This is, as Gary Marcus puts it, an “epistemic clusterfuck” in the making. Imagine it becomes a common experience to access the news through the meditation of a chatbot:
Whereas Marcus focuses on the dangers of a lack of ‘ground truth’, in the sense that Grok would be relying on Twitter discourse to fuel the product, I immediately thought about personalisation. If the metric for success here is engagement (and I think it would be given that we’ve moved towards a hybrid membership and surveillance model for social platforms) then technologies to increase engagement would be inevitable. The obvious way to do this is to profile the responses of the user, with a view to personalising the presentation in order to increase their engagement with the news product.
Even with the most benign motivations in the world (e.g. if this was a public interest foundation rather than Elon Musk) this would an incredibly dangerous road to go down, because the editorialising would be (a) black boxed (b) utterly privatised (c) potentially with escalating effectiveness over time. If we assume a world where motivations are not benign, it’s quite terrifying, getting us ever close to Neal Stephenson’s dystopia in the Dodge books where everyone who can’t afford a human curator is driven mad by the internet.
Mark Carrigan 