Aligning AI With Human Values
As ChatGPT took the world by storm in 2023, it became clear that AI systems will soon become ubiquitous in our lives. MSI PI Dongyeop Kang (assistant professor, Computer Science and Engineering) and MSI PI Mingyi Hong (associate professor, Electrical and Computer Engineering) are working on a project called “Aligning AI models with Human Values: Formulation, Algorithms and Applications,” that acknowledges the potential benefits in fields like medical diagnoses and virtual assistance, and so emphasizes the necessity for AI systems to align with human strategies, preferences, and evolving needs for effective and reliable interaction. The overarching goal is to build robust AI systems that align with human values and societal norms, addressing critical challenges in integrating human-generated data, adapting to evolving preferences, and ensuring transparency in decision-making by intelligent agents.
The project consists of two thrusts. The first thrust aims to design multifaceted alignment formulations, focusing on dynamically steering multi-dimensional human values and aligning multiple specialized agents collaboratively. The second thrust involves developing efficient alignment algorithms using optimization theory to achieve data- and compute-efficiency.
This project recently received a DSI Small Seed Grant. The Seed Grant program is intended to promote, catalyze, accelerate, and advance U of M-based data science research so that U of M faculty and staff are well prepared to compete for longer term external funding opportunities.
The program was updated in Summer 2024 to include three focus areas: Foundational Data Sciences; Digital Health and Personalized Health Care Delivery; and Agriculture and the Environment (new focus area). This project falls under the Foundational Data Sciences focus area. The types of awards are Rapid Response Grants and new types, Awards for DSI Faculty Fellowship and Data Sets (Data as an Asset).
Image description: Aligning AI with human values.
