The masses against the classicists?

In retrospect, it seems foreordained that the great Classics Twitterstorm (sorry – Xstorm) of 2025 happened at its close, with a guerrilla scholar going by “Roman Helmet Guy” (henceforth “RHG”) taking on — progressively — Marxist historiography, secondary sources in general, and, finally, the very concept of philology itself. 

Though that storm has subsided back into its teacup, the affair did raise two interconnected questions of wider interest: whether scholarly gatekeeping is valid and necessary, and whether AI has a leading role to play in bypassing it (or supplementing it). Let us consider how these two issues came up in the debate, and then try to find a broader view on them.

The spat began because RHG bragged of convincing Elon Musk of the barbarisation thesis — that the Roman Empire fell in the West because of a large influx of unintegrated barbarian migrants. (The contemporary political valence, by the by, is clear.) His recapitulation of this idea raised some ire, most significantly from those who noted the barbarisation thesis is no longer credited in scholarship, set aside in favour of models like those of Chris Whickham which look at plague and climate change as decisive.

RHG was not cowed, and here is where the initial question of scholarly gatekeeping came up. Why, he wondered, should he credit modern historiography by avowed Marxists, when he could simply read the primary source of why the Romans thought they were struggling?

This isn’t wholly unfair. We are well past the point where conservatives and reactionaries can be gulled about the neutral expertise of progressive academia. Whickham is, in fact, an avowed Marxist, and you worry that with that hammer in hand, he’ll find the nail of capital and other structural causes under every historical rock. RHG may be incurious and even uncharitable, but he is not merely shadow-jousting here.

Yet are these academic gatekeepers wholly to be set aside? RHG’s next cannonade might make us run back to the climate-change-obsessed kooks. Defending his prioritisation of primary source, he gave the specific example of how some modern scholars posit Emperor Septimius Severus as a black man – but the Historia Augusta says that Severus was so scared of black people that he saw one and thought it was a bad omen! 

But is the Historia Augusta a primary source? It is certainly a Roman source, but it was (probably) written 150-190 years after Severus’ reign. It is like a book written now about the American Civil War — and as much as I enjoy Shelby Foot, Steven Woodworth, and the rest, they are hardly primary sources, despite also being American. 

The dating of this “primary source” may have passed RHG by, however, because it purports to be written about 100 years before it probably was. Yet the vocabulary is anachronistic for that period, the naming conventions wrong, several of the events are of doubtful historicity, and, what’s perhaps worse, the text as we have it seems to interact with Ammianus Marcellinus, who lived c. 330-400. This sort of ferreting out of evidence, the parsing of references, the comparing of sources, is hard for the admirable autodidact (such as RHG), who does not have the time, the breadth, or depth — but it is exactly the hunting ground of the obsessive loser scholar, be they Marxist or not.

By this point in the debate, RHG understood he had gained what success he could over Marxist scholars and Septimius Severus, and shifted the battlefield: if you also prefer primary sources to Marxist historians, why not head to his Substack, where he is translating heretofore-untouched Venetian dispatches from the later Renaissance period?

Just as Socrates would be interrogated first by the proponent of one idea and then by the believer in another, RHG now saw his most vocal critics change, with the decidedly conservative Magister Ioannes and Yuan Yi Zhu joining the fray. Their objection? RHG was, in fact, simply using AI to translate the Venetian Dispatches — a fact he happily owned — and then tidying their style up in post.

I checked in with ChatGPT on this, asking “can AI translate Renaissance Venetian-language documents?”. Its summary answer:

Short answer: sometimes—but not reliably on its own.

Translating Renaissance Venetian-language documents is one of the harder tasks for AI, and results vary a lot depending on the text.

Why It’s Difficult

Venetian ≠ Standard Italian

Renaissance Venetian (15th–17th c.) is a historical variety with:

• • Different spelling conventions (no standardized orthography)

• • Vocabulary that has fallen out of use

• • Regional and administrative jargon

• • Latin influence (especially in legal and ecclesiastical texts)

AI systems are mostly trained on modern Italian, not historical Venetian dialects.

It went on to suggest that, under perfect conditions, AI could translate “personal letters” with a 60-75 per cent accuracy, and “administrative text” with 50-65 per cent accuracy. 

From the horse’s mouth, then. 25 per cent inaccuracy is vastly higher than any eccentric Vicar’s amateur translation from the 19th century. RHG’s defence was that, in the end, there were nowhere near enough actual philologists to ever get to any of this; Magister Ioannes agreed but seemed (in a snottily academic moment) to think that this was all quite correct and proper.

In essence, however, it is fair to say that RHG’s Venetian Dispatch translations are of the nature of a semi-informed guess. No serious historical work can be undertaken based on the translations themselves, and the likelihood of major errors affecting any given letter are high, rendering them dubious even for the total neophyte. There is also no immediate hope of vastly improved accuracy, because without more high-quality human work on Renaissance Venetian sources, the AI translators have nothing with which to improve their processes.

Our confidence in the Tweeting Autodidact, especially the one relying on AI, must at this point be rather shaken. The drive to self-improve is always admirable, and the inventive turn to modern technology appealing, but the results are dubious. Yet is there something that can be salvaged from the tale of Roman Helmet Guy who, like Varus, has gotten himself lost in a wood?

It’s one in the eye to Luddites like me, but yes. Perhaps the most exciting advance in Classics in recent years has been the “unlocking” of the flash-fried Herculaneum Scrolls, the only surviving library of its type from the ancient world, but heretofore completely unreadable. Opening the scrolls simply destroyed them. Modern technology offered a route forward to translating the 800 (!) scrolls — they could be “digitally” unrolled and scanned by 3D X-Ray. The problem remained that this only returned gibberish, because the many layers of the scroll were interposed over each other.

The Vesuvius Challenge, announced in 2023, offered a grand prize of$700,000 to whoever could provide a way of decoding them. By 2024, this prize had been claimed — by a team using AI. Three non-classicists, but experts in AI and robotics, put together a model which, in its initial run, decoded 2,000 Greek characters from one of the scrolls. These seem to belong to an Epicurean philosophical text, perhaps by Philodemus. If this success can be replicated en masse, AI will have provided the biggest literary breakthrough in Classics since the Renaissance.

Of course, philologists were involved in the final process, but the embrace of technology and the welcoming of outsiders has been vital in this great breakthrough. There is force in RHG’s complaint that there are so many untranslated older texts that philologists will never get to them, and as new tools emerge, brushing such complaints off is simply not going to work anymore. There have always been brilliant amateurs in Classics, and a blanket rule barring them based on credentialism can only do harm. The rediscoverers of Pompeii were, first, an architect (Domenico Fontana) and then a military engineer (Roque Joaquín de Alcubierre). The decoder of Linear B, Michael Ventris, was a bomber navigator-turned-architect, who decided to take up linguistic cryptology in his spare time.

Nevertheless, both RHG-Porta and the triumph of the Herculaneum Scrolls demonstrate the indispensability of expert human input, whether credentialed or not. The AI models cannot develop knowledge, only refigure it (though they can use this skill to reach brilliant conclusions — see their success in various Maths Prizes). At this stage, at least, human judgement will still be needed to judge the literary nature of the decoded scrolls at Herculaneum. Nevertheless, just as Will Robinson was better with Robot than without it, so too is the classicist better embracing technology.