As ChatGPT is appearing in more and more conversations I have, I thought it was time to address the topic in a newsletter. Most of us are experimenting with the possibilities of A.I. – and rightfully so. A couple of years ago I was already informed by an expert at a lecture on the subject that ‘you will not be replaced by A.I., but by someone who uses A.I. if you don’t”. Recently, it seems the algorithm is also increasingly being consulted on collecting strategies and all things art. Surprisingly, a lot of trust is placed in ChatGPT – as I’ll explain below, not always justified. Just because the algorithm is strong in one domain (for example travel tips for a 48-hour trip to Bangkok) does not mean it is equally efficient in guiding your collecting activities in the field of African art! Well, not yet. As a new generation of digital natives expects increased price transparency, while also exploring alternative ways to authenticate an artwork, a lot of trust is given to A.I. – without the need for a seasoned art expert. Ironically, it is the complicated market of African art (where even the experts themselves do not always understand everything) which is precisely why emerging collectors flock to ChatGPT and other tools. In 2023 venture capitalist Marc Andreessen suggested in his “The Techno-Optimist Manifesto” that “artificial intelligence is best thought of as a universal problem solver”. Indeed, our market has a lot of “problems”, so I understand the appeal of ChatGPT of course. But our market is not so universal, even very specialized, and that is still the algorithm’s Achilles’ heel.
Large Language Models such as ChatGPT are only as good as the data they were built with. That data sets the boundaries of what the model can do. It can also start making up things – which it does – just to please its user. ChatGPT would rather give a made-up answer than no answer. I have never heard it say it didn’t know something. While in the gallery, I am not afraid to admit that I just don’t know something. Let’s take the Metoko post I have for sale as an example. It is a much-admired object in the gallery and visitors just love its abstract composition and colors. When someone ask me what its use is, I can only mumble “initiation” – as the few references to them in the literature suggest they functioned in that context. However, if you ask ChatGPT, you’ll get a full essay about their use, an explanation which sounds entirely logical and valid! Asking the algorithm where it found that information, it lists three primary ethnographic sources, but only in a follow-up question does it admit it has no access to these obscure field-reports! Due to the lack of direct documentation on Metoko posts, it just followed the logic behind artworks from neighboring cultures and started to make things up, or hallucinate as techies call it.
Ok, I hear you, a Metoko post is such a rare type of object that I’m really testing the limits of the system. So, a fun second example. A couple of years ago, I co-authored a book on Baule monkey figures from Ivory Coast with Jean-Louis Danis. We dedicated a chapter to the fact that these have repeatedly been misidentified as “gbèkrè” (which is in fact the Baule word for “mouse”). Yet, this error is still widespread on the art market, and hence also online, as not everyone has read the book (yet). So, when you ask ChatGPT what a Baule gbèkrè is it will tell you it is a monkey-shaped statue! Poor thing didn’t have access to my book. Just to say that even when the algorithm has relevant data available, the quality of the data is not verified, and it happily repeats the mistakes of others. Let us also not forget the World Wide Web is flooded with inauthentic African art, which is not labeled as such and of course the algorithm cannot tell the difference.
Just for the record, I’m not a Luddite, you might even call me a techno-optimist. Surely the algorithm is an ideal sparring partner to bounce ideas off. Yet, we shouldn’t let it kill our creativity. If you are accustomed to its language, I’m sure that you too need to smile when you receive a newsletter from a gallery that was entirely written by the algorithm. With a bit of experience, these are easy to spot. I understand one wants to automate tasks, but not a personal message to your clients, right? In a few years, those artificially created newsletters will get artificially created automated replies and it is just machines talking to machines while we ignore our mailbox chilling on a beach?
It’s always good to look back, to get an idea of what’s ahead. In a different field (record collecting), I had to think of when in the early days of eBay I bought obscure disco records on eBay Italy. I had no idea how they sounded, but had strict criteria: they had to be from 1983 and from certain record labels (and producers). I had to wait until they arrived home to know whether it was a hit or a miss. Now, that element of surprise is gone. At the time, I could never have imagined all Italo disco songs would once be available on YouTube as they are now – killing the pleasure of discovering obscure forgotten records on your own. Will the same happen with African art expertise? Last week, I bought a terracotta chameleon in a French auction listed as Akan – but I recognized it immediately as Ife, many centuries older than the auction house thought and published in a book on the subject. That joy probably will be gone in a couple of years, when properly trained A.I. will avoid experts making such errors. The algorithm will have the correct attribution and valuation ready for each lot appearing at auction, ending the age of the auction sleeper.
I don’t think the profession of classical African art expert is in danger just yet. Evaluating the quality of an object can partly be quantified, yet, antique African artworks are far too heterogeneous, even within a certain type, to establish a quality axis that is universally accepted. Tech-savvy IT wizards surely are going to have a lot of fun designing new models, yet the algorithm can’t feel the presence of a sculpture. Or smell it – as a colleague always says: “if you smell smoked bacon, it is not kosher”. Concluding, it is clear the available data set used by ChatGPT online is still too limited (for now). I’m sure in a couple of years, things will have accelerated even more. Combining all articles published in African Arts & Tribal Art Magazine, together with all sales records from all auction houses, and the information held by archives like the Ross Archive of African Art & the AHDRC, and all collection records from museums will create a very clever algorithm. Trained on such high-quality data, and rigorously tested with the help of traditional experts, the resulting tools will become very valuable ‘co-pilots’.
But will they also have taste? Collecting art is so subjective, objects speak to us and stop us in our tracks. We obsess and dream about artworks for very personal reasons – a human thing. And I think in the current and coming machine age that ‘human’ element will become more important than ever. Only by contrasting the man-made with the digitally created, the human touch will be appreciated more than ever. African art will remind us of our shared human experience, and hence its market, in my humble opinion, has a very bright future.
A.I. can be useful, surely, but please resist the temptation to let an algorithm validate quality and authenticity, or set price expectations. African art still requires the eye, the nose, the intuition of someone who has handled thousands of objects, someone like me.
Yours truly,
Bruno