AI in Art Valuation: The Promise and the Pitfall

Chris Mann, Head of Data

The rapid progress of generative AI has generated significant discussion across the art and finance sectors about how these systems might support valuation and appraisal processes. Recent tools such as ‘Appraiser GPTs’ show impressive capabilities: For example, retrieving auction results directly from auction house websites.

However, recent testing comparing platforms such as ValuerPro and ChatGPT demonstrates that these systems are still not ready for professional use in the valuation of artworks. While they can offer useful insights for lower-value artworks (typically under USD 30,000–50,000), the outputs often contain inconsistencies such as incorrect auction data, outdated sales results, and, in some cases, entirely fabricated or ‘hallucinated’ information. These inconsistencies are most concerning when managing artworks of significant value, or if being used for business-critical decisions, such as for art-secured lending or insurance underwriting.

These issues underscore a fundamental challenge in applying large language models to valuation: ensuring factual accuracy. Valuations depend on data that is:

·      Accurate: Verifiably true and current. Data needs to reflect current market conditions.

·      Grounded: Sourced from external, authoritative databases rather than the model’s internal patterns.

At Overstone, we make extensive use of agentic AI systems and traditional machine learning across a range of operational, analytical, and data-management processes. These technologies help improve efficiency, data connectivity, and insight generation throughout our platform.

However, when it comes to critical tasks such as valuation, where 100% accuracy and auditability are essential, we do not deploy large language models. All valuations are completed by our team who have decades of combined experience at international auction houses, as well as training in how to conduct USPAP compliant valuations. Our valuations use structured, verified market data and we can provide commentary to elucidate how we come to our suggested valuations.

We continue to monitor the development of generative AI closely and recognise its potential in research, workflow automation, and information retrieval. But for now, the valuation of artworks must remain grounded in transparency, precision, and factual integrity - the same principles that underpin Overstone’s approach to every aspect of our technology.

For a free one-hour consultation for Family Offices and PWMs,

please contact: John Jessen: jjessen@overstoneassociates.com