[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fOuGcksCLspMaXsMhxfNlRaLOMM4ck7edAsBuiaucAAY":3},{"code":4,"msg":5,"data":6},200,"操作成功",{"id":7,"title":8,"content":9,"digest":10,"source":10,"coverPath":11,"thumbsCoverPath":12,"isTop":13,"isShow":14,"baseClick":13,"clickCount":15,"createTime":16,"typeId":17,"isNewest":18,"newsInfoTypeRespVo":19,"voiceUrl":22,"voiceSize":23,"taskId":24,"releaseTime":25,"titleEn":26,"contentEn":27,"voiceUrlEn":28,"taskIdEn":29,"voiceSizeEn":30},1504,"AWS神经符号AI突破 自动推理技术确保Agent可信度Nova Act达90%可靠度","\u003Cp>\u003Cstrong style=\"color: rgb(255, 153, 0); font-size: 18px;\">AWS Agent AI副总裁Swami Sivasubramanian在re:Invent 2025主题演讲中，特别邀请AWS杰出科学家、自动推理领域权威Byron Cook登台，阐述如何通过神经符号AI（Neuro-Symbolic AI）技术，确保AI Agent真正可信。\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002Fed907f99093848b99d1003844aacd9f3\u002F图片2.png\" width=\"485\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cspan style=\"color: rgb(255, 153, 0); font-size: 18px;\"> \u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">AWS神经符号AI突破 自动推理技术确保Agent可信度Nova Act达90%可靠度。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami坦言AWS早期构建Agent原型时，曾出现“模型在API调用时产生幻觉”的问题。这衍生一个根本性疑问：如何让Agent真正可信？他指出LLM可能在复杂规则面前出错，或推理存在逻辑错误，一旦场景涉及金融交易、客户信任等关键环节，单纯统计方法便远远不足。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002F7680b0e2229644ffb31f426545844317\u002F图片3.png\" width=\"471\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Byron Cook开场便提出尖锐问题：“你会把信用卡交给Agent吗？就像交给一个几岁的小童去购物一样，他可能会帮你买到想要的商品，但最终你也可能得到一大堆不需要的东西。”\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Cook详细阐述神经符号AI这一突破性方法。他解释自动推理是在数学逻辑中搜索和仔细检查证明的学科，与2,000年前欧几里得证明定理的方法相同。AWS在内部系统使用此技术已超过十年，现在将其应用于Agent AI。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Cook通过三个生动例子，解释如何将自动推理应用于AI Agent：\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">验证输出结果：自动推理工具会像严谨的数学老师一样，逐步验证LLM推理过程，若发现错误则推回重试，形成反馈循环。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">训练数据生成：就像用标准答案的习题集训练学生。使用Lean定理证明器创建无限量“标准答案”，让AI从一开始便学会正确推理\u003C\u002Fspan>。\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002Fc1b9af52c318457399d6783991029db3\u002F图片4.png\" width=\"459\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">约束解码：就像汽车的车道保持系统，当偏离车道时，方向盘会自动微调拉回。在AI系统中，当模型试图回答“法国首都是什么”并开始输出字母“B”时，自动推理系统会实时介入，像导航一样将其引跳转至正确字母“P（Paris）”。这种实时纠偏确保AI输出始终符合逻辑规则。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Cook也强调Kiro的规范驱动开发：Kiro可分析应用程序、识别验收标准并转换为规范，进而指导程序代码生成、测试生成，甚至证明程序正确性。他随后详解昨日发布的AgentCore Policy：只需用自然语言描述允许的操作，如阻止任何Agent在AWS生产账户资源上执行更新操作，系统便会将其自动转换为AWS两年前开源、语义经Lean定理证明器形式化的Cedar策略语言，且这些策略可通过自动推理验证主权、隐私、安全等要求。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(255, 255, 255); background-color: rgb(56, 56, 56); font-size: 18px;\">Cook总结：“形式化推理与生成式AI结合，是构建可信Agent的游戏规则改变者。”\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002F4b8bd16ab4224c08a327b8b062493aac\u002F图片5.png\" width=\"473\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami表示企业需要“值得信赖的Agent”，而非仅仅“能做事的Agent”。今天Agent可能第一次能完成任务，但重复执行时却会失败。问题根源在于传统机器人流程自动化（RPA）可靠但缺乏弹性，LLM具备弹性但协调复杂，需要构建错误处理和回溯机制，导致LLM在失败路径上运行很远才意识到错误。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">AWS给出的破灭方案，是将端到端集成与强化学习深度融合。今日正式推出的Amazon Nova Act，专为构建和管理自动化生产UI工作流程的Agent团队而设，在企业级工作流程场景中已实现90%高可靠度。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami解释Amazon Nova Act独特之处在于紧密集成的组件——模型、协调器、执行器、SDK端到端优化。更关键是端到端训练的理念：不是在“罐子里培养大脑”，而是让大脑和手脚一起训练，出厂就会走路。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami详细解释训练方法的创新。传统模仿学习让Agent观察和模仿专家行为，但Agent永远不会理解行为因果关系。因此AWS转向强化学习，并创建数百个RL“健身房”——复制模拟的真实企业环境，如CRM、HR系统、任务定位器等。在这些“健身房”中，Agent运行数千个工作流程，通过数十万次互动进行试错学习。每次成功完成任务Agent获得奖励，每次失败获得惩罚。通过这种方式，Amazon Nova Act学会可靠地解决真实世界企业用例。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">在RealBench和ScreenSpot等关键基准测试中，Amazon Nova Act表现与业界最佳模型相当或更佳。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002F43d5b145263e4748a563fe8773ac02b3\u002F图片6.png\" width=\"450\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">阐述完四大核心技术支柱后，Swami特别邀请AWS应用AI解决方案高级副总裁Colleen Aubrey登台，展示这些技术如何在实际客户服务场景中发挥作用。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Colleen开场即提出核心观点：“未来数年，“Agent队友”将如同身边同事一样不可或缺。”她强调真正效率并非“更少努力”，而是“新产品、新服务、更佳客户体验和新商业模式”。而实现这一切关键，是让AI成为团队一部分，嵌入到每个工作流程中。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Colleen以Amazon Connect为例阐明。这是一款云计算AI原生全渠道客户服务应用程序，沿用Amazon内部同一技术，在全球范围内已支持数十亿次客户对话。她通过生动现场演示，直观展示人工客服代表与AI Agent无缝协作处理信用卡欺诈核查的完整流程。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">在演示环节，Colleen将自己化身客户。当她发现账户上有多笔可疑交易时，AI Agent首先通过集成Amazon Nova Sonic神经声音的自然语音对话验证身份，锁定实体卡但保留Apple Pay功能。当需要深入调查时，AI Agent将Colleen无缝转接给人工调查员Hector，并自动共享所有上下文资讯，Colleen无需重复任何内容。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">更令人印象深刻的是，Hector的“Agent队友”在短短数分钟内便完成通常需要数小时甚至数天的欺诈验证工作。AI Agent自动分析交易地理模式，当地图上标注出可疑位置，并打开跨其他案例查找模式，迅速确认是一宗ATM监听欺诈。Hector甚至即场创建一个自订Agent，以简单自然语言提示定义其行为，让它持续监控Colleen所有账户，一旦发现可疑活动就发送安全消息。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">最后，AI Agent分析Colleen完整账户和交易历史，主动推荐一个更安全旅行账户，并指出她当晚在MGM Grand的晚餐预订可以享受更佳旅行奖励。整个过程流畅自然，展示人类与AI分析能力如何完美结合。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Colleen宣布，本周Amazon Connect发布8项新Agent AI功能，包括Amazon Nova Sonic声音集成、实时推荐Agent、AI驱动预测洞察以及多模态协作等核心功能，让人机客服团队能真正实现无缝协作。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Colleen总结道：“未来数年，人与Agent结合的团队模式将从根本上改变工作方式。这不仅是把同样事情做得更快，而是释放我们甚至无法想象的能力。”\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami在演讲结尾回归开场主题。他说：“还记得你第一次成功编写程序时的感觉吗？那种成就感、自由感、解锁新世界的兴奋感。”今天，借助AI Agent，全球构建者们每天都在体验这种感觉。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">无论你是在清理海洋、解锁人脑奥秘，还是解决尚未发现的挑战，你都拥有构建自由、从概念到影响的空前速度、解决看似不可能问题的自由。AWS正通过“易于构建、高效、可信、可靠”四大支柱，让每个人都能构建并运行企业级AI Agent。未来并非Agent能做所有事，而是我们能全然依赖它们做事。\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(136, 136, 136);\">【新闻来源】十轮网科技资讯\u003C\u002Fspan>\u003Ca href=\" https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002Fnews\u002Fother\u002Far-AA1RUyM6?ocid=msedgntphdr&cvid=6938c776403b4e2c83ca73876f5afd3d&ei=126 \" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(136, 136, 136);\"> \u003Cu>https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002Fnews\u002Fother\u002Far-AA1RUyM6?ocid=msedgntphdr&cvid=6938c776403b4e2c83ca73876f5afd3d&ei=126\u003C\u002Fu> \u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(136, 136, 136);\">（本网转发此文章，旨在为读者提供更多的信息资讯，所涉内容不构成投资、消费建议。文章事实如有疑问，请与有关方核实，文章观点非本网观点，仅供读者参考。）\u003C\u002Fspan>\u003C\u002Fp>","","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002F11f7d7b18c7a40d09dd9cccdf203a96f\u002Fd3ceb214-e320-4d27-b9ff-62bd5ab11fa0.jpg","https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002Fthumbs\u002F11f7d7b18c7a40d09dd9cccdf203a96f\u002Fd3ceb214-e320-4d27-b9ff-62bd5ab11fa0.jpg",0,1,50,"2025-12-11 16:58",2,false,{"id":17,"name":20,"enName":21},"芯位视野","Xinwei Vision","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3A166f8bf6-7692-4ba3-9e9a-f3c553f5e9a8%3A0.wav?Expires=1765513034&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=bDmqrpG3I44ujIYtvH1bJYcRLhQ%3D",14250086,"166f8bf6-7692-4ba3-9e9a-f3c553f5e9a8","2025-12-11 16:50","AWS Neuro-Symbolic AI Breakthrough Ensures Agent Trustworthiness Nova Act Achieves 90% Reliability","\u003Cp>\u003Cstrong style=\"color: rgb(255, 153, 0); font-size: 18px;\">At the re:Invent 2025 keynote, AWS Vice President of Agent AI Swami Sivasubramanian specially invited AWS Distinguished Scientist and authority in automated reasoning Byron Cook to the stage, explaining how to ensure that AI Agents are truly trustworthy through Neuro-Symbolic AI (Neuro-Symbolic AI) technology.\u003C\u002Fstrong>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002Fed907f99093848b99d1003844aacd9f3\u002F图片2.png\" width=\"485\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cspan style=\"color: rgb(255, 153, 0); font-size: 18px;\"> \u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">AWS Neuro-Symbolic AI Breakthrough Ensures Agent Trustworthiness Nova Act Achieves 90% Reliability.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami admitted that when AWS initially built Agent prototypes, there was a problem where \"the model produced hallucinations during API calls.\" This raised a fundamental question: How can an Agent be truly trustworthy? He pointed out that LLMs may make mistakes in front of complex rules or have logical errors in reasoning, and once the scenario involves critical areas such as financial transactions and customer trust, statistical methods alone are far from sufficient.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002F7680b0e2229644ffb31f426545844317\u002F图片3.png\" width=\"471\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Byron Cook started with a sharp question: \"Would you entrust your credit card to an Agent, just like handing it to a child to go shopping? He might help you buy what you want, but in the end, you might end up with a lot of things you don't need.\"\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Cook elaborated on this breakthrough method of neuro-symbolic AI. He explained that automated reasoning is a discipline that searches for and carefully checks proofs in mathematical logic, similar to the method Euclid used to prove theorems over 2,000 years ago. AWS has been using this technology in its internal systems for over ten years and now applies it to Agent AI.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Cook explained how automated reasoning is applied to AI Agents through three vivid examples:\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Validating Output Results: Automated reasoning tools will verify the LLM's reasoning process step by step, like a strict math teacher. If an error is found, it will be pushed back for re-trial, forming a feedback loop.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Training Data Generation: Like training students with standard answer exercises. Using Lean theorem prover to create an infinite number of \"standard answers,\" allowing AI to learn correct reasoning from the beginning.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002Fc1b9af52c318457399d6783991029db3\u002F图片4.png\" width=\"459\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Constraint Decoding: Like a car's lane-keeping system, which automatically adjusts the steering wheel to bring it back when it deviates. In the AI system, when the model attempts to answer \"What is the capital of France\" and starts outputting the letter \"B\", the automated reasoning system intervenes in real-time, guiding it to the correct letter \"P (Paris)\" like navigation. This real-time correction ensures that the AI output always complies with logical rules.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Cook also emphasized Kiro's specification-driven development: Kiro can analyze applications, identify acceptance criteria, and convert them into specifications, thereby guiding program code generation, test generation, and even proving the correctness of programs. He then detailed the AgentCore Policy released yesterday: simply describe the allowed operations in natural language, such as preventing any Agent from performing update operations on AWS production account resources. The system will automatically convert this into AWS' open-source Cedar policy language two years ago, formalized by the Lean theorem prover, and these policies can be verified for sovereignty, privacy, and security requirements through automated reasoning.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(255, 255, 255); background-color: rgb(56, 56, 56); font-size: 18px;\">Cook summarized, \"Combining formal reasoning with generative AI is a game-changer in building trustworthy agents.\"\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002F4b8bd16ab4224c08a327b8b062493aac\u002F图片5.png\" width=\"473\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami said that enterprises need \"trustworthy agents,\" not just \"agents that can do tasks.\" Today, agents may complete tasks for the first time, but they may fail upon repeated execution. The root cause lies in traditional robotic process automation (RPA), which is reliable but lacks flexibility, while LLMs have flexibility but lack coordination, requiring the construction of error handling and backtracking mechanisms, leading LLMs to run far before realizing errors on failure paths.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">AWS' solution is to deeply integrate end-to-end with reinforcement learning. Amazon Nova Act, officially launched today, is specifically designed to build and manage agent teams for automated production UI workflows, achieving a high reliability of 90% in enterprise-level workflow scenarios.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami explained the unique aspect of Amazon Nova Act, which is tightly integrated components—model, coordinator, executor, SDK end-to-end optimization. More importantly, the concept of end-to-end training: not cultivating a brain in a jar, but training the brain and hands together, so that it can walk right out of the factory.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami explained the innovation in the training method. Traditional imitation learning allows agents to observe and imitate expert behavior, but agents will never understand the causal relationships of the behavior. Therefore, AWS turned to reinforcement learning and created hundreds of RL \"gyms\"—simulated real enterprise environments, such as CRM, HR systems, task locators, etc. In these \"gyms,\" agents run thousands of workflows, learning through tens of thousands of interactions. Each successful task completion earns a reward, while each failure incurs a penalty. Through this method, Amazon Nova Act learns to reliably solve real-world enterprise use cases.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">In key benchmarks such as RealBench and ScreenSpot, Amazon Nova Act performs comparably or better than industry-leading models.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cimg alt=\"undefined\" src=\"https:\u002F\u002Fimage.51xinwei.com\u002F2025\u002F12\u002F43d5b145263e4748a563fe8773ac02b3\u002F图片6.png\" width=\"450\" height=\"undefined\" style=\"display: block; margin: auto;\">\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">After explaining the four core technological pillars, Swami specially invited Colleen Aubrey, Senior Vice President of AWS Applied AI Solutions, to the stage to demonstrate how these technologies work in actual customer service scenarios.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Colleen started with her core idea: \"In the coming years, 'Agent teammates' will be as indispensable as colleagues by our side.\" She emphasized that true efficiency is not about \"doing less,\" but about \"new products, new services, better customer experiences, and new business models.\" The key to achieving all of this is to let AI become part of the team, embedded into every workflow.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Colleen illustrated with Amazon Connect. It is a cloud-native, full-channel customer service application, using the same technology as Amazon's internal operations, supporting billions of customer conversations worldwide. She demonstrated through a vivid live demonstration the entire process of human customer service representatives and AI agents working seamlessly to handle credit card fraud verification.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">During the demonstration, Colleen took on the role of a customer. When she noticed multiple suspicious transactions on her account, the AI agent first verified her identity through natural voice conversation integrated with Amazon Nova Sonic, locking the physical card but keeping Apple Pay functionality. When further investigation was needed, the AI agent seamlessly transferred Colleen to human investigator Hector and automatically shared all context information, so Colleen did not need to repeat anything.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">More impressively, Hector's \"Agent teammate\" completed the fraud verification work that usually takes hours or even days in just a few minutes. The AI agent automatically analyzed transaction geographic patterns, marked suspicious locations on the map, and opened cross-case search patterns, quickly confirming it was an ATM skimming fraud. Hector even created a custom Agent on the spot, defining its behavior with simple natural language prompts, letting it continuously monitor all of Colleen's accounts and send security messages whenever suspicious activity was detected.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Finally, the AI agent analyzed Colleen's complete account and transaction history, actively recommending a more secure travel account and pointing out that her dinner reservation at MGM Grand that night could enjoy better travel rewards. The entire process was smooth and natural, demonstrating how human and AI analytical capabilities can perfectly combine.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Colleen announced that this week, Amazon Connect released eight new Agent AI features, including Amazon Nova Sonic voice integration, real-time recommendation agents, AI-driven predictive insights, and multimodal collaboration, enabling human-machine customer service teams to achieve seamless collaboration.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Colleen concluded, \"In the coming years, the team model combining humans and Agents will fundamentally change the way we work. It is not just doing the same things faster, but unlocking abilities we cannot even imagine.\"\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Swami returned to the opening topic at the end of his speech. He said, \"Do you remember the feeling when you first successfully wrote a program? That sense of achievement, freedom, and excitement of unlocking a new world.\" Today, builders around the world experience this feeling every day with AI Agents.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"font-size: 18px;\">Whether you are cleaning the ocean, unlocking the mysteries of the human brain, or solving challenges yet to be discovered, you have the freedom to build, the unprecedented speed from concept to impact, and the freedom to solve seemingly impossible problems. AWS is making it possible for everyone to build and run enterprise-level AI Agents through the four pillars of \"easy to build, efficient, trustworthy, and reliable.\" The future is not that Agents can do everything, but that we can fully rely on them to do things.\u003C\u002Fspan>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cbr>\u003C\u002Fp>\u003Cp>\u003Cspan style=\"color: rgb(136, 136, 136);\">【News Source】Ten Wheel Network Technology Information\u003C\u002Fspan>\u003Ca href=\" https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002Fnews\u002Fother\u002Far-AA1RUyM6?ocid=msedgntphdr&cvid=6938c776403b4e2c83ca73876f5afd3d&ei=126 \" rel=\"noopener noreferrer\" target=\"_blank\" style=\"color: rgb(136, 136, 136);\"> \u003Cu>https:\u002F\u002Fwww.msn.cn\u002Fzh-cn\u002Fnews\u002Fother\u002Far-AA1RUyM6?ocid=msedgntphdr&cvid=6938c776403b4e2c83ca73876f5afd3d&ei=126\u003C\u002Fu> \u003C\u002Fa>\u003C\u002Fp>\u003Cp class=\"ql-align-justify\">\u003Cspan style=\"color: rgb(136, 136, 136);\">（This article is forwarded by this website to provide readers with more information. The content does not constitute investment or consumption advice. If there are facts in the article that you are unsure about, please verify with the relevant parties. The views in the article are not the views of this website, and are for reference only.）\u003C\u002Fspan>\u003C\u002Fp>","https:\u002F\u002Fxinwei-dev-test.oss-cn-shenzhen.aliyuncs.com\u002Fintelligent\u002Faudio%3Acff1523f-2a24-410f-bcb9-8870b8613022%3A0.wav?Expires=1774838444&OSSAccessKeyId=LTAI5tNvY2RkKjZw4LLWsrPK&Signature=IYg71MzUbu%2BmhBo%2BPWCqYRizA%2F0%3D","cff1523f-2a24-410f-bcb9-8870b8613022",17476336]