On Wednesday, Feb. 18 Johns Hopkins Advanced Academic Programs hosted a discussion on AI and museums titled “Building Intelligent Museums.” The webinar, hosted online, focused on both AI policy for museums as well as how digital systems support AI-enhanced museum projects.
Karina Wizevich, associate program director and senior lecturer for the MA in Museum Studies program, opened the discussion. She introduced the topic of AI intersecting with the museum sphere and welcomed the event’s moderator, Dr. Joyce Ray, another Assistant Program Director and senior lecturer for the MA in Museum Studies program.
Ray then introduced the two main panelists: Nik Honeysett, now CEO of the Balboa Park Online Collaborative, and Alexandra Kron, a community manager at Terentia.
Honeysett, a former adjunct faculty member at Hopkins, was the primary lecturer for the first half of the event. He described his recent work, which focuses on AI in museums, and began with a discussion on whether institutions were willing to adopt AI tools in curatorial practice. Honeysett conducted a survey on this question and found that nontraditional individuals tend to be skeptical about AI.
“This is important when you think about it, because in any group of people, not only in an institution, a museum, you’re going to get this group of people, but also your audience is the same,” he said. “There are some people who are totally up for AI adoption and [who] embrace it, and there are others where it is the last thing they’d want to do.”
Honeysett then described the current scope of museums in relation to AI usage. Museums have for-profit corporate board members who embrace AI adoption. However, there’s a gap in which those very individuals in leadership lack the time necessary to delve into AI discourse. Honeysett continued on the idea of digitization as fuel for AI. He noted that museums should be adopting governance early with these systems, referencing Google’s search engines that influence how museum websites are trafficked.
“If you’ve been using Google certainly for the last four to six months, you’ll notice that when you search, rather than getting a set of results, you’re actually getting an AI interpretation,” Honeysett said.
This has had impacts on museum websites, where there was a spike in traffic per the AI “bots,” followed by a dip in that same traction because Google had taken away those AI possibilities. Honeysett expanded on this by explaining generative energy optimization (GEO), which is how websites are designed to be AI-friendly.
However, there are challenges with GEO. For example, Honeysett presented on a website called Living Museum. He described how a coder in Canada took the British Museum’s entire collection and attached an AI assistance tool to the service, available in any education level or language. While benefit is personalization, the drawbacks are nonnegligible — museums could lose control of their voice and historical accuracy.
“Museums need to be treating AI like a publication platform with all the editorial guidelines and goals and rigor behind it that they would do under any circumstances,” he said.
Green AI policy is also significant in the grand scheme of AI utilization. The goal is to use AI to enhance operational efficiency as well as visitor experience, all while minimizing the environmental impacts of its use. According to a chart presented by Honeysett, Google Gemini is the most efficient AI platform for the environment. Ultimately, he argued, AI is not a replacement for people, rather it is there to supplement human productivity.
“You are certainly seeing an overestimation of [AI’s] ability right now, but it is going to have a massive impact over the long term,” Honeysett concluded.
Kron then began her section, which focused on the role of digital asset management systems (DAMS) in building intelligent museums. She first outlined what occurs behind the scenes of AI tools, such as an engine that identifies dataset patterns, biased outputs and AI’s lack of context awareness. Products of AI are shaped by the context and data that we provide it with.
“As we're using those AI tools more and more, we're seeing a rise of what I'm going to call AI-enhanced projects, and by that, I mean projects where AI supports human work — not replacing it. In a museum context, this looks like generating or enriching metadata, suggesting keywords, possibly identifying people or places in images, transcribing handwritten documents, or converting audio and video into searchable text,” Kron said. “But AI's role isn't limited to efficiency. AI is increasingly also shaping search and discovery. It can influence which objects are surfaced, how materials are connected, and what patterns are becoming visible in your collections.”
AI in museums presents both opportunities and risks. With advances in efficiency, scale and insights, there are risks of a loss of context. The argument goes back to the fact that AI’s outputs are based on our own context. Kron referred to DAMS as a “truth anchor” for AI, suggesting that they can provide a home for files, metadata and version history. When AI operates within DAMS, it operates under governance. It can assist with tagging and transcription under institutional rules and human review. AI working outside the DAMS may rely on incomplete or decontextualized data.
Kron also highlighted the critical uses of DAMS based on institutional priorities. She called attention to The Theodore Roosevelt Presidential Library, a project in Medora, North Dakota.
“What makes this project really unique is that it began without a traditional collection... The library started with almost no physical materials on site and no existing digital infrastructure. So to build its collection, the team had to aggregate records and digital assets from more than 30 institutions, including Harvard, the Library of Congress, and the National Archives,” she said.
Like other projects, there were familiar challenges such as data consistency and rights agreements. The solution was to establish a single system of record, or a trusted AI digital repository that consolidates collections data, among other benefits. AI tools have archivists review outputs, approve confidence records and flag others for review. This performance will improve over time. The lesson, though, is that AI can accelerate projects only when it is influenced by strong governance and human oversight.
A question-and-answer session followed Kron’s demonstration. Honeysett deliberated on how museums can be more AI-friendly.
“The best training for an AI engine is a combination of data — so, collections data,” Honeysett said. “And there's some examples that we've done [in] a project with Strategic Air Command [where] collections data [was] augmented by recording volunteers giving tours around the institution and recording the questions that are asked of them and their answers.”
Another concern is education with AI tools. The solution is increased utilization. AI will not replace jobs, but those with more skills in using those very platforms will. This conversation extends into job applications, where there is a shift in making a resumé more AI- than human-friendly.
The panel concluded with a discussion of museums that are skeptical about using AI tools. The consensus was that being informed on AI itself and learning how to gain the most out of its advancements are key. When interacting with AI, they argued that the most effective approach is acting as a performance director, emphasizing a controlled environment and understanding how these engines are designed in the first place.
