‘Computer scientists have used the latest image processing techniques to analyse hundreds of works of art and unearth previously unconsidered sources of inspiration between artists’, reported Matthew Sparkes in yesterday’s Telegraph. The article was reporting on a paper put together by a bunch of computer scientists from the Computer Science department at Rutgers, State Uni. of New Jersey, and is available online here.
The gist of the article is this: the identification of similarities between works of art has long been the prerogative of art historians, but now computers, which are becoming ever more sophisticated, are ready to take their place, being capable of identifying instances of formal similarities between given works of art that have hitherto elided the experts.
‘One important task for art historians is to find influences and connections between artists’ say, quite rightly, Babak Saleh, Kanako Abe, Ravneet Singh Arora, and Ahmed Elgammal — the paper’s authors. We’re off to a good start.
But things quickly go awry . . .
‘It must be mentioned that determining influence is always a subjective decision. We will not know if an artist was ever truly inspired by a work unless he or she has said so.’ This is mad as far as I’m concerned. Michelangelo never declares that his conception of God from the Sistine Ceiling was inspired by Ghiberti’s similar airborne God from the Gates of Paradise, but on the basis of formal and circumstantial evidence, which is to say it looks a damn lot like it and Michelangelo will have seen Ghiberti’s sculptures daily in his youth, I think we can say that it’s probably the case. But this isn’t my real issue with the paper; the article does after all acknowledge that instances of artistic influence proposed by art historians are usually demonstrably right, even if there is no “proof”. For example, we might not know FOR CERTAIN that Francis Bacon ever saw Diego Velázquez’s portrait of Pope Innocent X (do we? Raphael’s Julius II makes for just as neat a comparison?), but since the former’s picture of a seated, grand, albeit tormented, bloke really does look like the latter’s Papal portrait, then there most likely IS a relationship. Hence the comparison has found its way into the mainstream literature on Bacon.
My problem instead is with some of the previously undiscovered, but for my money far-fetched, relationships between works of art that the computers have apparently managed to unearth, as well as some of the frankly flippant, if not wholly misguided, claims the authors make along the way. (An important caveat here: the authors do admit that ‘We are not asserting truths but instead suggesting a possible path towards a difficult task of measuring influence.’) Let’s look at some of them.
‘Although the meaning of a painting is unique to each artist and is completely subjective, it can somewhat be measured by the symbols and objects in the painting.’ Art historians will nowadays wince at those words, and Roland Barthes will probably have had chickens . . .
‘The earliest style is the Renaissance period with artists like Titian and Michelangelo during the 14th to 17th century.’ Notwithstanding the arbitrariness of the period style classifications that the article leans on more generally (Renaissance, Romanticism, Baroque, Pop, Abstract Contemporary, American Modernism, Post-Impressionism…etc.), this statement is a bit worrying… Pedantry, perhaps, but Michelangelo and Titian weren’t about in the 14th century or the 17th, and if we’re being picky, traditional narratives of art history don’t usually include 17th-century art under the rubric of the Renaissance.
And the most major problem, I think, is this clanger:
‘Paintings do not necessarily have to look alike, but if they do, or have reoccurring objects (high-level semantics), then they might be considered similar.’
My issue here is that they’re effectively saying that even if pictures don’t look alike to the eye, computers, with all their mathematical wizardry and algorithms, can nevertheless spot relationships that otherwise defy human perception. Problem here, of course, is that people make artworks, not computers, and so if two artworks by two artists don’t look alike to the eye, then it is really doubtful whether there ever was a meaningful relationship between them. Common sense, which computers don’t possess, dictates as much.
See: Frédéric Bazille’s Studio 9 Rue de la Condamine (1870) and Norman Rockwell’s Shuffleton’s Barber Shop (1950). The computers threw this up as a match, and ‘After browsing through many publications and websites, we concluded, to the best of our knowledge, that this comparison has not been made by an art historian before.’ The authors’ faith in the technology is thus vindicated. But, hang on, there’s probably a very good reason why art historians have never spotted a relationship between Bazille’s Studio and Rockwell’s Barber Shop, which is that they simply don’t look sufficiently alike to warrant the positing of anything more than a coincidental relationship between them. Which is precisely what the next sentence says: ”The painting might not look similar at the first glance, however, a closer look reveals striking similarity in composition and subject matter, that is detected by our automated methodology . . . [emph. mine]’ I don’t buy it. And what the authors neglect to mention is what we, that is to say art historians, call iconographic conventions. Bazille’s picture belongs to a rich tradition for showing artists working in their studios, and perhaps Rockwell did, either knowingly or inadvertently, look to that that tradition for his Barber Shop. That’s a sound art historical judgement. But it doesn’t mean that Rockwell was influenced by Bazille. The authors also fail to mention whether there is any inkling whatsoever that Rockwell knew Bazille’s work(s)? These are the kinds of questions art historians ask, whereas computers, it seems, do not. Or perhaps can’t ask?
Similarly, the paper heralds the similarity between Georges Braque’s Man with a Violin and Pablo Picasso’s Spanish Still Life as a “discovery”. Nah, not really. Braque and Picasso were immediate contemporaries, they knew one another(!!) and were pioneers of a movement retroactively called Cubism. A perusal of any monograph on cubism will generate Braque’s and Picasso’s names alongside one another and show ample similarities between their art. So, not a discovery after all. . .
If you read the paper, you’ll quickly find that there’s loads of technical jargon, equations and mind-boggling graphs which apparently bestow scientific robustness on the findings:
Erm…. I’m struggling. The tables and charts don’t exactly shed any more light, either:
And what on earth is this?
. . . answers on a postcard, please.
I can’t help but think that all this is a case of all fur coat and no knickers, and that the jargon simply conceals the fact that the computer’s supposed discoveries don’t stand up to the scrutiny of art historians. It’s nonsence masquerading as scientific art history. I really don’t think, for example, that Bazille’s and Rockwell’s pictures look sufficiently alike to warrant the claim that there IS a relationship. And this is where art historians differ from computers: art historians, or else, the good ones, weigh-up their proposals against a balance of probabilities, and posit relationships between artists and their works only where there is a demonstrable formal relationship that is meaningful, and, moreover, can be substantiated by consideration of the likelihood that X artist had seen Y’s Picture. That’s what art historians do, and, on the basis of this article, are capable of doing infinitely better than a machine.
Suffice it to say that I don’t think we can expect to see queues of bereft-looking art historians at the Job Centre any time soon . . .