About the Project
AI_CDT_DecisionMaking
Details
Large language models are increasingly used in multi-agent systems where specialised agents communicate through natural language, shared memory and multimodal representations, access external tools and make sequential decisions. However, uncertainty and errors can propagate across agent interactions, while communication may remove uncertainty or provenance from upstream evidence and produce unsupported consensus.
This PhD will develop methods for uncertainty-aware communication, alignment and adaptive orchestration in LLM-based multi-agent systems.
A first focus will be to quantify how different sources of uncertainty are represented and transformed as information passes between agents. Natural-language communication will be compared with structured uncertainty representations, confidence scores, shared memory, latent representations and hybrid approaches. Controlled perturbations of input evidence and agent outputs will be used to trace uncertainty and error propagation across multi-step text and multimodal reasoning.
System-level methods will then be developed to monitor uncertainty, evidential sufficiency, provenance and inter-agent disagreement. Evidence attribution and consistency measures will assess whether agents' claims and decisions remain supported by retrieved or tool-generated evidence, enabling the detection of uncertainty loss, conflicting beliefs and unsupported consensus.
Agent coordination will be formulated as a sequential decision problem. At each reasoning step, an adaptive orchestrator will determine which specialist or verifier agent to invoke, whether further evidence is required, whether previous reasoning should be revised, and when to stop, abstain or request human input. Decisions will be guided by uncertainty, disagreement, expected information gain, risk and computational cost. Reinforcement learning and related optimisation methods will be investigated for learning routing, verification and stopping policies from interaction trajectories and feedback.
Evaluation will use controlled and long-horizon benchmarks to assess uncertainty propagation, calibration, evidence alignment, error amplification, and robustness to missing, noisy or conflicting evidence. It will examine whether adaptive orchestration improves reliability and efficiency over standard multi-agent approaches, using measures of decision quality, computational cost, and appropriate verification, abstention and human escalation.
The methods will be investigated in environmental intelligence, where specialised agents integrate heterogeneous environmental, atmospheric and geospatial data to answer complex natural-language questions. This provides a challenging real-world setting involving multiple sources of uncertainty, conflicting evidence and decisions requiring different forms of expertise and human oversight, with pollution as a primary application area.
Desirable Student Background:
A strong background in computer science, artificial intelligence, machine learning and natural language processing is desirable. The student should have good programming skills, preferably in Python and modern deep-learning frameworks, and a strong interest in large language models and agentic AI. Experience with multi-agent systems, reinforcement learning, uncertainty quantification or multimodal learning would be advantageous. Strong analytical skills and an interest in developing and experimentally evaluating new AI methods are particularly important.
How to apply:
Please apply under the University of Manchester application portal and select 'PhD in Artificial Intelligence' (OAA Application form).
When applying, you’ll need to specify the full name of this project, the name of your supervisor, if you already having funding or if you wish to be considered for available funding through the university, details of your previous study, and names and contact details of two referees.
Your application will not be processed without all of the required documents submitted at the time of application, and we cannot accept responsibility for late or missed deadlines. Incomplete applications will not be considered.
You will be asked to upload the following supporting documents:
- Final Transcript and certificates of all awarded university level qualifications
- Interim Transcript of any university level qualifications in progress
- CV
- Supporting statement: A one or two page statement outlining your motivation to pursue postgraduate research and why you want to undertake postgraduate research at Manchester, any relevant research or work experience, the key findings of your previous research experience, and techniques and skills you’ve developed. (This is mandatory for all applicants and the application will be put on hold without it).
- Contact details for two referees (Please ensure referee email addresses are official university or workplace accounts, as we may need to verify the references.)
- English Language Certificate (if applicable)
We actively encourage applicants from diverse career paths and backgrounds and from all sections of the community, regardless of age, disability, ethnicity, gender, gender expression, sexual orientation and transgender status.
We also support applications from those returning from a career break or other roles. We are dedicated to supporting work-life balance and offer flexible working arrangements to accommodate individual needs. Our selection process is free from bias, and we are committed to ensuring fair and equal opportunities for all applicants.
We are dedicated to supporting work-life balance and offer flexible working arrangements to accommodate individual needs, including a part-time study option.
Join us and help grow the UK’s national capability in artificial intelligence, part of the UK’s modern industrial strategy. Everyone is welcome if you are passionate about uncertainty quantification, decision making with humans in the loop and decision making for machine learning systems and their applications in Science and Engineering, and looking to develop your research and innovation skills at the doctoral level to further your career in the AI sector.
Equality, diversity and inclusion is fundamental to the success of The University of Manchester, and is at the heart of all of our activities. We know that diversity strengthens our research community, leading to enhanced research creativity, productivity and quality, and societal and economic impact. (Equality, diversity and inclusion | The University of Manchester)
We actively encourage applicants from diverse career paths and backgrounds and from all sections of the community, regardless of age, disability, ethnicity, gender, gender expression, sexual orientation and transgender status.
Funding Notes
Fully funded AI UKRI CDT 4 year program; The Project will be based in University of Manchester.
The UKRI AI CDT in Decision Making for Complex Systems is part of the TechFirst offering stipends of £30,000 for students eligible for home fee status. The start date is September 2027.
References
Qian, L., Peng, X., Wang, Y., et al., Huang, J., & Ananiadou, S. (2026). When Agents Trade: Live Multi-Market Trading Benchmark for LLM Agents. WWW 2026 Industry Track.
Xiong, G., Deng, Z., Wang, K., et al., Huang, J., Ananiadou, S., & Xie, Q. (2025). FLAG-TRADER: Fusion LLM-Agent with Gradient-based Reinforcement Learning for Financial Trading. Findings of ACL 2025, 13921–13934.
Zhang, X., Wei, Q., Yuan, C., & Ananiadou, S. (2026). Reasoning Confidence is not Policy-Invariant: Calibrating Prefix Success under Model and Budget Shift. EMNLP 2026 Main Conference.
Yu, Z., Belinkov, Y., & Ananiadou, S. (2025). Back Attention: Understanding and Enhancing Multi-Hop Reasoning in Large Language Models. EMNLP 2025, 11257–11272.
Jiang, X., Georgaka, S., Rattray, M., & Álvarez, M. A. (2025). Scalable Multi-Output Gaussian Processes with Stochastic Variational Inference. Transactions on Machine Learning Research.
Jiang, F., Xia, J., Yu, J., Sun, M., Coe, H., Topping, D., Liu, D., Li, Z. J., & Zheng, Z. (2026). CAAL: Confidence-Aware Active Learning for Heteroscedastic Atmospheric Regression. arXiv:[contact details available in the full listing].
