AI epistemic risks has got to be one of the most interesting areas at the moment, and it highlights why epistemic *resilience* is so important. The paper of the year on this (and quite possibly the most important paper ever on this issue) highlights that these risks are "self-concealing", because it is hard to know how our thinking has changed retroactively. Which is why it is so important to measure, track, and actively influence epistemic effects. It also sorts these issues into three categories: First, AI can persuade and manipulate. It can be misused for economic or political manipulation, inciting crime or radicalization, and escalating conflict. It can also create unintentional harms, e.g. sycophancy and mental health risks. Second, AI's unique features enable cognitive offloading in a qualitatively different way from previous technologies, risking cognitive decline. Third, human-AI and AI-AI feedback loops can narrow the epistemic space which humans and AI systems draw from. This already drives homogenization and may potentially lead to fragmentation. The 30+ authors record the emerging empirical evidence for each mechanism and analyse how they likely amplify one another, creating compounding systemic risk. Full paper, 1000% worth a read: papers.ssrn.com/sol3/papers.… Paper by @Yoshua_Bengio @StephenLCasper @jonathanstray @camrobjones @natashajaques @brianchristian @hannahrosekirk @KobiHackenburg @ea_seger @lukebeehewitt @hauselin @MaartenSap @dhadfieldmenell @tomstello_ @DG_Rand @GordPennycook @Dr_Atoosa @KellinPelrine and others