Can ChatGPT Replace a Curator: The Nasher Museum’s AI Experiment

What happens when art meets algorithm? This question came to life at the Nasher Museum of Art at Duke University, where the curatorial reins were handed over to ChatGPT for a groundbreaking exhibition.

Spearheaded by the museum’s innovative team and faculty from Duke University, this project delves into the uncharted territory of AI-driven curation, exploring its possibilities and limitations in the traditionally human-centered world of art.

A Curatorial Experiment Begins with a Joke

The story of the AI-curated exhibition began almost by accident. Faced with an unexpected gap in their exhibition schedule, the Nasher’s curatorial team jokingly suggested that Artificial Intelligence could curate their next show.

What started as a lighthearted remark quickly evolved into a full-fledged experiment, pushing the boundaries of what was thought possible in museum curation.

Marshall Price, Chief Curator at the Nasher, recalls the initial proposal: “We thought, why not? It was an opportunity to explore AI’s potential in a space that has always been dominated by human intuition and expertise.” Alongside Julia McHugh, Director of Academic Initiatives, and Mark Olson, Associate Professor of Art History and Visual Culture, the team decided to let ChatGPT, a large language model developed by OpenAI, take the lead.

The AI Takes the Lead

From the outset, the Nasher team approached the project with an open mind and a spirit of experimentation. They began by simply asking ChatGPT to curate an exhibition using the museum’s collection. However, it quickly became apparent that the AI couldn’t access their publicly available online database directly, resulting in nonsensical suggestions of artworks not in the Nasher’s collection.

Recognizing the need for technical expertise, the team brought in Mark Olson and his students, who specialize in emerging technologies. With a combination of AI tools including LangChain, Streamlit, and ChromaDB, Olson’s team developed a custom chatbot that allowed ChatGPT to interact with the Nasher’s database in a meaningful way. This complex process involved translating the museum’s data into a format that the AI could understand and work with, essentially teaching the AI to ‘read’ the collection in its own language.

Crafting the Exhibition: Utopia, Dystopia, and Dreams

With the technical groundwork laid, the Nasher team tasked ChatGPT with curating a cohesive exhibition. They asked the AI to suggest themes suitable for a university art museum. Among the options, the AI frequently gravitated towards concepts of utopia, dystopia, the subconscious, and dreams—abstract themes that resonated with both the AI’s capabilities and the museum’s willingness to explore unorthodox ideas.

The resulting exhibition features 21 works, including familiar pieces like Salvador Dali’s “The Obsession of the Heart” alongside lesser-known artifacts from the Nasher’s collection, such as ancient American stone figurines and ceramics. Surprisingly, even though ChatGPT could only rely on textual data rather than visual analysis, it managed to make some intriguing formal connections between the artworks, such as the thematic links between Dali’s entwined figures and those in other selected pieces.

However, not all selections were immediately obvious or relevant. The inclusion of ancient artifacts initially puzzled the curators until they realized the AI might have associated these pieces with burial practices and afterlife concepts, loosely fitting them into the broader themes of utopia and dreams. These moments highlighted both the strengths and quirks of AI curation, revealing an underlying logic that sometimes diverged from human intuition.Navigating AI’s Challenges: Hallucinations and Missteps

Throughout the process, the Nasher team encountered the phenomenon of AI “hallucinations,” where ChatGPT generated inaccurate or misleading information. This included assigning incorrect accession numbers or descriptions to artworks, such as labeling a Dorothy Dehner painting as a sculpture. These errors underscored the limitations of AI and the importance of human oversight, particularly when AI bridges specific institutional data with broader internet sources.

Another challenge arose when ChatGPT attempted to sequence the exhibition layout. The AI suggested impractical setups, such as placing works in secluded niches or under dramatic lighting that the gallery couldn’t accommodate. Ultimately, the team had to balance AI’s creative suggestions with the physical realities of the exhibition space, blending AI-driven ideas with practical curatorial expertise.

The Human Touch in a Machine-Driven Curation

While ChatGPT played a significant role in curating the exhibition, the Nasher team quickly realized the indispensable value of the human touch. The AI not only selected the works but also generated introductory texts and object labels. However, its writing style often lacked the critical nuance of traditional curatorial commentary, favoring a promotional tone more suited to a travel brochure. To provide context and correct inaccuracies, the museum added human commentary alongside AI-generated text, creating a layered interpretative experience for visitors.

The exhibition sparked diverse reactions from visitors, including students, faculty, and tech enthusiasts. Feedback collected through surveys revealed a range of opinions on AI’s role in museums. Some praised the AI’s fresh perspective and the thought-provoking nature of the project, while others expressed skepticism about AI’s ability to capture the empathy and depth required for curation. “Curating is more than just selecting works,” noted one visitor. “It’s about storytelling and connecting with audiences in a meaningful way—something AI still can’t fully grasp.”

Looking Ahead: The Future of AI in Museums

As the landscape of AI evolves, so too will its applications in museums and other creative fields. The Nasher Museum’s AI-curated exhibition serves as a compelling case study, highlighting both the potential and the pitfalls of integrating AI into the arts. It underscores the importance of critical engagement with technology and the need for human oversight to ensure accuracy and meaningful curation.

Reflecting on the project, Marshall Price emphasized that while AI is unlikely to replace curators anytime soon, it offers intriguing possibilities for collaboration and experimentation. “This experiment showed us that AI can be a powerful tool, but it’s the human element that remains essential,” he said. “We’re excited to see how we can continue to explore this intersection of art and technology in the future.”

–

you are interested in learning more about how Museums can use Artificial Intelligence? Join us for the Museum AI Summit 2025.