The Differential ‘Wisdom of Crowds’: Using ChatGPT and the Twitter API to Interrogate Entire Cultures

ChatGPT hit the public consciousness in a large way around the turn of 2022–2023. Generative models of language have been around for decades; Hidden Markov Models, like ChatGPT, can generate natural language from a probabilistic model derived from large amounts of data. But the public rightly sat up and took notice of ChatGPT because of its truly unprecedented ability instantaneously to generate essay-length text, the logical cogency and linguistic grammaticality of which is consistently as good as, or better than, what many humans are able to produce. Time has not been lost by students and even lawyers looking to ChatGPT to help them find shortcuts in their writing assignments… and there has been consternation and pushback too. Is this kind of use ethical? Or do we need to re-evaluate what it means for humans to do creative writing? Another aspect of ChatGPT that has not gone unnoticed is the possibility of its producing factually inaccurate output. ChatGPT is an example of an application of Large Language Models (LLMs); since LLMs are trained on large corpora of human-generated text, the veracity of their output will be a function of the veracity of all the input into the LLM. A question thus arises as to whether the existence of any cultural groupthink in the ‘Large Language’ input might cause ChatGPT simply to reflect back to its user (and further entrench) any blind spots, or even mass delusion, that might already exist. We suggest that there is a way, using language and linguistics (and leveraging the diversity that necessarily exists between cultures and languages) to test this hypothesis, and in fact lean into the problematic tendency of ChatGPT and similar technologies to present dubious material as truth, in order to gain a macro-level view of culture-specific blind spots and biases – both our own and those of others. Archimedes once said, ‘Give me a place to stand, and a lever long enough, and I will move the world’ (δῶς μοι πᾶ στῶ καὶ τὰν γᾶν κινάσω). If ChatGPT is the lever here, what it allows, if used correctly, is a ‘place to stand’ – a culture and linguistic milieu independent of the assumptions of our own culture – to begin to highlight subtle and not-so-subtle differences between cultural shibboleths which might be based on disinformation. As light begins to be shed more on these, we can, in the words of the engraving in the CIA headquarters, ‘know the truth, and the truth will set us free’.