How Must UK Science Adapt to the Age of AI?
Agentic AI is happening. Models can already reason across large bodies of information, write and run code, operate software and coordinate sequences of work. But intelligence matters only when it changes what we can do in the real world.
The UK has exceptional strengths in AI, science, engineering, universities, national facilities, life sciences and the NHS. But many of the institutions through which we conduct science were built for a slower world—one in which analysis, writing, review and coordination were expensive. Simply adding agents to today's grant applications, papers and laboratory workflows may give us more activity without giving us more progress. We want progress and we want to UK to build new industries rapidly as a result of this.
This is a small, informal call to arms from the a group of us that think we need to coordinate how the UK responds to AI-in-Science. We want to bring together people who believe science will have to change as intelligent work becomes cheaper and more abundant—and people who disagree about what that change should look like.
Among the questions we want to explore:
Is it either feasible or desirable for grants and papers to continue to exist in their present form: or can we design better replacements?
Do we need to build major autonomous cloud labs as national infrastructure, or do we need a marketplace for automation from a multitude of providers?
How can agentic AI help with field building? And what new institutional designs are enabled by the agentic era?
Should the scientific literature be replaced with discourse graphs, making claims and assumptions explicit and machine-interpretable?
How could agentic peer review be used to avoid scientific fields making mistakes? E.g. how could the candidate gene era of genetics have been avoided, using agentic AI?
How can agentic AI shorten the innovation OODA loop, so can develop solutions to bottlenecks quicker?
What must the UK do now if it wants to turn AI capability into scientific and technological advantage?
There will be no panel and no sequence of presentations. We will meet around pub tables, begin with a short provocation and spend the evening testing it. The aim is not to force consensus. It is to expose the important disagreements, identify what evidence or experiments could resolve them, and find people willing to help build what is missing.
The discussion will help revise the Network's initial manifesto and shape a wider series of gatherings on autonomous laboratories, scientific funding, reproducibility, publishing and the institutions of the agentic era.
If you can see a part of this future clearly—or think the premise is wrong—we would like you in the room.