Competition Funded PhD Project (UK Students Only)
Application deadline: Friday, December 04, 2026
Supervisors: Dr Umang Bhatt, Dr Mateja Jamnik
Department: Department of Computer Science and Technology, University of Cambridge
About the Project
AI_CDT_DecisionMaking
Details
Building powerful AI agents is not sufficient for deployment in high-stakes settings such as education, healthcare, and public safety [0]. It is crucial to understand how human decision-makers incorporate agents into their workflows to improve productivity without introducing untrustworthy behaviour. Prior human-AI interaction work studies assistant-style agents [1], yet the infrastructure around agent handoffs, oversight, and escalation remains poorly understood. In high-stakes domains, humans assisted by AI agents may need additional information before having agents act, may require oversight before letting agents execute tasks, or may need explanations before approving agent recommendations. Understanding *how* and *when* agents are appropriate assistants has become a principal research need.
These challenges connect to well-known problems in machine learning around selective prediction and task allocation, since selecting which agent to use and when amounts to comparing expected rewards across multiple agents [2]. This PhD project will study and operationalise agent orchestration, the science of coordinating AI agents and human experts in real-world settings. The project addresses three guiding questions: first, when does orchestration outperform a single-agent approach, especially once multiple objectives such as cost, latency, policy, and risk are considered [3]; second, how can we estimate human expertise, adapt the division of labour, and provide agents with the right information at the right time; and third, how should we switch among agents, including humans, under uncertainty and decide when to route, parallelise, vote, escalate, or defer to AI agents or humans [3]. This project will extend the human-AI interaction literature to multi-AI, multi-human settings with a focus on productivity and reliability, and will be grounded in a workflow registry that has captured examples of handoff triggers, consensus rules, and escalation ladders [4].
The student will advance the science of orchestration, for example through contextual-bandit methods that model user needs with unstructured rewards, and will deliver production-ready tooling for agent orchestration. Understanding how agents represent uncertainty will clarify how to use them in practice, since selective prediction has been used to gate concept-based explanations [5], confidence estimates [6], or predictions themselves [2, 3]. When epistemic uncertainty is high, orchestration can trigger information-gathering actions such as cross-agent probes, schema checks, or selective human queries that target the highest-value features. For example, Selective Concept Models allow stakeholders to decide which features to verify and to balance cognitive effort and benefit [5], while Intervention-Aware Concept Models integrate user corrections when concept feedback is available [7]. These types of interventions can be ported to the multi-agent setting to determine where oversight is needed in sequential decision-making tasks to improve performance, confidence, and efficiency. The project will deliver a formal decision-theoretic framework and accompanying algorithms for when and how AI agents should invoke human oversight, deciding when to act, defer, or acquire information under cost, uncertainty, and policy constraints. Validated through simulations and user studies, the project will result in an open-source orchestration toolkit that quantifies productivity gains (accuracy, throughput, cost-of-error). The project will be co-supervised by Mateja Jamnik, whose expertise in human-like reasoning, explainable AI, and intervention-aware modelling will ground the project in technical rigour and real-world deployment.
Student Desirable Background:
machine learning, DNNs, LLMs, strong maths
By submitting an application for a University of Cambridge project, you consent to your documents being shared with the relevant supervisory team at the partner institution. If you have questions about this process aidecisionscdt@manchester.ac.uk before applying.
How to apply
Please apply under the University of Manchester application portal and select 'PhD in Artificial Intelligence' (OAA Application form)
When applying, you must include the project’s full name, your supervisor’s name, funding status, academic history, and two referees’ contact details. Applications with missing documents will not be processed. Late or incomplete submissions will not be considered, and the university is not responsible for missed deadlines.
After you have applied 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 mandatory one- to two-page statement explaining your motivation for postgraduate research, why you chose Manchester, relevant academic or work experience, key research findings, and the techniques or skills you've gained. Applications will be put on hold if this is not provided.
- 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)
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.
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.
Funding Notes
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.
The Project will be based in University of Cambridge.
References
- The GenAI Divide. MIT NANDA (https://nanda.media.mit.edu/). Preprint 2025.
- Towards a science of human-AI decision making: An overview of design space in empirical human-subject studies. Lai et al. ACM FAccT 2023. https://dl.acm.org/doi/abs/10.1145/3593013.3594087
- SelectIT: Selective Instruction Tuning for LLMs via Uncertainty-Aware Self-Reflection. Liu et al. NeurIPS 2024. https://arxiv.org/abs/2402.16705
- Learning Personalized Decision Support Policies. Bhatt, et al. AAAI 2025. https://ojs.aaai.org/index.php/AAAI/article/view/33555
- Documenting Deployment with Fabric: A Repository of Real-World AI Governance. Jorgensen et al. AAAI/ACM AIES 2025. https://ojs.aaai.org/index.php/AIES/article/view/36636
- Selective Concept Models: Permitting Stakeholder Customisation at Test-Time. Barker et al. AAAI HCOMP 2024. https://ojs.aaai.org/index.php/HCOMP/article/view/27543
- Modulating Language Model Experiences through Frictions. Collins et al. Preprint 2024. https://arxiv.org/abs/2407.12804
- Learning to Receive Help: Intervention-Aware Concept Embedding Models. Espinosa Zarlenga et al. NeurIPS 2023. https://arxiv.org/abs/2309.16928
