The Future of Fundamental Science Led by Generative Closed-Loop Artificial Intelligence

TL;DR

This work explores and investigates aspects of an AI-driven, automated, closed-loop approach to scientific discovery, including self-driven hypothesis generation and open-ended autonomous exploration of the hypothesis space.

Abstract

Recent advances in machine learning and AI, including Generative AI and LLMs, are disrupting technological innovation, product development, and society as a whole. AI's contribution to technology can come from multiple approaches that require access to large training data sets and clear performance evaluation criteria, ranging from pattern recognition and classification to generative models. Yet, AI has contributed less to fundamental science in part because large data sets of high-quality data for scientific practice and model discovery are more difficult to access. Generative AI, in general, and Large Language Models in particular, may represent an opportunity to augment and accelerate the scientific discovery of fundamental deep science with quantitative models. Here we explore and investigate aspects of an AI-driven, automated, closed-loop approach to scientific discovery, including self-driven hypothesis generation and open-ended autonomous exploration of the hypothesis space. Integrating AI-driven automation into the practice of science would mitigate current problems, including the replication of findings, systematic production of data, and ultimately democratisation of the scientific process. Realising these possibilities requires a vision for augmented AI coupled with a diversity of AI approaches able to deal with fundamental aspects of causality analysis and model discovery while enabling unbiased search across the space of putative explanations. These advances hold the promise to unleash AI's potential for searching and discovering the fundamental structure of our world beyond what human scientists have been able to achieve. Such a vision would push the boundaries of new fundamental science rather than automatize current workflows and instead open doors for technological innovation to tackle some of the greatest challenges facing humanity today.

Figures (3)

• Figure 1: A quantitative framework of domain-agnostic acceleration of scientific discovery with AI, its relationship with human-carried science, and the combination of human and machine. 'All knowledge' can be interpreted as potential knowledge if it were discoverable by humans through AI or by themselves. AI time $T'$ to conduct certain tasks is traditionally taken to be faster than $T$ by orders of magnitude King2004a as it is also more scalable. Still, its domain-dependency and relationship to $T"$ (human-machine hybrid) are likely highly domain-specific and has traditionally ignored closed-loopness or the removal of any human input.
• Figure 2: Bubble landscape of current approaches to AI from and for science. Bubbles may occur more than once when related to several larger domains. Some approaches may have alternative names or have been re-branded in certain contexts. Neuro-symbolic models have sometimes been referred to as 'intuitive' while some statistical-driven approaches have been labelled as 'cognitive computing'. Generative AI (GenAI) has made little to no contributions to fundamental science so far but has great potential. Large Language Models (LLMs) may significantly tap into and contribute to the exploratory capabilities of the scientific hypothesis space, given their capabilities to process human language in which all human science has been written. GenAI and LLMs are approaches of statistical nature, but it remains unexplored to what extent they may develop symbolic capabilities from statistical (e.g. linguistic) patterns.
• Figure 3: Visual representation of closed-loop full experimentation cycle for scientific discovery pathways, adapted and combining ideas from Kitano2016 and King2004a. LLMs can now facilitate closing this loop but require help to connect each module and process in a causal rather than only a statistical fashion.