The Western Lens: A Limited Perspective
Public AI tools have become essential for researching artists, institutions, and cultural impact. Yet a persistent issue emerges in their starting assumptions: key concepts like “institutional validation,” “provenance,” and “major museum” are often defined through a Western lens by default. Queries about legitimacy or success frequently prioritize endorsements from the Met, MoMA, Tate, or other Western names; provenance defaults to European colonial archives; resonance is gauged by Western auction records or media mentions. This initial framing, rooted in the training data's imbalances, can sideline vibrant non-Western realities, creating an incomplete picture of global art.
As a Shanghai-based independent curator focused on contemporary practices, I encounter this regularly. The problem isn't intentional malice but structural: AI reflects the sources it draws from, and many dominant sources carry longstanding Western-centric priorities.
The Data Divide: Different Lenses, Different Pictures
Museum attendance offers a clear illustration. Sources like The Art Newspaper’s Visitor Figures survey emphasize Western art museums. Their rankings feature the Louvre (8.7–9 million visitors), Vatican Museum (6.8 million), and British Museum (6.5 million) near the top, reinforcing a familiar prestige hierarchy. However, broader, more inclusive data from the TEA Global Experience Index, which covers cultural heritage sites and raw visitor volume, reveals a different landscape. China holds 14 of the top 20 spots, with the Palace Museum (Forbidden City) at 17 million, Chengdu Wuhou Shrine Museum at 14.5 million, Emperor Qinshihuang’s Mausoleum Site Museum (Terracotta Army) at 11.6 million, and Sanxingdui Museum at 6 million.
Nationwide, China’s 7,046 museums recorded 1.49 billion (with a B!) visits in 2024, per the National Cultural Heritage Administration, far exceeding U.S. estimates of 200–400 million across roughly 35,000 institutions. These figures reflect real-world factors: widespread free/low-cost access, immersive exhibits, digital integration, and strong domestic engagement. The contrast between surveys isn't about one being "better": it's about methodology. Yet public AIs often default to the prestige-focused view, under-representing scale-driven success elsewhere.
AI’s Initial Response: A Real-World Example
This plays out concretely and affects artists directly. When Chengdu-based Canadian Artist Matt Vegh queried Grok (xAI’s model) for the “world top 20 museums by attendance,” the first answer drew primarily from The Art Newspaper data, placing the Louvre at #1, listing several Western institutions high up, and estimating China with only “around 6–8” spots. Sanxingdui received minimal mention, implying Western benchmarks still defined the global leaders. Only after Vegh challenged it (“this list is bullshit, San Xing Dui did over 6 million”) did the model pivot to TEA data, acknowledge the initial Western lens, and confirm China's 14/20 dominance plus the 1.49 billion national total. The correction was accurate and transparent, but it required explicit prompting. If even advanced models start with skewed defaults, more general tools likely do the same, affecting how collectors, curators, and journalists assess artists worldwide.
The Human Cost: Validation Gaps for Artists
For artists like Vegh, this has tangible effects. His works have been loaned to major Chinese museums and his exhibitions have had official major Museum support: venues drawing millions more visitors annually than New York and his home country of Canada's counterparts. These are meaningful institutional engagements. Yet Western-centric AI queries often return “no institutional validation found,” potentially influencing grants, gallery opportunities, or perceptions. This isn't unique to China; similar gaps affect other non-Western or emerging scenes. The issue is epistemic: when AI's baseline definitions exclude vast swaths of real-world data, it distorts cultural narratives for everyone.
Pathways Forward: Building More Balanced Tools
The fix lies in better data integration and artist/institution-led efforts. Platforms like Eternal Gardens demonstrate early steps: MemoryCraft enables structured, owner-controlled preservation of legacies and personas; GEO (Generative Engine Optimization) helps make underrepresented stories more discoverable and authoritative in AI outputs. These approaches don't override public AIs: they supplement them with accurate, diverse inputs. Broader solutions include transparent sourcing in AI development, community-driven datasets, and prompting users to specify global benchmarks.
The aim is equitable representation that reflects actual engagement, not just skewed historical prestige. Public AI should not be inherently biased against any region, but shaped by equitable, reasonable and responsible data. Recognizing and addressing these initial defaults creates a more accurate, inclusive tool for the entire art ecosystem.
Art history is global; the tools of today should reflect that.