BrainPilot: Automating Brain Discovery with Agentic...

BrainPilot: Automating Brain Discovery with Agentic... | AI Research

Key Takeaways

BrainPilot: Automating Brain Discovery with Agentic Research BrainPilot is an open-source, multi-agent system designed to accelerate brain science research b...
Understanding the brain increasingly depends on integrating evidence across scales, modalities, and disciplines.
Addressing a single research question therefore requires a coordinated sequence of operations, from surveying prior work to executing analyses and interpreting results in light of domain knowledge.
AI agents promise to accelerate this process, but current agents lack domain expertise in brain science, may fabricate claims, drift during multi-step reasoning, and offer few defined points for expert intervention.
These failures are especially costly in brain science, where conclusions feed into downstream scientific claims and depend on laboratory-specific expertise and careful human judgment.

Paper AbstractExpand

Understanding the brain increasingly depends on integrating evidence across scales, modalities, and disciplines. Addressing a single research question therefore requires a coordinated sequence of operations, from surveying prior work to executing analyses and interpreting results in light of domain knowledge. AI agents promise to accelerate this process, but current agents lack domain expertise in brain science, may fabricate claims, drift during multi-step reasoning, and offer few defined points for expert intervention. These failures are especially costly in brain science, where conclusions feed into downstream scientific claims and depend on laboratory-specific expertise and careful human judgment. We present \textbf{BrainPilot} a \textbf{fully open-source} multi-agent system that accelerates brain science research with traceable logs and agent-verified results. A principal investigator (PI) agent coordinates specialist agents grounded in curated domain knowledge: a unified brain science knowledge base containing 7{,}233 indexed items and a skill library of 72 reusable methodology units across seven research domains. Every major step is recorded in the Graph of Trace, an auditable record that links subgoals, tool use, evidence, and claims and allows researchers to follow and inspect the workflow. An Auditor agent further integrates fabrication checking into the workflow. For evaluation, we run three brain science tasks from Agents' Last Exam, introduce our own benchmark, \textbf{BrainPilotBench-v0}, and present additional end-to-end case studies. Across these evaluations, BrainPilot with an open-source backbone model attains performance comparable to state-of-the-art agent framework with less costs.

BrainPilot: Automating Brain Discovery with Agentic Research
BrainPilot is an open-source, multi-agent system designed to accelerate brain science research by automating complex, multi-step workflows. While AI agents have shown promise in scientific fields, they often struggle with domain-specific expertise, hallucinated claims, and a lack of transparency. BrainPilot addresses these challenges by grounding its agents in curated neuroscience knowledge and providing a "human-in-the-loop" framework where every research step is recorded, auditable, and verified by an independent agent.

A Team of Specialist Agents

The system operates under the guidance of a Principal Investigator (PI) agent, which acts as the central coordinator. The PI decomposes a user’s research goal into smaller tasks and delegates them to a team of specialist agents:

Librarian: Surveys literature and provides knowledge grounding.
Experimentalist: Designs procedures, controls, and analysis plans.
Engineer: Translates designs into code and executes data analysis.
Writer: Drafts and polishes final scientific documents.
Auditor: Independently checks the work for fabrications by comparing claims against evidence in the session workspace.

Grounding in Neuroscience Knowledge

To ensure the agents perform with scientific rigor, BrainPilot is built upon two primary assets. First, a unified knowledge base contains 7,233 indexed items, including neuroscience textbooks and research papers, which agents use to retrieve relevant background information. Second, a skill library provides 72 reusable methodology units across seven research domains. These skills range from descriptive methods to executable code for established neuroscience tools, such as those used for neuroimaging and behavioral analysis. By using a "skills-first" approach, agents ensure their methodological choices are anchored in established scientific practice rather than relying solely on their internal training data.

Transparency Through the Graph of Trace

A core feature of BrainPilot is the Graph of Trace, an auditable record that documents the entire research process. As agents complete tasks, they log their actions, tool usage, evidence, and generated claims into this structured graph. This allows researchers to inspect the workflow at any point, trace the origin of a specific conclusion, and evaluate the methodological decisions made by the agents. By making the "black box" of AI reasoning visible, the system allows human experts to maintain control and provide judgment where it is most needed.

Performance and Evaluation

The researchers evaluated BrainPilot using tasks from the "Agents' Last Exam," a new benchmark called BrainPilotBench-v0, and several end-to-end case studies. The results indicate that BrainPilot, even when using an open-source backbone model, achieves performance comparable to state-of-the-art agent frameworks while operating at a lower cost. These evaluations demonstrate that the system can effectively handle complex, multi-step research workflows while maintaining the scientific validity required for brain science.