The article "Empowering Biomedical Discovery with AI Agents" by Shanghai Gao et al. explores the potential of AI agents to revolutionize biomedical research. The authors envision "AI scientists" as systems capable of skeptical learning and reasoning, integrating AI models, biomedical tools, and experimental platforms to enhance discovery workflows. These AI agents combine human creativity and expertise with AI's ability to analyze large datasets, navigate hypothesis spaces, and execute repetitive tasks. Key features of these agents include: 1. **Skeptical Learning and Reasoning**: AI agents can break down complex problems into simpler tasks, use large language models (LLMs) and generative models for structured memory, and incorporate scientific knowledge and biological principles. 2. **Task Proficiency**: AI agents can perform various tasks, plan discovery workflows, and self-assess to identify and mitigate knowledge gaps. 3. **Collaboration**: Agents can cooperate through conversations with humans and other agents, seeking feedback and critique. 4. **Adaptability**: AI agents can adapt to new biological insights and incorporate the latest scientific findings, refining hypotheses based on experimental results. 5. **Ethical Considerations**: The authors emphasize the need for safeguards to prevent harm and ensure responsible implementation. The article also discusses the evolution of data-driven models in biomedical research, including databases, search engines, machine learning models, and interactive learning models. It highlights the advantages of AI agents over traditional methods in terms of efficiency, automation, and continuous, high-throughput research. The authors propose two main approaches for building AI agents: LLM-based agents and multi-agent systems. LLM-based agents are programmed to perform various roles, while multi-agent systems combine specialized agents with human experts to address complex tasks more effectively. The article provides examples of AI agents in genetics, cell biology, and chemical biology, illustrating their capabilities at different levels of autonomy. Finally, the authors outline a roadmap for building AI agents, focusing on modules such as perception, interaction, memory, and reasoning. They emphasize the importance of multi-modal perception and interaction capabilities to enable agents to interact with diverse data modalities and environments. The article concludes by discussing the challenges and limitations of AI agents in biomedical research, emphasizing the need for responsible development and deployment to ensure scientific integrity and avoid overreliance on agents.The article "Empowering Biomedical Discovery with AI Agents" by Shanghai Gao et al. explores the potential of AI agents to revolutionize biomedical research. The authors envision "AI scientists" as systems capable of skeptical learning and reasoning, integrating AI models, biomedical tools, and experimental platforms to enhance discovery workflows. These AI agents combine human creativity and expertise with AI's ability to analyze large datasets, navigate hypothesis spaces, and execute repetitive tasks. Key features of these agents include: 1. **Skeptical Learning and Reasoning**: AI agents can break down complex problems into simpler tasks, use large language models (LLMs) and generative models for structured memory, and incorporate scientific knowledge and biological principles. 2. **Task Proficiency**: AI agents can perform various tasks, plan discovery workflows, and self-assess to identify and mitigate knowledge gaps. 3. **Collaboration**: Agents can cooperate through conversations with humans and other agents, seeking feedback and critique. 4. **Adaptability**: AI agents can adapt to new biological insights and incorporate the latest scientific findings, refining hypotheses based on experimental results. 5. **Ethical Considerations**: The authors emphasize the need for safeguards to prevent harm and ensure responsible implementation. The article also discusses the evolution of data-driven models in biomedical research, including databases, search engines, machine learning models, and interactive learning models. It highlights the advantages of AI agents over traditional methods in terms of efficiency, automation, and continuous, high-throughput research. The authors propose two main approaches for building AI agents: LLM-based agents and multi-agent systems. LLM-based agents are programmed to perform various roles, while multi-agent systems combine specialized agents with human experts to address complex tasks more effectively. The article provides examples of AI agents in genetics, cell biology, and chemical biology, illustrating their capabilities at different levels of autonomy. Finally, the authors outline a roadmap for building AI agents, focusing on modules such as perception, interaction, memory, and reasoning. They emphasize the importance of multi-modal perception and interaction capabilities to enable agents to interact with diverse data modalities and environments. The article concludes by discussing the challenges and limitations of AI agents in biomedical research, emphasizing the need for responsible development and deployment to ensure scientific integrity and avoid overreliance on agents.