AI Epistemic Risks: Emerging Mechanisms & Evidence
Stephen Casper, Jonathan Stray, Anna Gausen, Cameron Jones, Brian Christian, Bálint Gyevnár, Siao Si Looi, Joshua Levy, Elizabeth Seger, Hannah Kirk, Kobi Hackenburg, Matt Kowal, Michelle Malonza, Dan Zhao, Zhonghao He, Luke Hewitt, Hause Lin, Maarten Sap, Dylan Hadfield-Menell, Atoosa Kasirzadeh, Gordon Pennycook, Jean-François Godbout, Kellin Pelrine, Reihaneh Rabbany, Natasha Jaques, Jasmine Li, Yoshua Bengio, David Rand, Thomas Costello, Mick Yang
SSRN
Abstract
Advances in artificial intelligence pose risks to humanity’s collective capacity to form accurate beliefs, reason well, and maintain a healthy information environment: risks we term epistemic risks. These risks arise from AI’s integration into the infrastructure through which individuals and societies think, form beliefs, and make sense of the world together. We highlight three primary mechanisms by which AI can cause systemic epistemic decline: persuasion and manipulation, cognitive offloading, and feedback loops. AI can persuade and manipulate through misuse for economic or political manipulation, inciting crime or radicalization, and escalating conflict, and can also cause unintentional harms such as sycophancy and mental health risks. We record the emerging empirical evidence for each mechanism and analyze how they likely amplify one another, creating compounding systemic risk. The evidence suggests that AI could be an unprecedented lever for improving epistemics, but this will not happen by default.