The First Workshop on Scientific Agents in the Digital World (SciAgentVerse)

About the workshop

Scientific discovery proceeds through repeated cycles of gathering evidence, proposing hypotheses, and testing them against observations. Scientific agents offer a way to coordinate these steps — connecting scientific literature and databases with computational analysis, simulation, and experimental validation, and using each result to guide the next investigation.

This workshop focuses on scientific inquiry conducted through digital environments, where agents interact with scientific knowledge and software to pursue a research objective. These environments let agents explore candidate explanations, conduct controlled computational tests, and examine results before committing to further investigation. We ask how agents can connect scientific reasoning with executable tests, reason about the limits of simulators and learned models, respond to unexpected observations, and preserve the rationale behind earlier decisions in multi-stage investigations.

How can agents use digital environments to produce findings that are scientifically meaningful and independently verifiable?

1

Evidence to hypotheses

How should agents combine literature, structured scientific data, and domain knowledge to formulate hypotheses with traceable evidence?

2

Planning under uncertainty

How should agents plan retrieval, simulations, and experiments under uncertainty and resource constraints — and decide when a hypothesis needs revision?

3

Evaluating discovery

How should evaluation connect the quality of individual agent decisions with scientific validity, novelty, reproducibility, and discovery efficiency?

Call for Papers

We invite submissions on agents that pose scientific questions, run computational experiments, and update their conclusions as new evidence comes in. Research papers, position papers, work-in-progress papers, and demos are all welcome, whether the focus is on methods or on applications to problems from any field of science.

Important dates

November 30, 2026Paper submission deadline
December 18, 2026Final notification to authors
December 25, 2026Camera-ready submission
February 19, 2027Workshop day at WSDM 2027, Cordis Hong Kong

All deadlines are 11:59 PM Anywhere on Earth (AoE).

Topics of interest

Topics include, but are not limited to:

Scientific knowledge and hypothesis generation. Methods for formulating testable hypotheses from scientific literature and domain knowledge, including identifying research gaps, reconciling conflicting findings, and using evidence to assess novelty and refine or reject hypotheses.
Simulation and experimental planning. Agent approaches to selecting informative tests under resource constraints, using scientific simulators and learned predictive models to guide exploration across scientific domains, including balancing candidate investigation with testing uncertain assumptions and determining where further validation is needed.
Agent workflow execution and memory. Systems that execute research workflows through scientific software while maintaining hypotheses, modeling decisions, and intermediate results; diagnosing failed runs and revising workflows while preserving scientific context and decision rationale.
Learning and adaptation for scientific investigation. Approaches that improve agent capabilities through demonstrations, interaction, and feedback, including foundation model adaptation, learning from unsuccessful attempts, continual learning, and transferring knowledge and skills across tasks, agent tools, and environments.
Collaborative scientific inquiry. Collaboration among specialized agents and researchers to coordinate expertise, examine hypotheses, and resolve conflicting interpretations, with attention to effective human intervention and clear responsibility for scientific decisions.
Evaluation and trustworthiness of scientific agents. Evaluation of scientific validity and reproducibility through traceable evidence and independent scrutiny of reported findings.

Submission guidelines

Format

Up to 8 pages for the initial submission, plus unlimited pages for references and supplementary materials.
Manuscripts must be submitted in PDF using the ACM conference template (sigconf), as used by WSDM 2027.

Review

Reviewing is double-blind. Submissions that are not properly anonymized will be desk-rejected.

Non-archival

The workshop is non-archival. Accepted papers will be posted on this website but will not be included in any proceedings.

Presentation & open science

WSDM 2027 is an in-person event; at least one author of each accepted paper should present on site in Hong Kong.
We encourage sharing code, datasets, simulation configurations, workflow descriptions, and evaluation resources where possible.