ELLIS UnConference 2026 · Paris

Multi-Agent AI for
Scientific Discovery

Learning, Coordination and Evaluation

Paris, France Two-hour workshop
Project pitches & posters Register through ELLIS

Invited talks · Project pitches · Panel discussion · Posters

About the workshop

When does collaboration
improve scientific discovery?

Scientific discovery involves generating hypotheses, choosing experiments, analysing results and checking conclusions. Teams of AI agents could combine complementary expertise and challenge each other’s reasoning—but they also introduce communication costs, duplicated effort and the risk of false consensus.

This workshop brings together researchers in multi-agent learning, human–AI collaboration and AI for Science to ask when agent teams deliver meaningful scientific gains. We will connect learning and coordination with shared memory, experimental design and rigorous evaluation under matched resource budgets.

Call for contributions

Bring an idea. Share your research.

We invite contributions on multi-agent AI for scientific discovery, including coordination, shared memory, human–AI collaboration, experimental design and evaluation. Choose the format that best fits what you want to discuss.

5-minute pitches

Call for Project Pitches

Ideas, open problems & collaboration

Share an early-stage or ongoing project, a concrete research question, or a promising direction that would benefit from community feedback. Completed results are not required.

  • What scientific problem are you tackling, and why might multiple agents help?
  • What is your proposed approach or main bottleneck?
  • What feedback, expertise or collaboration are you looking for?

Selected pitches will take place during the workshop: 4 minutes to present and 1 minute for a question.

Main ELLIS UnConference poster session

Call for Poster Presentations

Research contributions & in-depth discussion

Present research on methods, systems, benchmarks or scientific applications, with a concrete contribution to discuss. Posters provide space for detailed conversations about your approach, evidence and open questions.

  • Explain the research question, approach and key findings.
  • Highlight what multi-agent collaboration contributes.
  • Share limitations and questions for further discussion.

Workshop organizers will select posters for the main ELLIS UnConference poster session, outside our two-hour workshop.

Submit a pitch or poster

The form takes 5–10 minutes to complete.

Submission deadline
18 November 2026
Notification
23 November 2026

Submitting a contribution does not register you for the event. Presenters must also register through ELLIS.

Invited speakers

Perspectives across the field

Confirmed speakers
Talk titles to be announced

Samuel Kaski

Samuel Kaski

University of Manchester
Aalto University

Founding Director, ELLIS Institute Finland

Jun Wang

Jun Wang

University College London

Multi-agent learning and decision making

Tentative programme

Two hours of shared inquiry

Session time and room
to be announced

The programme combines invited talks, short project pitches, a moderated panel and a closing discussion. Times below are minutes from the start of the session.

Tentative workshop schedule; elapsed minutes from session start
MinutesSession
00–10Opening

When does multi-agent collaboration improve discovery?

Yali Du · King’s College London

10–70Invited talks

Learning, coordination and scientific evaluation

Samuel Kaski · Jun Wang · Zafeirios Fountas

Three 20-minute contributions, including brief questions. Speaker order and talk titles will be announced.

70–85Project pitches

Ideas, open problems and opportunities to collaborate

Three selected projects · Five minutes each

Four minutes to present an ongoing idea, followed by one minute for a question. See the call for pitches.

85–115Panel & audience Q&A

When does multi-agent collaboration improve scientific discovery?

Invited speakers · Moderated by Yali Du

Four discussion themes, concrete examples and open questions from the audience.

115–120Collective synthesis

Evaluation criteria and research priorities

A shared discussion to identify three to five priorities for future research.

Programme subject to updates. If an additional speaker is confirmed, the invited-talk block will comprise four 15-minute contributions.

Selected posters will be presented in the main ELLIS UnConference poster session. See the main event schedule for the latest timing.

Panel discussion

Four questions to explore together

When do agent teams outperform a single agent?

What demonstrates a coordination benefit under matched compute and experiment budgets, beyond simply running more attempts?

How should agents divide work and share knowledge?

Which roles need specialisation? How can shared memory, evidence provenance and productive disagreement support a team?

Where should scientists remain in the loop?

When should agents seek expert input, and how should people guide hypotheses, experimental decisions and unexpected findings?

How can we evaluate and trust collective discoveries?

How should benchmarks assess validity, novelty and reproducibility while exposing error propagation and false consensus?

Join the conversation

Meet us in Paris

Part of the ELLIS UnConference 2026.
Registration is managed by the main event.

Registration & event information
Date
Venue
Palais des Congrès de Paris
2 Place de la Porte Maillot, 75017 Paris
Session
Two hours · Time and room to be announced

Organizers

Bringing the community together

Yali Du

Yali Du

King’s College London

Organizer & session chair

Get in touch

Questions about the workshop?

Contact Yali Du at yali.du@kcl.ac.uk.

Back to top