AI for Decision-Making under Behavioural Feedback

Many real-world AI systems influence the behaviour of the people they serve. In EV charging networks, for example, pricing decisions influence travel patterns, which in turn affect congestion and future pricing decisions. This feedback loop poses significant challenges for conventional machine learning and optimization methods, as the underlying data distribution depends on the decisions being made. Developing AI methods that explicitly account for this interaction is therefore essential for trustworthy and citizen-centric decision making.

This project develops AI methods for decision-making in large-scale networked systems, drawing on decision-dependent optimization and machine learning. The first phase of the project developed a new framework for optimal network pricing with oblivious users, together with scalable stochastic optimization algorithms with theoretical convergence guarantees. By combining ideas from machine learning, optimization, and algorithmic game theory, the project enables AI systems to make reliable decisions under uncertainty while accounting for human behavioural responses, supporting more efficient, trustworthy, and citizen-centric management of EV charging and other critical infrastructures.