mandela Effect — the phenomenon of large groups of people sharing the same false memory. It asks visitors to test, with their own minds, how far memory can be trusted. Nobody stays a passive observer.

Pairing that exhibition with a passive audio guide would have been a mismatch. We started from a single question: could an audio guide pull visitors deeper into the show rather than narrate it from the outside?

Approach

No dedicated hardware. The whole experience runs in the browser on the visitor’s own smartphone, so there is nothing to lend, collect or recharge, and visitors use their own earphones.

It begins by asking for a photograph of the visitor’s face. We do not explain why — not yet.

Voice actor Yuki Kaji then narrates the exhibition chapter by chapter. Visitors select a letter shown beside each work and play the matching track, across 24 chaptered tracks in total. Captions run alongside the audio so the content stays followable where sound is difficult.

Just before the final chapter, the experience pauses. From there, the face taken at the start flows into an AI-driven sequence that connects each visitor to the exhibition’s story in a different way.

What remains is a single image that belongs to that person alone. What happens in between is best left to the room.

Technology

Consent designed ahead of the spectacle

Because the experience processes the visitor’s own face with AI, immersion and risk management collide head-on. We consistently prioritised the latter: the explanation of AI use, the consent flow and the link to the privacy policy all sit before the staging begins. Declining remains a real option, and that path branches into a version of the experience that never uses a personal photo.

The same experience in two languages

The English version is not the Japanese one with subtitles bolted on. Reception, chapter structure and narration are all complete in English. Narration is handled by “Soyogi Soyogi,” the voice AI character produced by Yuki Kaji himself — so the voice stays the same while the language changes.

Written to the exhibition’s own rules

Script and staging were tuned together with Yami Inc., the exhibition’s planner, applying the show’s own editorial rules to the web experience: how numerals are written, how chapter titles are set, how long a warning is allowed to hold before it moves on.

Results

At the Tokyo venue, it was sold as an audio-guide ticket at ¥3,290 against the ¥2,300 standard admission. Roughly 3% is regarded as an average take-up rate for audio guides at museums; this guide reached about six times that level (based on the full Tokyo run).

On social media, the most common reaction during the run was a strong recommendation to buy: “a hundred times more fun with it,” “absolutely get one.” Close behind were posts framing it not as an add-on but as part of the work itself — “this is part of the exhibition,” “this is the main feature.” Many mentioned the face registration and the ending while carefully avoiding spoilers.

The exhibition’s official account also posted that the guide, “including a certain device you can only experience here, has been very well received in our visitor surveys.”

What’s next

We want to bring this immersive audio guide beyond exhibitions — to theatre, live performance and other formats. Rather than fitting content into an off-the-shelf system, we design and build each one to the world and the pacing of the work itself.