Some of the biggest advances in artificial intelligence have emerged from industrial research labs with enormous computing resources, large multidisciplinary teams and the freedom to pursue ambitious ideas at speed. UK universities now want to see whether academia can compete by changing the way AI research itself is organised.
The University of Oxford, University College London and Imperial College London have officially launched the British Open-ended Learning and Discovery programme, or BOLD, bringing five established research groups together into a single collaborative AI laboratory.
Supported by UK Research and Innovation (UKRI), the programme is intended to give academic researchers access to the teams, computing power and shared infrastructure needed to pursue high-risk ideas that might be difficult for an individual university research group to tackle alone.
BOLD also wants to look beyond the large language model approach that currently dominates much of artificial intelligence. Its research interests include new ways for machines to learn, human-centred AI and more resource-efficient systems, with researchers encouraged to pursue fundamental questions rather than simply making existing models larger.
Advertisement
The programme brings together the FLAIR and WhiRL groups at Oxford with DARK and LASP at UCL and AIRL at Imperial. Technology companies including NVIDIA and AMD are supporting the initiative through computing resources, infrastructure and technical expertise, while other corporate partners are funding postgraduate researchers.
Access to computing power has become an increasingly important issue for academic AI research. Training and experimenting with advanced models can require costly specialist hardware, creating a substantial gap between what individual university groups can attempt and the resources available to the world’s largest technology companies.
BOLD’s response is to pool those resources. Instead of researchers operating within separate groups led by individual principal investigators, teams will share infrastructure, computing capacity and research practices across institutions. The ambition is to combine the openness of university research with some of the scale and speed associated with major commercial AI laboratories and start-ups.
Open science is intended to remain central to that model. BOLD plans to use open-source software, transparent research practices and shared infrastructure so that results can be scrutinised, reproduced and built upon by researchers outside the programme.
There is also an education and skills dimension. BOLD plans to recruit researchers at different career stages, including senior BOLD Fellows, while corporate partners are supporting PhD positions. The programme hopes that concentrating talent and resources across several universities will help the UK attract and retain researchers who might otherwise be drawn towards better-resourced commercial AI laboratories.
Advertisement
The initiative was formally introduced through the BOLD Launch Festival, which brought researchers and industry partners together for talks, collaborative sessions and the development of new research ideas. Its deliberately open-ended format reflected the programme’s wider ambition to encourage researchers to challenge established assumptions about both artificial intelligence and the way scientific research is conducted.
Whether the model produces major AI breakthroughs remains to be seen, but BOLD represents an interesting experiment in its own right. At a time when cutting-edge AI increasingly requires enormous amounts of computing power, specialist expertise and collaboration, the programme is asking whether universities need to redesign the research lab itself to keep pace.