为多智能体人工智能系统实现人机回路控制.pdf

1、 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.C N S 4 2 8Implementing Human-in-the-loop controls for multi-agent AI systemsDhiraj MahapatroPrincipal Specialist Solutions Architect,Amazon BedrockAmazon Web Servic

2、es(AWS)2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.AgendaEvolution of Agentic AIImportance of Human-in-the-loop(HITL)Implementing HITLTakeawaysResources 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Evolution of Agentic AIINCREASING AUTONOMY AND BUSINESS I

3、MPACTGenerative AI assistantsGenerative AI agentsAgentic AI systemsFollow a set of rules Automate repetitive tasksAchieve a singular goalAddress broader range of tasksAutomate entire workflowsFully autonomousMulti-agent systemsMimic human logic and reasoningM O R EH U M A NO V E R S I G H TL E S SH

4、U M A NO V E R S I G H T 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Foundation model inference on AWSAny ComputeFoundation ModelUserAmazon BedrockAmazon SageMakerAmazon Elastic Kubernetes Service(Amazon EKS)Amazon Elastic Container Service(Amazon ECS)Amazon Elastic Compute Cl

5、oud(Amazon EC2)2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.LLM inference with chat historyAny ComputeModelUser 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.LLM inference with history and context(RAG)Any ComputeModelUserEmbedding ModelVector StorePrompt+Co

6、ntext 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Real-time agent working with LLMAny ComputeModelUserEmbedding ModelVector StorePrompt+ContextAgentTools 2025,Amazon Web Services,Inc.or its affiliates.All rights reserved.Real-time agent with security and observabilityAny Compu