Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Agentic Discovery: Closing the Loop with Cooperative Agents

J. Gregory Pauloski, Kyle Chard, Ian T. Foster

TL;DR

This paper argues that human bottlenecks in hypothesis generation, experimental design, and resource management limit the pace of data-driven science. It advocates agentic discovery via federations of cooperative agents that can autonomously navigate the scientific process, demonstrated through a case study in carbon-capture MOFs using the MOFA workflow. The authors outline a concrete architecture for agent roles (objective, knowledge, prediction, experiment, analysis, publish) and discuss key technical challenges (interfaces, governance, infrastructure, mobility, provenance) and risks (trust, security, attribution). They foresee a near-term path of incremental agent augmentation toward fully autonomous discovery within five to ten years, with substantial implications for scalability, reproducibility, and interdisciplinary collaboration.

Abstract

As data-driven methods, artificial intelligence (AI), and automated workflows accelerate scientific tasks, we see the rate of discovery increasingly limited by human decision-making tasks such as setting objectives, generating hypotheses, and designing experiments. We postulate that cooperative agents are needed to augment the role of humans and enable autonomous discovery. Realizing such agents will require progress in both AI and infrastructure.

Agentic Discovery: Closing the Loop with Cooperative Agents

TL;DR

This paper argues that human bottlenecks in hypothesis generation, experimental design, and resource management limit the pace of data-driven science. It advocates agentic discovery via federations of cooperative agents that can autonomously navigate the scientific process, demonstrated through a case study in carbon-capture MOFs using the MOFA workflow. The authors outline a concrete architecture for agent roles (objective, knowledge, prediction, experiment, analysis, publish) and discuss key technical challenges (interfaces, governance, infrastructure, mobility, provenance) and risks (trust, security, attribution). They foresee a near-term path of incremental agent augmentation toward fully autonomous discovery within five to ten years, with substantial implications for scalability, reproducibility, and interdisciplinary collaboration.

Abstract

As data-driven methods, artificial intelligence (AI), and automated workflows accelerate scientific tasks, we see the rate of discovery increasingly limited by human decision-making tasks such as setting objectives, generating hypotheses, and designing experiments. We postulate that cooperative agents are needed to augment the role of humans and enable autonomous discovery. Realizing such agents will require progress in both AI and infrastructure.
Paper Structure (12 sections, 2 figures)

This paper contains 12 sections, 2 figures.

Table of Contents

  1. What is an Agent?
  2. An Era of Agentic Discovery
  3. A Case Study in Materials Discovery
  4. Closing the Loop with Agents
  5. Evolving Human Responsibilities
  6. Key Technical Challenges
  7. Discovery and Interfaces
  8. Access Control and Sharing
  9. Infrastructure
  10. Agent Mobility
  11. Provenance and Reproducibility
  12. Uncertainties and Risks

Figures (2)

  • Figure 1: The discovery cycle of metal-organic frameworks (MOFs) for carbon capture is largely human-driven (orange stages). While some aspects have been automated (blue stages), such the AI generation and simulation of MOFs in the agentic MOFA system, human responsibilities limit the rate of MOF discovery.
  • Figure 2: The scientific method is an iterative process (stages depicted in the central loop). Specialized agents (depicted as boxes with corresponding stages indicated by color) can carry out the stages autonomously. Agents can also transcend stages to enable long-term planning, exploration, and safety.