iDiscovery: Agentic AI

Agentic AI systems are AI systems that can pursue complex goals with limited direct supervision on behalf of a user by interacting with a variety of external digital tools and services, such as navigating the web, mimicking real users, or negotiate with other Agentic AI systems. Contrasting the opportunities presented by Agentic AI systems are numerous challenges, including resource and computational requirements, legal issues, data poisoning, and hallucinations. This course will focus on the potential opportunities, challenges, and guardrails of Agentic AI by addressing key questions associated with understanding advanced Agentic AI systems from societal, technological, and security perspectives. Academic and industrial approaches are explored through presentations and discussions lead by experts from across the industry and academia. For this course, the topics of interest include major results and latest developments in authorization and security of AI Agents, the alignment between intended outcomes and actions, trustworthiness and stability of AI Agents.  Industry leaders will provide perspectives on the emerging trends around hyper automation, AI decision intelligence, personalization, democratization and ethics.  Finally, students will gain hands-on experience with Agentic AI systems, Large Language Models and containerized approaches on desktops and the MIT SuperCloud supercomputing system. This course is by invitation only. Invitations are issued based on collaborations and not on an individual basis. If you are currently enrolled, you can view the course content by visiting your Dashboard. To access your Dashboard, click on your username on the toolbar at the top of your screen and select it from the dropdown menu.