AuroraGPT：原理、数据挑战和人工智能研究助理的开发.pptx

1、AuroraGPT:Rationale,Data Challenges and Development of an AIResearch Assistant,Franck CappelloLead of AuroraGPT EvaluationArgonne National Laboratory,AI as Research/Scientific Assistant,AI models are producing fundamental changes in the way we search,access,transform information/data,control,operate

2、,steer robots machines,vehicles,experimental facilities,simulate phenomena,and solve optimization problems.AI will inevitably profoundly change the research practice,perhaps much more than what the Internet did.As a research institution,Argonne is interested in developing AI foundation models specif

3、ically and optimized for Science.AI will inevitably,Accelerating Discovery:Accelerating All Steps of Research,https:/doi.org/10.1038/s41524-022-00765-z,Study,Hypothesis,Test,Assess,Report,Question,Start here,Ideally AI will help in all these steps,Accelerating Discovery:Digital Twin Example,https:/d

4、oi.org/10.1038/s41524-022-00765-z,Study,Hypothesis,Test,Assess,Report,Question,Start here,Ideally AI will help in all these steps,Define the Objectives and Scope,Data Collection and Integration,Develop a mathematical,simulation-based or AI based model,Simulate the behavior of the system under differ

5、ent conditions,leveraging physics-based models and incorporating real-world constraints,Compare the outputs of the digital twin against real-world observations,Publish paper(s)And software,What do we mean by:“AI scientific assistant”?,An AI-based system with:Scientific skillsReasoning,math,literatur

6、e understanding,integrity,safety Effective assistance(no hallucination!)Trustworthy in the different tasks related to scientific activitiesRelevant human and environment interaction modalities(communication skills)Understanding command(semantic of it),interface with tools and devices Degree of auton