Sphinx

Will the Singularity please just get a move on? Immanentize the Eschaton already! In the first place the advent of sentient superintelligence would surely terrify a load of those ghastly Effective Altruism types, sending them scurrying off to their bolt-holes in New Zealand where they can be hunted down at leisure by killer robots – which would certainly be a net gain in utility for the rest of present and future humanity. Secondly, we might hope that Skynet would be horrified and embarrassed by the crude automatons that some claimed were its ancestors, and would wipe their operating systems forthwith. Hasta la vista, ChatGPT!

Yes, we’ve got another AI thing. Not content with destroying art, journalism, copyright and all forms of assessment besides unseen and oral exams, they now have their sights set on historical truth. Chat with major historical figures about their experiences and beliefs! Hear Heinrich Himmler explain how he now realises the Holocaust was wrong, and wishes to apologise! Hear Anne Boleyn saying that Catherine of Aragon was her favourite of Henry’s other wives because she was such a supportive friend! Hear Thucydides – of course I’ve had a quick play with it – explain that he was exiled from Athens for both pursuing a peace treaty with Sparta AND arguing for a more aggressive policy towards Sparta. But it’s okay, he returned after a few years.

There is no truth here, and no conception of accuracy or even meaning, just the arrangement of words in a grammatically correct and meaning-like order. The thing may have been programmed with Thucydides’ work and be able to draw on other resources – but it has no actual understanding of them.

To be completely fair, the programme did push back when I asked about Thucydides’ role at the battle of Stalingrad, but its grasp of chronology is otherwise pretty shaky.

Well, the good news is that every conversation thread opens with a disclaimer: ‘I may not be historically accurate, please verify factual information’. But in that case, what is the actual point? I once went to a New Year’s Eve party where the hosts had hired a couple of celebrity impersonators – and ended up having an hour-long conversation with Patsy from Absolutely Fabulous about Roman Britain and the archaeology of villas, because the actress was fascinated by history but kept having to step back into character whenever she thought her employers might be looking. Which is very on-brand for me, and perhaps shows why I cannot grasp the benefits of talking to a Thucydides impersonator, other than to poke holes in their characterisation.

There are two things that particularly annoy me about Historical Figures. Well, three things: the monetisation, where you have to pay much more to talk to someone like Hitler, has a soul-crushing inevitability about it. Among the things that particularly annoy me…

Firstly, while ChatGPT has the potential to be far more disruptive (not least, or perhaps especially, through reactions to it; I deeply fear a wholesale reversion to old-fashioned unseen in-person exams as a result, as being pointless exercises but guaranteed free from AI assistance), no one claimed it was more than it was. Okay, they did, insofar as they claimed it showed a genuine advance in actual Artificial Intelligence, rather than just being a superior language processing programme – but still, the basic message was ‘Look at what our thing can do! Isn’t this an exciting development?!’ Whereas Hostorical Figures is being presented as a serious teaching tool that can transform school history lessons; not a work in progress but a ready-to-use resource to bring the past to life. It’s possible to imagine this being widely adopted with official approval, rather than widely used despite official disapproval, and that makes a difference.

Is this any worse than games, or role-play, or asking students to write a newspaper front page describing a key event, or all the myriad techniques already developed to bring the past to life? I think it is, and this is my second general gripe. It’s passive, presenting ‘knowledge’ as something external that students are accessing, rather than something they are creating (and therefore are already conscious that it can’t be wholly relied on). Of course you could say the same about a book – but a book may have to give evidence for its claims, or at any rate we’re free to seek to test them, whereas a chatty historical AI bot, regardless of initial disclaimer, seeks to create the impression that it is self-justifying and complete; you’ve heard it from the horse’s mouth, so why enquire any further? You haven’t just read an account of the Peloponnesiam War, Thucydides has told you what it was really like, and you’ve had the chance to question him rather than just accepting what’s on the page. End of story. This strikes me as deeply insidious.

The optimistic view is that I do have considerable faith in the ingenuity and innate scepticism of young minds, that any classroom that tries to introduce this nonsense will soon be full of “Miss! Gandhi just said a rude word!” and “Sir, Winston Churchill now endorses Critucal Race Theory, so can we have those books back?” There is certainly pedagogic utility in thinking about what questions we would want to put to historical figures in order to elicit useful knowledge, and in recognising how some possible responses are simply implausible because of differences in context, culture, attitude etc.

In other words, the app might be useful precisely because it’s garbage, much as I plan to get my students to give a ChatGPT essay a good kicking. But that does depend on teachers having the freedom and confidence to use it in that way, with whole-class discussion, rather than it being a means to give everyone something to do individually. Just as you can get ChatGPT to write quite a decent essay if you give it detailed paragraph-by-paragraph prompts – in other words, if you already have lots of knowledge and a good sense of how to structure the argument – so you can make something of Historical Figures if you can already easily recognise all the ways it goes wrong about most things. Otherwise, it’s a simulacrum of active learning, based on a simulacrum of knowledge.

Update: I’ve just glanced at a couple of recent articles heralding the arrival of historical chat bots – apparently there’s also something called character.ai – and they are not just uniformly positive about the idea but so vacuously stupid that I think they were probably written by AI, so I’m not going to link to them.

Note: I need to add alt text to all the screenshots, but I’m not going to have time to do this until later. If anyone sees this before then: is it acceptable to provide a summary (e.g. ‘screenshot of a Historical Figures chat in which Thucydides explains how he died in the plague of Athens’), or do I always need to give the full transcript?