Deeper Understanding

I’ve been continuing to think about ‘AI’ and LLMs and the like; partly through sheer annoyance at some very silly ‘This Is How AI Will Build Utopia’ and ‘This Is How AI Will Destroy Us All’ articles (it’s a two-man con, isn’t it? They set this up as THE debate we should be having, with both sides calling for more funding for research in this area, drawing attention away from the systems and structures shaping these developments), partly because I was putting together a funding application for a small project on how to respond to Chat-GPT and its ilk within the assessment of historical skills, and partly because the philosopher John Holbo has been posting some very interesting, thought-provoking pieces on Crooked Timber, prompting some high-quality discussion.

One of the significant themes emphasised in the latter is the role of the human response; we are the ones ascribing intentionality, consciousness, intelligence and emotions to the unknown thing that is generating stuff that reminds us of our own outputs – even when disparaging them, by talking about ‘mimicry’ or ‘hallucinations’. There must be a shedload of relevant research in the anthropological study of ‘animism’ – a human propensity to identify patterns of regularity, predictability and even consciousness within the mass of data we receive about the world around us, and for the most part to anthropomorphise it all. If horses could draw their gods, they would look like horses, as Xenophanes remarked; humans conceive of the inscrutable processes of generative AI either as quasi-human or as a very human, crude science fiction image of what aliens or supercomputers would be like, i.e. part-exaggerated and part-deficient humans (intelligence without emotion, aggression without pity etc.). We all need to (re-)read Stanislaw Lem’s Solaris…

This also brought to mind a post I wrote about six years ago, that I think has aged pretty well (without having had the slightest impact): we need some serious study of Greco-Roman cultures of enslavement to broaden our sense of how societies respond to such a situation. Extensive debate and confusion – philosophical, legal, cultural – about whether our ‘thinking tools’ are human, or semi- or deficient humans, or merely ‘two-footed things’ analogous to ‘four-footed things’ like cattle and dogs. Widespread fear of revolt or violence – combined with a willingness to hand over sensitive tasks, personal information and financial responsibility, upbringing of children etc. to these alien, hostile, resentful subjects.

The insistence that even when one of them has attained full legal status as human, and amassed wealth and social standing, there is always something that marks them out as inferior (masking the anxiety that perhaps there isn’t); this might be more of a problem for the future, but is it stretching the analogy too far to see something of this mindset in reactions to Chat-GPT outputs, the idea that there are easy ways of telling them apart from ‘real’ writing (even if, as has been noted in discussions of using ‘gAI’ to detect the products of other LLMs, the principles also serve to exclude non-native speakers, the less well-educated etc.)?

Obviously Greco-Roman societies were actually founded on the enslavement of full human beings, whereas we are not (and ‘enslavement’ is clearly not the right term). But for the purposes of understanding the reactions and thinking of the established elite (today: western humans generally), this doesn’t matter; the presence of ‘things’ that appear to possess some human-ish qualities and that are capable of taking on many complex tasks makes this a decent enough analogy, not least because neither we nor the ancients are sure what we’re dealing with, what we’re living alongside, what we’re ceding power and responsibility to, and that’s precisely the issue.

So, if I wasn’t trying to write a book about something completely different, I would be looking for some suitable funding calls for projects on the social and cultural dimensions of ‘AI’, because I’m pretty sure they don’t yet realise how much they need input from Greco-Roman history. If anyone wants to run with this idea, be my guest – I might have the capacity to be a CI, but I’d rather see this happen without me than not happen at all.

Incidentally, since I drew the title of this piece from it, the 2011 original of Kate Bush’s song is SO much better than the 2018 remake, perhaps just because voice software has got a lot better in the last five years so the 2011 version seems more prescient… https://m.youtube.com/watch?v=JVU6eFBvQ0g