Overview
Multi-agent LLM systems are fully instrumentable: every message, tool call, and memory write can be logged. Transfer entropy and directed information are the natural language for directed flow, with recent estimators (TREET; AGM-TE) and early applications to LLM-MAS cascade monitoring. Separately, causal audits of latent channels show that end-task performance does not identify whether receivers actually use transmitted content (message permute / drop / other-example interventions). An information topography needs topographic quantities — conductance proxies, saturation, judgement-junction load — that survive causal checks. This project builds that instrumentation layer on DOAgent-quality traces and refuses to treat a TE heatmap as a result.
Prerequisite
L172 Information, Energy and Intelligence (IEI), or equivalent preparation in information theory, maximum entropy, and information geometry.
FAQs
- What are the prerequisites?
L172 Information, Energy and Intelligence (IEI), or equivalent preparation in information theory, maximum entropy, and information geometry.
- What will I learn in this Project?
You will learn directed information and transfer entropy (TE), modern neural / generative TE estimators, causal message-intervention audits for multi-agent LLM systems, and data-oriented instrumentation with DOAgent. You will build validated metrics for channel utilisation, saturation, and judgement-junction load — not heatmaps that fail under intervention.
- What is the objective of the project?
(1) Instrument a multi-agent system (DOAgent preferred; AutoGen/MetaGPT acceptable) with full logs across communication, memory, tool, and execution channels. (2) Implement at least two TE/DI estimators; calibrate on synthetic graphs with known flow. (3) Run causal message interventions (permute, drop, replace with other-example messages); require claimed flows to survive these checks. (4) Define and validate saturation and judgement-junction load metrics; package a small reusable library with documentation and evaluation. Deliverables: library, calibration + intervention study, thesis chapter.
- How does this fit into the bigger picture?
This is a core target for instrumenting an information topography. We need deployable tooling so that topographic interventions can be evaluated in engineered systems and later in partner organisations. The project extends the S4 / DOAgent agenda from interpretability of decisions to quantitative information topography, and complements existing DOAgent projects on ICU decision support and multi-agent interpretability.
Related Group Projects
Interfaces
The Interfaces research programme develops interpretable, self-sustaining multi-agent AI systems, treating software as the interface between socio-technical needs and AI capabilities, with validation in healthcare and other high-stakes settings.
Data-Oriented Architectures for AI-based Systems
Data Oriented Architecture (DOA) is a software architecture pattern that creates data-driven, loosely coupled, decentralised, and open systems. DOA achieves these goals by exposing systems' data as first class citizen to distributed, stateless, and asynchronous systems’ components. These design decisions enable DOA-based systems to achieve desirable properties such as data availability, reusability, and monitoring, as well as systems adaptability, scalability, and autonomy.