Introducing 'Accelerating Astrophysical Discovery'

Accelerating Astrophysical Discovery is an open-science consortium steering how humans and machines will jointly study the cosmos.

Science was not built to work with AI. Its workflows, infrastructure, and norms developed around human researchers. Making human–machine collaboration productive requires us to rethink how science is done, while protecting the practices that make scientific knowledge trustworthy and useful.

Our two pillars

Formalising science

We aim to make the structure of science explicit: its concepts, models, assumptions, hypotheses, evidence, and the relationships between them. A shared scientific formalism should let humans and machines express ideas, trace how claims are supported or challenged, and test proposed connections, with uncertainty and empirical validation at its core.

Guiding a multimodal joint-representation reasoning model

We aim to guide the development of a multimodal joint-representation reasoning model capable of accelerating the discovery of new links within this scientific formalism. By learning across observations, simulations, language, and their context, the model should help researchers propose and test relationships that might otherwise remain unexplored.

The formalisation and the model go hand in hand in physics, much as Lean and LLMs do in mathematics. The formalism gives reasoning a shared structure; the model helps explore new connections within it. In physics, those connections must also be tested against empirical evidence. Neither pillar is developed in isolation: each informs the other.

Accelerating serendipity, in a way that is healthy for science

Our ultimate goal is to accelerate serendipity: to help researchers uncover unexpected connections and test ideas faster. We want to guide a future in which human–machine collaboration deepens our understanding of the universe, with humans steering scientific priorities and exercising judgement, while preserving scientific rigour, transparency, reproducibility, and the ability to teach what we learn to the next generation.