AI for Critical Decision-Making: Uniting Logic and Probability

Advancements in AI could be used to aid in decision-making. Algorithms that make fast, accurate, data-driven, and optimal decisions can make the difference between life and death in critical moments, such as in natural disasters. These decisions require both symbolic reasoning, used for well-defined, rule-based problems and statistical reasoning, which handles uncertainty.

Progress has been made in symbolic decision-making and statistical inference individually, but Yexiang Xue, assistant professor of computer science at Purdue University, recognized the need for a tool that combines them to optimize decision-making. 

This innovative research aims to solve high-stakes real-world challenges, such as disaster preparedness and infrastructure planning, where accurate decision-making can mean the difference between success and failure.

“The integration of symbolic and statistical AI was rooted in the two prevailing schools of thoughts of AI since its creation – one, symbolism, which believes AI can be solved using symbolic reasoning, and the other, connectionism, which leverages probabilistic predictions made by neural nets. This award will explore a new way to unite these two schools of thoughts,” said Xue.

Satisfiability Modulo Counting (SMC)

Xue’s research introduces a novel approach known as Satisfiability Modulo Counting (SMC). SMC merges two powerful methods: satisfiability solvers, which handle symbolic reasoning, and weighted model counting, a technique used in statistical reasoning. 

By combining these two methods, SMC provides tighter guarantees and more reliable results than previous approaches, making it particularly well-suited for applications where both precision and probabilistic reasoning are essential.

For example, in planning for disaster preparedness, SMC can help city officials determine the optimal placement of emergency shelters while ensuring that, with high probability, residents can reach safety during extreme weather events. This blend of logic and probability enables decision-makers to create plans that not only meet deterministic requirements but also account for uncertain, high-impact scenarios.

Xue’s research is expected to have broad applications beyond disaster preparedness, extending to fields such as game theory, operations research, and AI-driven social good projects. His research is poised to significantly advance the integration of symbolic and statistical AI.

“Our hope is that the extensions of classical satisfiability solvers a new modulo theory of weighted model counting will allow us to make data-informed decisions with confidence in a diverse set of domains,” Xue said. 

NSF CAREER Awards

NSF CAREER awards are the organization’s most prestigious awards given to junior faculty who embody the role of teacher-scholars through research, education and the integration of those concepts within the mission of their organizations. CAREER awards support promising and talented researchers in building a foundation for a lifetime of leadership. Receiving this award reflects this project’s merit of the NSF statutory mission and its worthiness of financial support.

Yexiang Xue is an assistant professor of computer science at Purdue University. His research interests include developing intelligent systems that tightly integrate decision-making with machine learning and probabilistic reasoning under uncertainty. Prior to coming to Purdue, he received his Ph.D. degree in the Department of Computer Science at Cornell University.

Article first originally appeared at cs.purdue.edu