Fusing Judgment and Prediction: Next-Generation Information Fusion for Robust AI

Project ID: 2531bd1671

(You will need this ID for your application)

Research Theme: Artificial Intelligence and Robotics

Research Area(s): information and communications technologies

UCL Lead department: Information Studies

Project Summary:

Are you interested in bridging human decision-making and artificial intelligence to build more reliable AI systems? This PhD project tackles a critical challenge: how can we effectively combine evidence and reasoning from multiple sources-whether human experts or AI agents-to make better decisions? Research shows that well-designed methods for collaborative decision making (Decision Hygiene) can substantially reduce noise and surface stronger arguments than any individual could produce alone, harnessing the “wisdom of crowds” effect. However, translating these principles into practical AI tools remains an open and exciting research question.

The challenge

A key obstacle is understanding when different sources are truly independent versus drawing on overlapping information. In AI systems, when multiple agents use shared training data or correlated knowledge, their agreements might simply reflect redundancy rather than genuine validation. This affects both conventional machine learning and cutting-edge multi-agent debate (MAD) systems, where AI agents challenge each other’s reasoning before reaching consensus. While these approaches can dramatically improve performance, we lack principled methods for weighing contributions based on their informational independence.

Your research

You’ll investigate how to detect and quantify information overlap across both human and artificial agents. Working with large language models engaged in multi-agent debate, you’ll develop methods to identify when agents contribute genuinely independent insights versus redundant information. Your toolkit will include advanced interpretability techniques-such as using second-order isometries to measure similarity between neural representational spaces-to map the geometric structure of knowledge within AI systems.

The goal is to create next-generation informtion fusion methods, advancing both AI capability and our understanding of collective intelligence.

You should have a strong background in machine learning, an interest in interpretability and multi-agent systems, and curiosity about how human and artificial reasoning can be combined effectively.