VirSci - A multi-agent AI scientific research tool launched by the Shanghai Artificial Intelligence Laboratory

What is VirSci?

VirSci (Virtual Scientists) is a multi-agent AI scientific research tool launched by the Shanghai Artificial Intelligence Laboratory. It accelerates scientific innovation by simulating the collaborative process of a team of scientists. Based on large language models (LLMs), the system uses organized agent teams to collaboratively generate, evaluate, and refine research ideas, demonstrating innovation and influence in scientific idea generation that surpasses traditional single-agent systems. VirSci advances scientific discovery and, as a scientific research tool, explores the impact of different team compositions on innovation.

VirSci's main functions

• Collaborator SelectionSimulate the process of forming a team of scientists and selecting suitable collaborators to join the research team.
• Topic DiscussionTeam members discussed the research topic and determined the research direction.
• Idea GenerationTeam members proposed and refined research ideas, generating multiple potential research concepts.
• Novelty AssessmentThe novelty of the proposed ideas is evaluated, and the most innovative ideas are selected based on a comparison with existing literature.
• Abstract GenerationBased on the selected idea, generate an abstract of a scientific paper, including an introduction, objectives, methods, expected results, and conclusions.
• Self-reviewAfter generating the final abstract, a self-review is conducted to pre-check its novelty and ensure that it is not too similar to existing research.
• Knowledge base construction: Construct a knowledge base containing background information on scientists to provide intelligent agents with the necessary information for effective cooperation and discussion.
• Team discussion mechanismImplement team discussions at each step to improve output quality through iteration and refinement.

VirSci's technical principles

• Large Language Models (LLMs)VirSci leverages the advanced capabilities of large language models such as GPT-4o and Llama-3.1, which excel in understanding and generating natural language and can handle complex scientific discovery tasks.
• Multi-agent systemThe system consists of multiple intelligent agents, each simulating a scientist, and can collaborate, communicate, and solve research tasks, mimicking the dynamics of human teamwork.
• Digital twin technologyBased on the Retrieval Augmentation Generative (RAG) framework, VirSci creates digital twin agents of real scientists, which can access and use the scientists' knowledge base.
• Knowledge base and databaseThe system constructs a knowledge base containing background information on scientists, as well as a database containing past and contemporary papers, to provide agents with the necessary information for effective cooperation and discussion.
• Team discussion mechanismImplement a "team discussion" mechanism to support iterative internal and external refinement dialogues between agents during the generation of ideas and summaries, thereby improving output quality.
• Novelty assessmentThe novelty of ideas is assessed by comparing the generated abstracts with databases of past and contemporary papers, using metrics such as historical difference (HD), contemporary difference (CD), and contemporary influence (CI).

VirSci project address

• Project official website:open-sciencelab.github.io/Social_Science
• GitHub repository:https://github.com/open-sciencelab/Social_Science
• arXiv technical paper:https://arxiv.org/pdf/2410.09403

Advantages of VirSci over AI Scientist

• Simulation of collaborative natureVirSci is a multi-agent system based on large language models (LLMs) specifically designed to mimic the teamwork inherent in scientific research. Compared to single-agent systems like AI Scientist, VirSci organizes teams of agents to collaboratively generate, evaluate, and refine research ideas, more closely resembling how scientists work as teams to solve complex problems in the real world.
• Generation of innovative scientific ideasAccording to the experimental results, VirSci outperforms state-of-the-art single-agent methods such as AI Scientist in generating novel and impactful scientific ideas. VirSci, through multi-agent collaboration, improves alignment with contemporary research trends by an average of 13.8% and increases potential impact by 44.1%.
• Emergence of social behaviorThe VirSci experiments revealed social behavior among agents, consistent with important findings in the field of "science within science," such as how fresh teams often generate more innovative research. This suggests that VirSci, as a tool, has the potential for further exploration in studying collaborative mechanisms.
• End-to-end scientific collaboration processVirSci is the first multi-agent system to enable scientific collaboration in an end-to-end pipeline, from team organization to the generation of novel scientific ideas. This system can more comprehensively simulate the entire scientific research process, while AI Scientists focus more on automating single tasks.
• Experimental verificationVirSci has conducted extensive experimental validation in many aspects, including the team setup and the novelty of the generated scientific ideas, which provides more empirical support for its application in actual scientific research.

VirSci application scenarios

• Automated scientific discoveriesVirSci is used to automate the scientific discovery process, from hypothesis generation to experimental design, accelerating the progress of scientific research.
• Team research collaborationIn multidisciplinary teams, VirSci simulates collaboration among scientists, helping team members explore and solve complex research problems together.
• Generate innovative ideasVirSci generates new research ideas and concepts, providing innovative perspectives and solutions for scientific research.
• Education and trainingIn the field of education, VirSci serves as a teaching tool to help students understand the scientific research process and cultivate innovative thinking and teamwork skills.
• Research Project ManagementIn project management, VirSci helps researchers plan research projects, assign tasks, and track project progress.