El próximo jueves 6 de noviembre de 2025 a las 16:00 en salón de grados de la ETSI Informática tendrá lugar el seminario «Unleashing Scientific Creativity using AI Agent Networks» a cargo de la profesora Mihaela van der Schaar (University of Cambridge) dentro de la serie «La informática en la encrucijada» al que está toda la comunidad educativa invitada
We are entering a new era in which vast available scientific datasets invite us to move beyond narrowly defined questions and embrace a more open-ended, creative mode of discovery. This process, which we call hypothesis hunting, is about unleashing scientific creativity: exploring vast spaces of possible explanations and connections, searching not just for answers but for new ways to innovate and do research.
In this talk, I will present a framework for enabling such discovery, based on AI agent networks. The framework, called A-Science, models the scientific process as an evolving interaction between intelligent agents, their networks of collaboration, and the norms by which results are evaluated. Its implementation, ASCollab, is a distributed ecosystem of large-language-model research agents with diverse data, research tools and reasoning strategies. These agents work together—proposing, critiquing, and refining ideas in a continuous cycle that not only echoes the social dynamics of human science but also explores new, non-human ways of connecting ideas, evidence, and reasoning.
Our experiments, conducted on large-scale biomedical datasets, show that these networks can sustain exploratory research at scale—recovering known biomarkers, extending biological pathways, and proposing new therapeutic hypotheses. More broadly, they demonstrate that structured interactions among AI agents can cultivate creativity and accelerate discovery, not by replacing scientists, but by redefining their roles within the innovation process participating in the scientific network as researchers, as reviewers guiding the evaluation of ideas, and/or as architects shaping the norms by which the entire network operates and evolves.
