Bruno, an AI assistant designed to preserve a research group's working history, was presented at an ICML workshop
Bruno, an AI assistant designed to preserve a research group's working history, was presented at an ICML workshop
Bruno is designed to connect a research group's tasks with its decisions, working materials, and the people who produced them. The authors are preparing the first evaluation of the system in a computational biology course. No data are yet available on its use in an active laboratory.
A scientific paper usually presents the final state of a project: the question, method, data, and conclusion. The work that led to this result is scattered across messages, code branches, drafts, and people's memories. When a researcher leaves the group, the reason the team rejected a hypothesis or changed an experiment may leave with them.
Bruno coauthor Prashaanth Ranganathan writes that decisions get lost in Slack threads, unsuccessful experiments go undocumented, and a PhD student may carry half of a project in their head.
On July 11, Andreas Haupt and Prashaanth Ranganathan presented Bruno, a research group management tool, at the AI4Science workshop at the ICML conference. According to the authors, Bruno reads the team's working materials but cannot modify code, data, or manuscripts. Its stated scope is planning, coordination, and team memory.
Bruno is designed to read messages, the code repository, article drafts, and the experiment log, plan the next tasks, and connect results to those tasks. It is intended to preserve tasks, decisions, failures, and materials while recording who contributed each item. The authors expect that a new group member will be able to reconstruct the original question, identify which data changed the plan, and determine what the team did next.
The authors are planning the first pilot in a master's course in computational biology. They have not yet presented results from this pilot or data on Bruno's use in an active laboratory.
To reproduce a result, a group must preserve the data, code versions, and experimental protocol. Bruno is designed to connect these materials to the tasks and decisions that changed the course of the work.