See you at MCN 2026! Book a meeting

October 23, 2026·4 min read

Ethical Use of AI in Exhibit Design: A Case Study and Ongoing Conversation at the National Museum of the United States Navy

Brie Logsdon
Brie Logsdon
Co-Founder
Ethical Use of AI in Exhibit Design: A Case Study and Ongoing Conversation at the National Museum of the United States Navy

Art created by Matthew Love.

At AAM’s Annual Meeting in Philadelphia, Kyle Bernard, Director of Exhibits at the National Museum of the United States Navy, discussed how AI was used in the creation of its traveling exhibit, Navy 250: The Legacy. Bernard also outlined other ways AI has been utilized by the institution, and the times it fell short.

“The proliferation of artificial intelligence into our personal, professional lives has been astounding in the last few years,” Bernard said.

To create Navy 250: The Legacy on a short six-month deadline, the team expanded an existing exhibit called We Have the Watch, which focuses on the everyday sailor throughout history, highlighting that as much as the world and technology has changed, the core experience of being a sailor has remained the same.

Bringing history to life

One significant exhibit component is a 360-degree immersive theater. At 8’ tall and 20’ in diameter, the immersive theater was inspired by the Van Gogh and Dali experiences, both in scale and approach.

The theater was intended to be the “big hook” in attracting visitors who otherwise may not be interested in maritime history, Bernard said.

“Our thought was: What if we could put a visitor on the deck of navy vessels acting as engaged witnesses to history, as pivotal moments of American naval heritage have pulled all around,” Bernard said. “Could we use naval artwork in a way that hadn't been done before, that would engage visitors while still staying true to both the artist's intent and the historic record? Could we leverage navy archival photos and video in a new way?”

Enhanced Digital utilized a “fine-tuned” AI model based on Stable Diffusion at the time. Stable Diffusion is important to prevent model degradation, Bernard said, “or else AI beacons become a photocopy of a photocopy of a photocopy of a photocopy.” By the end of development, they had seven vignettes.

AI tools were used to activate artwork or photographs, stabilize film footage, color correct, and generate voiceovers.

Setting internal standards

One vignette features AI-generated voiceover narrating Captain John Paul Jones’ own after-action battle report, while activating waves, flames, and smoke of paintings from the Navy’s collection. There is also one image featured that is an AI-generated composite of all the original paintings. This project prompted long internal discussions about artist intent and interpretation.

“The movement of a portrait or the animation of smoke does not, does not fundamentally alter the intent of the work presented, nor does it change the interpretation of the battles,” Bernard said. “Also, when discussing historical accuracy, remember that the artists themselves were not at this battle, many not even alive when it took place to begin with. These works of art were an attempt at recreating history in a way to bring them to life for the public.”

In another clip, the team “unwittingly created deep fake audio.” After having a retired Navy Captain record audio for a series of recent video recordings, the team realized the audio was too poor quality to use, and an impending deadline meant there was no time to re-record. With the Captain’s permission, the team sent two other recordings of the Captain speaking to their partner Enhance to use as reference in recreating the poor quality recording. The result was a near-perfect recreation of his voice, and a high quality voiceover narration.

Considerations around these vignettes helped the team sketch the beginnings of an ethical AI use policy.

The AI sandwich

To prevent AI slop and hallucinations requires human input, verification, and editing in every instance of use, Bernard said.

“There needs to be a human on both ends,” Bernard said. “The AI sandwich. The AI sandwich concept served as our guiding principle during the initial foray into AI use in our exhibit design.”

Transparency has also been critical in every internal discussion and external document. The introduction panel also disclosed use of AI in the production of the exhibit.

“I argue that a majority of that risk comes from trying to pull one over on the public,” Bernard said. “Being upfront and transparent about use allows the public to make an informed choice on whether they want to interact with technology or not.”

Other AI experiments prove the need for this approach. An attempt to use Google’s Gemini 3.0 Pro to compile a list of Navy Medal of Honor recipients organized by date of action did in fact produce a list—one rife with mistakes. Incorrect ranks, dates, and service for nine of the first 10 on the list and beyond that there were completely fabricated names, untraceable to any U.S. service.

Future AI plans

Bernard said the biggest takeaway is that AI is a tool, not a solution. A tight deadline was the primary factor in using AI to support the creation of Navy 250: The Legacy.

“Looking back, there are edits I wish we could make, but overall, for the situation we were in, and the commitments made, the final production met and exceeded expectations,” Bernard said. “Will we continue to use generative media? Probably, but not images or video in exhibits or public-facing material.”

Additionally, the team is still discussing generative audio as an acceptable technique for the museum. If it is, is it only okay to employ it to do voiceover work for deceased individuals?

“The conversations continue,” Bernard said.

Be the first to build your own Zibit.