While I hesitate to amplify, I find it harder to sit idly in the face of another article touting artificial intelligence as a panacea to art valuation. Moreover, this salacious column is predictably written by someone (Daniel Grant) with no firsthand experience buying, selling, or appraising art, and it is perhaps unsurprisingly riddled with misinformation, inaccuracies, and elisions.
As a small selection:
1. Any responsible mention of the use of AI to authenticate Old Masters should be accompanied by an acknowledgment of its severe limitations and its potential for misattributions and dispute (lest we forget the Raphael episode).
2. Appraisers do not "set prices" -- we set ascribe values. If Mr. Grant does not know the difference, he should not be writing such an article.
3. The statement that "only a small percentage of artworks are sold at public auction" is misleading. A very significant percentage of high-value works of art are sold at auction.
4. Mr. Grant asks how insurers are "supposed to write fine art policies for collections if the appraisers they rely on have little or no access to the prices paid for artworks." It's patently absurd to contend that appraisers have "little or no access" to private realized price data, when in fact it's precisely our job to research and report on such markets. All good appraisers do this -- and insurers know (or should know) to rely on valuations performed by those appraisers. Technology will not help with this.
5. Mr. Grant uncritically quotes an AI professional who says that "A.I. can understand why and when the value of an artist's work changes over time" However, Mr. Grant curiously does not quote a single appraiser. As a reminder, AI is only as good as what it is fed. AI has neither relationships nor private knowledge, nor any real intelligence in making the necessary decisions to such a pursuit as art appraisal.
etc., etc., etc.
Wikipedia is not a reliable source
Today I discovered that Wikipedia does not have a page on “art appraisal.” They do have one on "art valuation,” though its contents, perhaps not surprisingly, are stilted.
In the first paragraph, its author-less voice proclaims that art valuation “is more of a financial rather than an aesthetic concern” — as if our financial valuations of art are not defined by well-honed “aesthetic concern,” usually over decades. The reader is then informed that our “subjective views” play a part as well, when in fact we are compelled by USPAP to perform our valuations in an independent, impartial, and objective manner. [NB: Opinions of value may vary (even significantly) even among seasoned appraisers, but any USPAP-compliant appraiser is bound to perform valuation services in this manner.]
Also worth adding, there is no Wikipedia page on the Appraisers Association of America, the premier association of personal property appraisers who focus on fine and decorative arts. There are two other major associations of appraisers, namely the International Association of Appraisers (ISA), which is also not represented on Wikipedia, and the American Society of Appraisers, which is represented by a brief Wikipedia page with no mention of art appraisal.
Why should we care? We all know that Wikipedia is not a reliable source, and we yet we all use it to some varying degree for expediency, trying to block out how fundamentally flawed it is. Misrepresentations and omissions of content are already a problem when it comes to personal consumption of information, but the problem is now compounded in the age of AI, which scrapes from all corners of the Internet, including Wikipedia.
What if you asked your AI tool how art is valued? What if it scraped that page, and then you used it in a professional capacity? What if you asked it who the people are who make their living by valuing art?
There are clearly vastly larger problems on Wikipedia with regard to bias, misrepresentation, and exclusion, but this is my corner of the world, so it is my example to highlight with respect to the potential for promulgation of misinformation. Yes, we should edit Wikipedia to rectify this tiny injustice, but the larger principle will remain that today, with AI scraping Wikipedia, it is nothing short of an echo chamber of unattributed statements, many of which are plainly wrong.
AI in Art Appraisal (part 2)
The recent episode (# 162) of The Appraisal Foundation’s “Appraiser Talk” deals with the topic of AI in appraisal. It is worth a listen, though I disagree with most of what the guest, technophilic real-estate appraiser Jim Amorin, asserts, above all, his comment that AI is “like having a crystal ball for the market.”
AI is only as good as as what it scrapes, and we don’t have market data from the future. Appraisers do analyze trends. And yes, technology can help organize data that shows past trends, which at times may provide insight into forecasting future outcomes — though such forecasting may often be tangential to an appraisal assignment with the given task of assessing a past or present value.
While such propensity for pattern recognition is presumably what emboldens Mr. Amorin to feel like he has a crystal ball, one must worry that the machine is not always right, either by virtue of having been trained on an insufficient quantity or type of data, or, critically, because artificial intelligence, certainly in its present form, lacks the nuance for judgement of real human intelligence.
Mr. Amorin’s assertion that “these AI algorithms excel at detecting patterns that might not be immediately obvious to the human eye" has proven, and will, I believe, continue to prove untrue. Perhaps the most notorious recent example was the AI / facial-recognition-driven attempt at attributing a would-be Raphael based on the machine’s detection of a “97 percent similarity” between the figure of this painting and the de Brécy Tondo — the attribution was subsequently questioned by another AI-based team.
Indeed, facial-recognition software might rightly perceive a 97% percent visual similarity between none other than the Mona Lisa and a well-painted copy (of which there are many), but the the 3% difference would presumably cause a 99%+ difference in value.
One must question whether AI, in its present form is at all “very good at analyzing patterns and irregularities,” as Mr. Amorin contends. On the subject of Leonardo, the example in the illustration above came up in the Photos app of my computer — and yes, this is, in fact, a kind of artificial intelligence. The machine’s error of mistaking the Salvator Mundi with a Frida Kahlo self-portrait is certainly funny, but the absurdity of this mistake —ostensibly resulting from insufficient data and training in the model — is also a keen reminder of how flawed, at base, these systems are. (Should we read some deeper meaning into the fact that Photos cropped out the Salvator Mundi’s crystal ball? )
[NB: Similarly problematic statements riddle Mr. Amorin’s self-published book The Generative Shift: Preparing Appraisers for Artificial Intelligence Models Like ChatGPT. While one might be tempted to dismiss a book that has not undergone due editorial process or peer review, it is alarming to see that Amorin’s voice was regarded as relevant enough for his inclusion in the Appraisal Foundation’s recent convocation, “Artificial Intelligence & USPAP Shaping Future Standards and Ethics Forum.” I also recognize that some aspects of real-estate appraisal may differ from those of personal-property appraisal, but we follow the same set of professional standards, and my concerns remain.]
AI Art Valuations
I am hardly alone among appraisers to be blistering over Daniel Cassady’s article published in ARTnews last week, initially titled “AI Art Valuations Are Starting to Give Some Market Players an Edge,” then quietly subsequently changed on the website to “AI Art Valuation Companies Think They Can Give Market Players an Edge.” [NB: Notwithstanding the title change, almost all of the article remains intact, and the initial title is unchanged on the magazine’s Instagram account.]
It is extremely problematic to peer into our profession, as Mr. Cassady does, and blithely suggest that certain appraisers have an “edge” over others, and further, that appraisers using AI can “appraise works more quickly and accurately than ever before.” Mr. Cassady offers no grounding for such sweeping claims about the relative speed or accuracy with which we do our jobs. Instead, he relies on unsubstantiated testimonials from a small group of tech-forward actors who appear to be advocating for their own business interests.
To the tech-as-panacea crowd, there is seemingly always an innovation to be hailed as transformative, but art appraisal relies on real—not artificial—intelligence, diverse lived experience in the art world, networks of relationships in an array of allied professions, and a large body of knowledge cultivated over time.
In addressing the use of technology in appraisal, Mr. Cassady should offer concrete examples of how, if at all, the blockchain, AI, or other technology can meaningfully augment appraisal. He should also address the major limitations of technology in a field in which individual expertise is critical. Mr. Cassady does his readers a disservice by failing to include commentary from relevant professionals who feel that the above-mentioned technology does not meaningfully augment apprasial, and may be a superfluous business cost.
The most egregious issue in the article is, in my view, Mr. Cassady’s uncritical use of unreliable sources, above all, appraiser Carolyn Taylor. Mr. Cassady writes: “Along the way, I became an appraiser and was shocked at how that industry ran,” Taylor told ARTnews. “After a few years, it became very clear there were no neutral appraisers. Every appraisal firm is also a dealer.”
Ms. Taylor is also a member of the Appraisers Association of America. As such, she must be compliant with the current (2024) edition of the Uniform Standards of Professional Appraisal Practice (USPAP), which states in its “Definitions” section that an appraiser is “one who is expected to perform valuation services competently and in a manner that is independent, impartial, and objective” (p. 3). The language of neutrality is found throughout USPAP. The “Conduct” section of the Ethics Rule of USPAP reiterates that an appraiser “must perform assignments with impartiality, objectivity, and independence, and without accommodation of personal interests” (p. 9).
It is not only wrong but frankly unprofessional for Ms. Taylor to assert that there are “no neutral appraisers,” when, in fact, all serious professional appraisers in this country, herself included, have in essence, taken a vow of neutrality with respect to the works they are appraising, by complying with USPAP.
An appraiser may indeed have commercial interest in works not being appraised at the time. Many professional appraisers also work in an advisory capacity, acting in the interests of their clients for purchases and/or sales; these assignments, when undertaken properly, do not compromise the appraiser’s “neutrality,” nor do they necessarily make these parties “dealers.” It should be noted, further, that many appraisers do not work in an advisory capacity, nor as private dealers.
Mr. Cassady errs again when he uncritically cites Taylor a second time: “Appraisal Bureau’s neutrality has fostered strong relationships with galleries, according to Taylor, who are more willing to share data with a company not involved in sales.”
Mr. Cassady offers no evidence to support this statement, nor is the assertion balanced with comments from appraisers whose experience proves otherwise. I, for example, routinely work with my professional gallery contacts – or make new ones – to learn about relevant private sales, as necessary, to give credible, accurate valuations. The fact that I also assist clients in an advisory capacity for other works does not impact my neutrality as an appraiser, nor does it compromise my ability to secure the necessary data for an assignment.
Appraisal Bureau’s claimed “neutrality” is not in fact a point of difference in a profession in which all reputable appraisers have vowed impartiality, and to suggest as such, is misleading. Mr. Cassady should have further researched this matter before using a source who apparently distorted the truth to promote her own business in a quotation.
As for the larger question of the use of AI in art appraisal, I have much more to say about the severe limitations of its applicability. More to come on this topic.