British Open-ended Learning and Discovery (BOLD) Research Studentship

3.5-year DPhil/PhD studentship 

Project title: Multiple projects available - information on specific projects will be made available as soon as possible.

Supervisors: Prof Jakob Foerster

At the core BOLD is about two bets:

1.  Paradigm changing breakthroughs in AI that differentially improve the chances of a pluralistic, distributed, open-source AI stack which serves and empowers humanity.

2. New modes of operation for AI research in academia that are more focussed, better resourced, collaborative, agile, and transparent than existing models. These will both allow us to make these paradigm changing breakthroughs by empowering groups of people to work towards shared ambitious long-term goals and will also act as blueprints for others to follow.

We are now looking for students who are excited to turn this bold vision into a reality! Note, this call is for the main deadline in the current recruitment cycle, so if our mission resonates with you please apply in time. Furthermore, all specific studentships will be aligned with our overall research directions and approach outlined below.

Specifically, BOLD was created in June 2026 by folding 5 internationally leading labs into one effort—FLAIR and WhiRL (Oxford), DARK and LASP (UCL), and AIRL (Imperial). These labs have jointly published hundreds of papers in top venues and won best-paper awards at ICML, AAAI, and GECCO, and represent one of the highest concentrations of talent in Europe and globally. The founding faculty bring deep experience from the frontier labs shaping AI today and have started several high impact companies.

BOLD consolidates these existing world-class academic groups into a single co-located lab in Oxford, formally linked to all three universities. A lightweight three-phase research funnel—broad exploration, selective scale-up, then deep focused missions—lets students rapidly develop promising ideas into larger research projects while exploring new directions with minimal barriers, building on the team’s track record of turning early experiments into breakthroughs such as EGGROLL within months. The labs have a proven pipeline from research to real-world impact: alumni have gone on to OpenAI, Anthropic, DeepMind, and the UK AI Security Institute, the foundation-model unicorn Reflection AI, and have co-founded companies including Weco.ai (whose AIDE agent has been used by OpenAI and Meta), Helical, and a number of current companies in stealth.

We also have short introduction videos for BOLD and the PIs.

There are three initial research pillars:

Beyond Backpropagation

Rethinking the foundational neural network optimisation and exploring architectures beyond NN. This pillar targets limitations of backpropagation (high communication overhead, poor handling of non-differentiable components, reduced plasticity) by developing zeroth-order, population-based, low-rank, and hybrid methods (e.g., the EGGROLL algorithm) to unlock massive compute efficiency, decentralized, privacy-aware training, and novel architectures.

Human-Centric Learning & Discovery

This pillar develops methods and tools that empower groups of humans to work towards shared goals and values. We do this via two main approaches: The first one is by developing tools (including our BOLD Operating System) and novel machine learning approaches that empower all members of BOLD to be more productive. The second avenue is to develop algorithms that allow the incentive-compatible, decentralised training of foundation models. Importantly, by deploying the findings of this pillar within BOLD we can turn BOLD into a recursive co-improvement testbed for AI that empowers groups of humans.

Embodied Learning

Developing resource-agile paradigms for data- and compute-constrained physical systems (robotics/autonomous vehicles). Instead of monolithic cloud-dependent models, this focuses on autonomous skill discovery in simulation (e.g., OMNI-EPIC), resilient adaptation, modular hierarchies, and edge-optimized scalability for real-time onboard deployment. Crucially, this pillar also provides grounding of our overall efforts by deploying methods from Pillar 1 on real world problems.

Eligibility

These studentships are funded through a number of funders and are open to Home and overseas students (full award – home/overseas fees plus stipend).

Award Value

The studentship covers the full course fees plus a stipend (tax-free maintenance grant). The stipend is the UKRI minimum stipend for the first year plus £4,000, and at least this amount for a further two and a half years (pro-rata).

Candidate Requirements

Prospective candidates will be judged according to how well they meet the following criteria:

• Applicants are normally expected to be predicted or have achieved a first-class or strong upper second-class undergraduate degree with honours (or equivalent international qualifications), as a minimum, in computer science, engineering, physics, mathematics, statistics or other related disciplines. A previous master's qualification is not required.

• Excellent English written and spoken communication skills.

Application Procedure

If you have any technical questions about the DPhil Studentship, please email boldexec-team@eng.ox.ac.uk.

Candidates must submit a graduate application and are expected to meet the graduate admissions criteria. Details are available on the course page of the University website.

Please quote 27ENGIN_JF in all correspondence and in your graduate application.

Application deadline: Noon on 2 December 2026 (In line with the University admissions deadline set by the University).

Start date: October 2027