合成队友：利用数字孪生优化关键任务中的人工智能性能.pptx

1、Synthetic Teammates:Challenges Developing and Validating Human Digital Twinsto Optimize Human-AI Performance in Critical Missions,Svitlana Volkova,PhDChief of AIOffice of Science and TechnologyAptima,Inc.,New York Scientific Data Summit 2024:Addressing Data Challenges in Digital TwinsNew York,NY Sep

2、tember 16,2024,How?Compound AI system with LLMs,MFMs,agents,tools etc.to create high-fidelity HDTs Key Components:HDT development using models,agents and toolsEvaluation of HDT fidelity and HAT performanceEmotional-cognitive measures in HDTsCausal investigations into factors affecting HDT and HAT pe

3、rformanceOperational Impact:Revolutionizing training,decision-making,and teamwork in high-stakes environments,Overview,Paradigm Shift in Human-AI Collaboration HDTs represent a transformative approach to integrating AI into human teams,Human,HDT,interacts,Initiates,Small-scale real-world training(ob

4、servational data),Knowledge base,Large-scale simulated training(interventional data),Past Training,RAG,Semanticmemory,Declarativememory,Episodicmemory,Experiences,AI Agents,HDT,Interventions,interacts,Predictive tools:Anticipate,Causal tools:Explain,Shared Mental Model,Profile:PersonalityCompetencie

5、s.,Real-world,Simulations,Compound AI Systems1 Models,Agents,Tools and AIP calls,AI Agents,1https:/bair.berkeley.edu/blog/2024/02/18/compound-ai-systems/,Volkova et al.,Compound AI Ecosystem:Agents and Tools to Improve Training and Learning.Interservice/Industry Training,Simulation and Education Con

6、ference2024(To Appear).,Operational Scenarios,1 Freeman,J.,Huang,L.,Wood,M.,&Cauffman,S.J.(2022).Evaluating Artificial Social Intelligence in an Urban Search and Rescue Task Environment.In N.Gurney&G.Sukthankar(Eds.),Computational Theory of Mind for Human-Machine Teams(pp.7284).2 McNeese,N.J.,Demir,