Artificial Intelligence & Epistemic Security

Course Description

This course will explore how the emergence of generative AI is affecting issues of epistemic security (misinformation, influence operations, media consumption, academic integrity, etc.) and how advances in AI could shape our epistemic futures. Students will learn the basics of how foundation models, LLMs, image generation models, and multimodal models work, and how choices across the AI lifecycle, from development to deployment, can cause harmful outputs and/or could contribute to epistemic decline (e.g.: quality of sources on the Internet, issues with model confabulations, academic integrity). Furthermore, students will apply their knowledge of influence operations and misinformation gained from previous courses to understand how malicious actors could use AI to threaten epistemic security, as well as learn more about the current AI malicious actor ecosystem (for both state and non-state actors). Lastly, the course will delve deep into potential solutions to mitigate epistemic crises, from technical mitigations within AI models, content authentication approaches, to broader whole of society efforts (participatory governance, information literacy, etc.). Students put themselves in the shoes of various actors in the current AI ecosystem, specifically: large AI developers, social media platforms, and policymakers across the US government, to produce targeted outputs for the course. In addition, students will engage with and use generative AI tools to understand various types of harms and will also learn unique insights into the challenges of trust & safety in the AI space. Course outputs could include: an internal policy enforcement protocol for disinformation for a large AI developer, a policy memo for the Director of OSTP on AI and information integrity, an exercise to generate known mis/disinformation narratives using generative AI models, and more.

Code

SA.501.116

Credits

4

Subject

Technology and Innovation