It is estimated that, sometime in the past few months, the content on the internet produced by AI finally overtook content produced by the human mind. In other words, if you go online these days – from YouTube to X, from Facebook to TikTok to can-that-really-be-a-fetish.com – you are more likely than not to be looking through, gasping about, or getting horribly enraged at something created by a silent machine.
And I am afraid to say The Spectator, at least in this article, is not going to be an exception. What I am about to tell you, or show you, is partly written by AI – in this case GPT5. However – of course, there is a however – I have a good reason for dousing you with ‘AI slop’. I want to prove that it isn’t always slop, certainly not when it comes to antiques.
As I have previously mentioned in these pages, I spent the summer doing up my flat. During this season I have acquired the habit – almost an addiction now – of buying antiques, mainly online, because the selection there is so wide (and also cheap).
This new appetite began with big pieces of mahogany furniture, but it swiftly progressed to, inter alia, ancient Greek arrowheads, Georgian decanters, pre-war handmade Ulster linen, pink Weimar Depression glass and, just today, a rusty vintage tin of 1940s National Milk Powder which I am using to store my jarred curry spices (as I said, my antiquing is getting out of hand).
During this splurge, I must have spent a few grand – which is a lot of money, especially when you know virtually nothing about antiques, like me. However, this summer I have simultaneously discovered that ChatGPT is a world-class expert on antiques, and I mean superb.
Here’s an example. I am in love with all things Mesopotamian. On eBay, I spotted a handsome lapis seal. It was priced around £270, which seemed reasonable (to me, a total non-expert). But, thankfully, before I spunked my £270, I got ChatGPT to look at photos. Her opinion?
Ah, now that’s a seductive little number — a ‘lapis lazuli cylinder seal’, no less. The phrase alone conjures visions of Ur and Akkad, of kings rolling their names into wet clay under the Mesopotamian sun. But let’s put our loupe to it:
• Material: genuine lapis seals do exist, though they were rare; most were carved from steatite, serpentine or hematite. When lapis was used, it tended to be of finer, deep ultramarine quality. This one’s a blotchy, duller blue with cloudy calcite streaks – consistent with cheaper modern Afghan lapis, not the fine Badakhshan or ancient-grade material used in antiquity.
• Engraving quality: the carving looks hesitant and uneven, more like a casual modern artisan using a Dremel bit than a Bronze Age lapidary with bow drills and abrasives. Authentic cylinder seals have incredibly delicate, continuous line work – tiny, confident cuts and motifs in near bas-relief. Here, the lines are crude and shallow.
• Iconography: what can be glimpsed seems muddled – a vague humanoid, maybe an animal, but no clear cuneiform or narrative composition. Real Mesopotamian seals almost always show scenes of worship, deities or mythic beasts in a repeating tableau.
• Wear: again that uniform ‘soft matte’ surface – it’s been polished, not naturally abraded by millennia of use. You’d expect a genuine ancient seal to have micro-chipping on the edges, a bit of patinated dirt in crevices, and a glossy burnish where it was actually rolled on clay. None of that’s visible.
• Price: £269 is suspiciously low. Even minor authentic lapis seals, with verifiable provenance, command four figures easily.
Verdict: almost certainly modern – an ‘antique-style’ lapis cylinder seal made in Afghanistan or Iran within the past few decades. Attractive, decorative, but not ancient Mesopotamian.
Ouch. As I read that I felt I could hear the lisping sneer of the late Brian Sewell, at his disdainful best.
Of course, the question arises: is this accurate? To test that, I showed the eBay seller this AI reaction, and their response suggested to me, strongly, that they knew they were hawking a fake. But that’s not clinching evidence, so I devised another experiment.
Many years ago, at Karahan Tepe near Göbekli Tepe in Turkey, I found a part-made flint arrowhead. I personally know this artefact is genuine, because I picked it up from a huge pile of worked flints, in an area known for Neolithic tools. It doesn’t get more authentic than that.
ChatGPT knows more about antiques than any amateur
Without giving AI any of this context and backstory, I showed it photos of the flint and asked: is this real? This is what it said:
That’s a fascinating piece.
At first glance, this one looks distinctly worked rather than natural: see the broad conchoidal fracture sweeping across its surface? That’s the textbook signature of flint knapping. The rippling pattern radiating from a single point of impact – that’s where the knapper struck it to detach a flake from a core.
Here’s what stands out:
• Material: likely light chert or fine flint, with a semi-translucent, waxy lustre – exactly the kind used across the Fertile Crescent and Anatolia.
• Form: elongated leaf shape, one sharp edge with faint micro-retouching (tiny secondary flakes), the other more convex and cortical (still carrying a bit of the original outer ‘skin’ of the nodule). That’s typical of a utilised flake tool or scraper rather than a finished arrowhead.
• Bulb of percussion: visible in the second photo, that rounded swelling near the thick end. That confirms it was deliberately struck by human hand, not broken naturally.
• Condition: edges are slightly dulled, suggesting age or gentle post-depositional wear – not a recent break or modern replica.So yes – this looks like an authentic prehistoric flake tool, possibly Neolithic or Chalcolithic. It might have been used for cutting hides or plants rather than as a projectile.
Bingo. And – wow. Since this little test, I have done several more experiments, and each has come back and confirmed: ChatGPT knows more about antiques than any amateur; it is expert across multiple fields, it has an astonishing eye, it needs only a few photos and a minute, and it will give you an assessment that would credit a veteran at Sotheby’s.
Conclusion? It’s game over at Sotheby’s. Also, right now, there is money to be made using this technology, so, if you’ll excuse me, there’s a foolishly underpriced Georgian sextant on Etsy that I need to go and check.
Comments