Hawaiian Electric CIO Tavares Rethinks Grid AI to Solve Decision Overload

夏威夷电力CIO Tavares重新审视电网AI以化解决策过载 公用事业企业在运营仪表盘上浪费了数百万美元，因为真正的可靠性取决于利用AI将孤立的电网数据转化为面向人工操作员的情境化决策智能。 针对公用事业企业面临数据过载这一普遍误解，Edward Tavares指出，真正的挑战在于如何依据现有数据做出有效决策。他批评了依赖孤立仪表盘并将人工操作员作为“中间件”的做法，强调企业面临的核心是决策智能问题，需要AI为操作员提供切实的情境化决策支持。 Edward Tavares 指出，公用事业公司面临的核心挑战并非数据或 AI 部署不足，而是“决策智能”问题。企业即使投入数十亿美元完成 SCADA、GIS 和 AMI 等数字化建设，操作员面对仪表盘依然缺少明确的下一步行动指引。 (0:04) 迫使操作员在相互割裂的系统间充当“人类中间件”会导致严重的认知过载并增加人为失误风险，这在夏威夷电网等极端复杂场景下尤为明显——当地峰值日的分布式能源（DER）占比高达 80% 至 90%。 (0:33) 决策智能框架按照“数据 → 信息 → 上下文 → 决策智能 → 行动 → 结果”层层递进。AI 的定位是结合上下文生成经严格验证的行动建议以强化人类判断，而非脱离人类掌控直接自动运行电网。 (6:21) 可视化地图和仪表盘只是交互界面，而非运营 AI 的最终产品。未来公用事业的领军者不会取决于部署了多少 AI 工具，而取决于能否借助决策智能在野火防治、抢修调度与风暴响应等核心场景中做出更优决策。 (9:44)

Edward Tavares argues that utilities have solved data collection but now face a 'decision intelligence problem,' where operators suffer cognitive overload acting as manual middleware between siloed dashboards. He contends that AI's primary value lies not in autonomous grid control or consolidated maps, but in generating context-rich, actionable recommendations that bolster human decision-making. Cognitive Overload and 'Human Middleware' (0:33) Utilities have invested up to billions of dollars centralizing grid data, integrating supervisory control and data acquisition (SCADA), geospatial information systems (GIS), advanced metering infrastructure (AMI), weather tracking, and cybersecurity tools into central control rooms. Despite these investments, Tavares observes that operators viewing fragmented operational screens are still left wondering what action to take next. When unexpected grid emergencies occur, operators are forced to function as 'human middleware,' manually synthesizing disparate data streams across isolated dashboards. Tavares warns that relying on humans as integration middleware creates cognitive overload and fosters environments where errors can jeopardize safety and life. This operational friction is intensified by modern grid dynamics; in regions like Hawaii, distributed energy resources (DERs) supply 80% to 90% of grid power on peak days. Because every customer outcome reflects thousands of daily operational choices, Tavares argues that utilities must recognize they are fundamentally managing decisions rather than just managing physical assets or outages. The Sequential Decision Intelligence Framework (6:21) To transition beyond backward-looking displays, Tavares outlines a five-tier Decision Intelligence framework: Data (explaining what happened in the past), Information (organizing data to reveal patterns), Context (explaining why those patterns matter), Decision Intelligence (directing what to do next), and Action and Outcome (executing the decision). Within this model, AI operationalizes IT systems by linking disparate feeds, uncovering cross-system relationships, and diagnosing underlying root causes. Crucially, Tavares cautions against turning grid control over to fully autonomous AI. The framework designates the human operator as the ultimate accountable party, positioning AI strictly as an enabling mechanism to reinforce human judgment and preserve grid reliability, resilience, and safety. Operational AI Beyond Consolidated Dashboards (9:44) Tavares emphasizes that visualizations, maps, and consolidated dashboards are not the end product, but merely the interface through which operators interact with Decision Intelligence. Rather than isolating AI in disconnected pilot projects, utilities require operational AI integrated directly into systems operation centers and power supply operation centers. This operational model fuses nine core utility workflows—including storm and sprawl response, crew dispatch, cyber risk management, DER orchestration, and wildfire analytics—into a real-time framework. Instead of merely superimposing weather, asset, and crew data onto a single screen, operational AI delivers vetted, actionable options directly into operator workflows, accelerating response efficiency and safety. Tavares concludes that customers and utility commissions do not measure success by the number of AI agents deployed or screens monitored, but by practical outcomes such as power restoration speed and operational efficiency. Consequently, industry leadership will belong not to utilities that deploy the highest volume of AI tools, but to those that make the best decisions using AI.

演讲者爱德华·塔瓦雷斯（Edward Tavares）指出，现代电网公用事业面临的核心挑战并非数据采集或算法部署，而是“决策智能（Decision Intelligence）”的缺失；企业必须改变让调度员在孤立系统间充当“人工中间件”的现状，利用融合情境的AI可视化工具支撑高质量决策。 数据过载与调度员的人工中间件困境 (0:33) 公用事业企业在过去数十年投入了数百万甚至数十亿美元建设调度控制室，将SCADA系统、GIS地理信息、物理资产数据、高级量测体系（AMI）、气象监测、野火分析及网络安全工具进行了深度整合。然而，尽管数据采集和系统互联已被攻克，面对众多由AI生成的仪表板，调度员依然频繁面临“下一步该做什么”的困惑。各独立屏幕虽然提供了特定维度的精确可靠数据，却无法呈现全局优先事项，系统仅优化了信息传递，并未优化决策过程。 在电网遭遇突发事件或紧急状态时，调度员被迫扮演不同孤立系统之间的“人工集成中间件”，依靠人工合成多屏信息。塔瓦雷斯强调，这种运作方式不具备可扩展性与弹性，会导致调度员处于极度的认知过载之中，不仅极易引发人为误操作，甚至可能危及生命安全。尤其在夏威夷等分布式能源（DER）渗透率极高的地区（峰值日DER供电占比达80%至90%），日益复杂的电网运行态势使这种传统人工整合模式难以为继。 决策智能框架：从原始数据到实际结果 (6:21) 为了超越单纯的数据罗列，塔瓦雷斯提出了连接原始数据与现实成效的五级结构化决策智能框架：第一级为“数据（Data）”，以回顾视角说明过去发生了什么；第二级为“信息（Information）”，对数据进行梳理以揭示规律并提炼知识；第三级为“情境（Context）”，阐明该信息为何至关重要；第四级为“决策智能（Decision Intelligence）”，在结合情境的基础上指明下一步该采取何种行动；第五级为“行动与成效（Action & Outcome）”，将建议转化为具体的运维操作并产生业务成果。 在此架构中，AI的核心作用在于持续丰富数据、发现跨系统关联、生成内容并解释潜在根本原因。塔瓦雷斯特别告诫，切勿将电网控制权完全移交给自主运行的AI系统，人类操作员必须始终是承担最终责任的主体。AI应当作为强化人类专业判断的使能工具，协助调度人员做出维护系统可靠性、弹性和安全性的最优决策。 超越仪表板：运营型AI与业务流融合 (9:44) 将气象预警、客户报修、设备资产、网络风险、DER状态与抢修班组位置简单叠加到单一地图或屏幕上，并非转型的终局。塔瓦雷斯强调，“可视化不是终点，地图不是最终产品，仪表板也不是产品”，可视化本质上仅是人类与决策智能进行交互的界面。真正的运营型AI必须直接建立在受信任的企业数据之上，为调度员提供明确且具备可操作性的具体方案。 公用事业需要将AI直接嵌入系统操作中心（SOC）和供电操作中心，而非仅仅运行孤立的技术试点。决策智能框架需要横跨并融合九大核心运营工作流，包括野火缓解与分析、外勤班组调度、网络风险管理、分布式能源管理以及风暴灾害应对等。通过在数字化电网系统中叠加智能情境，调度员能够以易于理解的形式吸收信息，进而显著提升电网运营效率、作业安全性和用户满意度。 衡量AI成效：停电恢复与决策质量 (13:33) 塔瓦雷斯主张从实际业务成效而非单纯的技术指标来衡量AI的落地价值。当电网发生停电故障时，电力用户和公共事业委员会关心的从来不是企业部署了多少个AI智能体或调度员查看了多少仪表板，而是电力何时能够恢复，以及企业如何有效提供价值与效率。 AI对公用事业的深远变革不在于替代人类员工，而在于赋能调度员每天更快、更好地做出决策。塔瓦雷斯断言，未来在AI时代占据领导地位的公用事业企业，绝非那些部署了最多AI工具的公司，而是那些能够利用AI做出最优决策的企业。 塔瓦雷斯总结指出，用户的最终供电体验实质上是电网每天数千次微观运营决策累积的结果。该演讲明确了AI赋能决策而非取代人类的底线，但并未就如何在高比例DER并网的技术演进中具体划分算法建议与人工干预的法律责任提出细化方案，这仍是公用事业推进深度自动化时有待厘清的问题。

## Background 该视频由 Ai4 于 2026 年 8 月 18 日发布在 YouTube 平台上，内容为夏威夷电力公司（Hawaiian Electric Company）CIO 兼 CISO Edward Tavares 的演讲。演讲介绍了他如何推动公用事业从单纯的“数据堆砌与仪表盘展示”跨越到“AI 驱动的决策智能平台”，旨在整合电网、气象、网络安全等多源数据，辅助调度员与高管做出更快、更可靠的决策。 ## TL;DR 演讲指出公用事业的根本矛盾在于“数据过载”与“决策瘫痪”，传统独立系统与仪表盘迫使调度员充当了脆弱的“人类中间件”。为此，演讲者提出了从历史数据到行动成效的五层演进框架（数据、信息、情境、决策智能、行动与成效），强调 AI 的真正价值不是生成报告或全自动接管电网，而是通过富化情境为人类提供明确的可执行方案。最终，AI 时代的领军企业将不再取决于部署模型的数量，而在于利用 AI 做出高质量决策的能力。 ## Atmosphere & style 演讲整体呈现出专业、务实且充满洞察力的行业沉浸感。演讲者结合自身 32 年以上的从业经验，以沉稳冷静的语气直击行业痛点，摒弃了盲目追捧 AI 的浮躁心态。整场分享逻辑紧密、层层递进，既有对一线调度人员认知负担的深刻体恤，又具备战略层面的宏大视野，展现出极强的人本关怀与严谨的主导气场。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=4s) 公用事业行业面临的核心痛点从来不是缺乏数据或 AI 技术，而是缺乏能够指导行动的“决策智能”（Decision Intelligence）。 - [00:04](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=4s) 行业在 SCADA、GIS、AMI 等数据聚合与精美仪表盘上耗资巨大，但单一维度的屏幕并未解决调度员“下一步该做什么”的根本困惑。 - [04:00](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=240s) 让调度员在孤立系统间充当“人类中间件”极不具备扩展性与系统韧性，认知过载极易诱发致命的人为操作失误。 - [04:00](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=240s) 公用事业的本质并非单纯管理物理资产或应对停电，而是动态管理围绕电网资产和停电所做出的数以万计的日常运行决策。 - [06:21](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=381s) 数据是记录“发生了什么”的历史后视镜，而决策智能的价值在于注入情境（Context），为操作员明确指出“下一步该做什么”。 - [06:21](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=381s) AI 转化为实际运营价值的关键，在于不断富化情境并推演出具体可靠、经过验证的可执行行动方案（Courses of Action）。 - [08:08](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=488s) 电网绝不能完全脱手交给 AI 自主控制，AI 的核心定位是强化人类判断（Strengthen judgment）而非取代人类调度员。 - [08:08](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=488s) 单纯将气象、资产与网络等数据融合到一个新大屏上只是“又做了一个仪表盘”，真正的运营决策智能必须直接输出具体的行动指引。 - [08:08](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=488s) 公众和监管机构在突发大面积停电时，并不关心企业部署了多少 AI Agent 或仪表盘，核心诉求始终是电力如何以及何时得以恢复。 - [13:56](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=836s) AI 时代的领军公用事业企业，不会是部署 AI 模型数量最多的机构，而是那些能够利用 AI 做出最高质量决策的组织。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=4s) 重新审视核心问题：从数据堆砌到“决策智能” 在长达 32 年以上的能源与电信公用事业从业经历中，行业投入了数百万、数千万甚至数十亿美元，只为了收集数据、聚合数据，并将信息呈现在控制室的屏幕上供调度人员使用。然而，当前关于人工智能（AI）的讨论普遍陷入了误区——大家总在问“如何在我现有的数据上部署 AI？”，但这完全问错了方向。核心问题从来不是公用事业缺乏数据，也不是缺乏 AI 技术，而是缺乏**决策智能（Decision Intelligence）**。 公用事业和多数企业面临的痛点不是“信息不够”，而是“如何做出更好的决策”。现代电网已经实现了高度数字化：从遥控所有设备的 SCADA 系统，到地理信息系统（GIS）与资产数据的图层叠加，再到整合客户服务、高级计量架构（AMI）、气象监控、火灾分析、网络安全防御工具、发电与开关监控以及智能变电站等系统。行业基本解决了数据收集和连接的问题，如今又引入 AI 抓取数据并制作出各种仪表盘。然而，当把这些数据可靠、界面精美的独立屏幕交给一线调度员时，所有人依然在问同一个问题：“下一步我该做什么？” ```mermaid flowchart TD accTitle: 数据整合演进与核心瓶颈 subgraph DataCollection["多源数字化数据接入"] A["SCADA & 智能变电站"] B["GIS 地理与资产数据"] C["AMI & 客户服务系统"] D["气象/火险/网络安全"] end subgraph CurrentState["当前误区：信息过载"] E["传统 AI / 多维仪表盘 Dashboards"] F["人类调度员（充当集成中间件）"] end subgraph Bottleneck["面临问题"] G["认知过载 (Cognitive Overload)"] H["产生人因失误 / 危及安全"] end DataCollection --> E E --> F F --> Bottleneck ``` - 公用事业过去数十年来在数据聚合上耗资巨大，但单一维度的精美屏幕并不能直接告诉调度员“当前最重要的事情是什么”。 - 传统做法优化了“信息的呈现”，却没有优化“决策的制定”。 --- ### [04:00](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=240s) 拒绝让调度员沦为“人类中间件”：重新定义公用事业的本质 在突发复杂事故时，如果各大独立系统之间缺乏有机协同，调度员就不得不充当各系统之间的“集成层”或“人类中间件（Middleware）”。这种模式既不具备扩展性，也不具备系统韧性；尽管运维人员具备相应的专业技能，但这种认知过载（Cognitive Overload）极易诱发人为失误，在关键基础设施场景下，错误的决策甚至可能付出生命的代价。 公用事业的本质并非单纯“管理资产”或“管理停电”，而是**管理围绕资产和停电所做出的决策**。无论风暴响应、网络安全事件、资本规划，还是分布式能源（DER）运营，核心都在于决策。以夏威夷群岛为例，电网复杂度极高，部分时间段内 80% 至 90% 的电力供应依赖于 DER。这绝非单纯管理 DER 设备，而是在动态管理电网运行的每一个决策。客户所体验到的所有结果，都是现代公用事业每天成千上万次运行决策累积出来的产物。 - 行业必须停止单纯优化孤立系统，转而优化**决策框架与决策流**。 - 调度人员不应充当系统数据拼接的“胶水”，而应获得能够辅助判断的决策支撑。 --- ### [06:21](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=381s) 决策演进框架：从后视镜数据到业务成效 为了清晰定义 AI 在业务中的定位，建立一个从底层数据到最终成效的层级模型至关重要。如果只记住一个核心逻辑，那就是数据、信息、情境与决策智能之间的转化链条： | 层次 | 核心功能与定义 | 特性与价值 | |:--- |:--- |:--- | | **数据 (Data)** | 记录“发生了什么” (What happened) | 属于“后视镜”式的历史记录，静态且已被存储。 | | **信息 (Information)** | 整理与归纳发生的事情 | 从数据中提取模式，将离散数据转化为结构化知识。 | | **情境 (Context)** | 解释“为什么这很重要” (Why it matters) | 结合电网与业务环境，注入关联性与重要性评估。 | | **决策智能 (Decision Intelligence)** | 指引“下一步该做什么” (What to do next) | 利用 AI 引擎持续富化并推荐可靠、经过验证的行动方案。 | | **行动与成效 (Action & Outcome)** | 执行明确行动并达成最终结果 | 将智能化建议落地为可衡量的业务价值与电网韧性。 | ```mermaid flowchart LR accTitle: 从数据到成效的决策演进链 subgraph CognitivePipeline["认知与决策层级"] D["数据 (Data)<br/>记录发生的事实"] --> I["信息 (Information)<br/>识别规律与知识"] I --> C["情境 (Context)<br/>解释原因与重要性"] C --> DI["决策智能 (DI)<br/>指引下一步行动"] end DI --> ACT["行动 (Action)"] ACT --> OUT["成效 (Outcome)"] ``` - AI 的真正价值不是生成静态报告，而是不断富化数据与情境，为操作员提供经过推演、结果明确的备选行动方案（Courses of Action）。 - 这正是 AI 转化为实际“运营价值（Operational Value）”的关键切入点。 --- ### [08:08](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=488s) AI 赋能人类判断：构建可落地的运营型决策体系 AI 应当贯穿在现代公用事业的各个环节，用以连接孤立系统、挖掘关联性并解释事件背后的根因。然而，技术绝不能演变为电网的完全自主脱手运行——“我们绝不能把电网完全交给 AI 自主控制”。责任的核心始终在于人类调度员；AI 的定位不是取代人类判断，而是**强化人类判断（Strengthen judgment）**，为一线人员提供足以保障人身安全、提升供电可靠性与韧性所需的高质量决策支持。 单纯将气象、用户、资产、网络安全、DER 及抢修班组等数据融合到一张新的地图或可视化大屏上，并不是真正的决策智能；那只是“又做了一个仪表盘”。可视化只是一种人机交互的呈现形式，真正的运营级决策智能（Operational Decision Intelligence）需要直接生成清晰可行的行动建议。 ```mermaid flowchart TD accTitle: 运营决策智能框架 subgraph OperationalWorkflows["融合 9 项典型运营工作流"] W1["风暴与灾害响应"] W2["抢修班组智能调度"] W3["网络安全风险管控"] W4["DER 运行与消纳"] W5["野火防范预警"] W6["其他核心运营流..."] end subgraph DecisionFabric["情境网络与智能叠加"] DF["情境架构 (Context Framework)"] AI["AI 关联挖掘与推演"] end subgraph OutputAction["最终输出"] H["赋能人类调度员 (Accountable)"] REC["可执行行动方案 (Courses of Action)"] end OperationalWorkflows --> DF DF --> AI AI --> REC REC --> H ``` - 决策智能不应被视作又一个强行塞进企业的孤立软件应用，而应是注入系统运行中心（SOC）或电力调度中心（PSOC）的关键运营能力。 - 涵盖风暴响应、班组调度、网络安全、DER 管理、野火防护等 9 类核心工作流，均可接入该情境框架中，实时揭示实质性影响。 - 当大面积停电发生时，监管机构和公众从不会过问“你们部署了多少个 AI Agent”或“调度员看了多少个仪表盘”，所有人关心的只有最根本的问题：“电力何时恢复？具体怎样恢复？” ### [13:56](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=836s) 从数据提供到决策行动：AI 赋能决策智能的核心逻辑 在公用事业的实际运营中，管理层和利益相关者最关心的核心问题始终是：“你如何实现高效运转？你如何为客户创造实际价值？”（"How are you efficient? How are you bringing value to the customer?"）。然而，一线运维团队面临的痛点往往不是缺少数据，而是被海量数据淹没后不知道下一步该做什么。运维人员常常提出疑问：“你给了我这么多数据，我下一步到底该做什么？”（"What should we do next? You're giving me data, what do I do next?"）。这正是人工智能（AI）能够发挥决定性作用的地方——将孤立的数据转化为可执行的行动指南。 演讲者坚信，AI 将彻底重塑现有的公用事业运营模式，但其核心绝非取代人类员工，而是“帮助人类做出更明智的决策”（"it's not going to replace people. I believe it's going to help people make better decisions"）。在即将到来的 AI 时代，行业的领军企业不会是那些部署了最多 AI 模型或技术的机构，而是那些“能够利用 AI 做出最佳决策的组织”（"The organization that leads in the AI era won't be the ones that are deploying the most AI. They'll be the ones that are making the best decisions using AI."）。这就是决策智能（Decision Intelligence）的本质精髓。 ```mermaid flowchart LR accTitle: AI-Enabled Decision Intelligence Workflow subgraph S1["传统困境"] A["海量孤立数据"] --> B["运维人员迷茫:<br/>下一步该做什么?"] end subgraph S2["决策智能转型"] C["AI 上下文整合"] --> D["辅助人类精准决策"] D --> E["提升效率与客户价值"] end B -.-> C ``` - **数据到行动的鸿沟**：单纯提供监控数据无法直接解决业务问题，一线人员需要明确的“下一步操作”建议。 - **人机协同的核心定位**：AI 的使命是作为决策赋能工具，始终将人类置于关键决策的中心。 - **AI 时代的竞争优势**：企业的真正竞争力不在于部署 AI 工具的数量多寡，而在于利用 AI 做出决策的质量与成效。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=4s) Utilities don't have a data problem—they have a decision problem. — Edward Tavares > [00:04](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=4s) 如何在我现有的数据上部署 AI？ — Edward Tavares > [00:04](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=4s) 下一步我该做什么？ — Edward Tavares > [04:00](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=240s) 管理围绕资产和停电所做出的决策 — Edward Tavares > [08:08](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=488s) 我们绝不能把电网完全交给 AI 自主控制 — Edward Tavares > [08:08](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=488s) 强化人类判断（Strengthen judgment） — Edward Tavares > [08:08](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=488s) 电力何时恢复？具体怎样恢复？ — Edward Tavares > [13:56](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=836s) How are you efficient? How are you bringing value to the customer? — Edward Tavares > [13:56](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=836s) What should we do next? You're giving me data, what do I do next? — Edward Tavares > [13:56](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=836s) it's not going to replace people. I believe it's going to help people make better decisions — Edward Tavares > [13:56](https://www.youtube.com/watch?v=MHj2qIbEZ5U&t=836s) The organization that leads in the AI era won't be the ones that are deploying the most AI. They'll be the ones that are making the best decisions using AI. — Edward Tavares ## Mentioned Edward Tavares, Hawaiian Electric Company (夏威夷电力公司), Ai4, SCADA, GIS (地理信息系统), AMI (高级计量架构), DER (分布式能源), SOC (系统运行中心), PSOC (电力调度中心) ## Video description Edward Tavares - CIO and CISO, Hawaiian Electric Company Utilities don't have a data problem—they have a decision problem. This talk will illustrate how artificial intelligence can transform disconnected operational and enterprise data into trusted Decision Intelligence by providing the context needed to make faster, more informed decisions. I will demonstrate how AI-powered visualization can create a single operational picture of the grid by integrating assets, weather, outages, customer impacts, cybersecurity, and other critical information into an intuitive environment that supports operators and executives alike. Attendees will learn a practical framework for evolving beyond dashboards toward an AI-enabled Decision Intelligence platform that improves reliability, resilience, operational performance, and customer outcomes while keeping humans at the center of every critical decision. Join this session to see how AI becomes an enabler of better decisions—not just another source of information. [Source](https://www.youtube.com/watch?v=MHj2qIbEZ5U)

[0:04] Edward Tavares: Utilities don't have a data problem—they have a decision problem. [0:04] Edward Tavares: 如何在我现有的数据上部署 AI？ [0:04] Edward Tavares: 下一步我该做什么？ [4:00] Edward Tavares: 管理围绕资产和停电所做出的决策 [8:08] Edward Tavares: 我们绝不能把电网完全交给 AI 自主控制 [8:08] Edward Tavares: 强化人类判断（Strengthen judgment） [8:08] Edward Tavares: 电力何时恢复？具体怎样恢复？ [13:56] Edward Tavares: How are you efficient? How are you bringing value to the customer? [13:56] Edward Tavares: What should we do next? You're giving me data, what do I do next? [13:56] Edward Tavares: it's not going to replace people. I believe it's going to help people make better decisions [13:56] Edward Tavares: The organization that leads in the AI era won't be the ones that are deploying the most AI. They'll be the ones that are making the best decisions using AI.

[0:04] Edward Tavares: What do I do next? [9:44] Edward Tavares: visualization isn't the end. The map isn't the product. The dashboard isn't the product. [13:33] Edward Tavares: When the lights go out, nobody asks, 'How many agents did you deploy?' They ask, 'When are you going to restore the power?' [13:33] Edward Tavares: The organization that leads in the AI era won't be the ones that are deploying the most AI. They'll be the ones that are making the best decisions using AI.

[0:04] So, we shorten that a lot to something that's more  digestible. That was a really, really long slide. [0:13] So, from data to better decisions. That's  that's really the point of this talk. [0:19] You know, energy utilities have evolved, right?  We've spent decades collecting information, [0:27] training people to use that information. You think  about it, utilities, both telecom and energy, [0:34] um which I have worked with for over 32 years,  we have spent [snorts] literally millions, [0:41] in many cases tens of millions, in some  cases hundreds of millions, and in a few [0:45] cases billions of dollars to aggregate our  data, to make it available in rooms like this, [0:53] to give it to people that can manage and  operate the grid effectively and efficiently. [1:00] Lately, we've spent years talking about AI.  But I think in those conversations, in all [1:07] of that work that we've done, in the collecting  of that data together, we've missed the point. [1:14] We're asking the wrong question. The question  isn't how do I deploy AI on the data I have? [1:21] The question is, how do I make better  decisions? Think about that for a second. [1:26] Utilities don't have a data problem. Most  businesses don't have a data problem. [1:31] And they don't really have an AI problem. What  they've got is a decision intelligence problem. [1:38] So, today I'd like to share with everyone  why I believe decision intelligence [1:44] is the missing operating model for AI in critical  infrastructure. You know, if you think about [1:51] the way utilities have evolved over the years,  right? We digitized the utility industry. We build [1:58] utilities that now have SCADA connection to just  about everything that you can control remotely. [2:05] We've taken geographic information systems, or  sorry, geospatial information systems, and we've [2:09] overlaid the information about geospatial with the  information that we've collected about our assets. [2:17] And then we've taken those that  asset data, and we've essentially [2:25] There we go. And we've essentially fused it with  information that comes from customer service, [2:32] with information that comes from our advanced  meter infrastructure, from our weather monitoring, [2:37] from our fire analytics, from our cybersecurity  defensive tools, from all the asset management [2:42] monitoring, the generation, the switching, the  intelligent substations, and all of the other [2:47] stuff that we've put in. So, we've solved, for  all intents and purposes, the problem of what [2:53] do we do, how do we collect the data, how do we  connect it together. And now [snorts] we're doing [2:58] AI. We're using AI to basically scrape that  information, to manipulate that information, [3:03] and show us dashboards. Yet, every single operator  is asking me the same question. What do I do next? [3:12] So, you give them the screen, and every  one of those screens is excellent. [3:18] It has It has trustworthy, reliable, secure data  behind it. And each of us tells Each of those [3:26] screens tells the operator something very, very  important about what is happening in the grid. [3:33] But none of them tells the operator what  matters most. See, what we've done is [3:39] we've optimized the information, but we  haven't optimized the decision-making. [3:46] And that's the challenge. How do you present the  operator with something that they can use? Because [3:52] at the end of the day, when bad things happen, the  operator ends up becoming the integration layer. [4:00] So, think about that for a second.  Humans have become middleware. [4:05] Is that where we want to put our grid  operations team, our sysops folks? [4:12] We want them to be middleware? I don't think so.  I don't think that's a good place for them to be. [4:16] I don't think that's scalable. I don't think it's  resilient. I think they have the skill to do it, [4:24] but I think they're we're putting them  in a position to have cognitive overload. [4:29] And that's when bad decisions are made, or  that's when human error comes in, or that's when [4:34] mistakes potentially can cost a life. That's where  decision intelligence begins. So, it begins with [4:44] looking at the data in a whole different light.  You know, we often say in the utility business, [4:51] we manage assets. I don't think we do. I think  we manage decisions about assets. I think [5:00] I hear people say, "Well, I'm a I manage outages.  We're very efficient at managing the outage." [5:05] Are we really managing the outage? No, we're  making decisions based on the outage parameters. [5:11] We're managing the decisions around the outage.  Same thing for storm response. Same thing for a [5:18] cyber incident. Same thing for capital planning.  We're managing the decisions around those lanes [5:24] in the company. You think about DER operations  and how that has exploded in utilities across [5:31] the United States. The complexity that we have in  the Hawaiian Islands is second to none. 80 90% of [5:38] the grid operating off of DER on some days. The  complexity behind that, we're not managing DER, [5:44] we're managing the decisions around the grid  operations on that day, everything begins with a [5:50] decision. Every outcome your customer experiences,  every outcome you as a customer experience starts [5:59] with a decision. It's the accumulated results  of thousands of operational decisions that are [6:05] happening in the modern utility every single day.  So, maybe let's stop trying to optimize systems. [6:13] Let's start optimizing the decision framework.  Optimize the flow. That starts with data. [6:21] This And this is the model I want you to walk  away with today. If you don't like pay attention [6:25] to any slide, please pay attention to this one.  Data tells you what. What happened. It's rear [6:34] rear view mirror kind of data. Or rear view mirror  kind of information. It's past. It's recorded. [6:42] Information organizes what happened. So, this  is where you start to understand patterns. [6:49] You start to get and extract knowledge from  that data. That's the information layer. [6:56] Context explains why that's important. [7:03] Decision intelligence tells you what  you should do next about all of that. [7:09] Then once you know what you  should do next or you're prompted [7:14] as to what to do next, you can act. That action  becomes outcome. It's deceptively simple, [7:23] but I don't think we always look at business  this way. When you think about it, the engine [7:28] is constantly enriching using AI, driving down  into that decision intelligence framework, [7:37] and then providing courses of action. Basically,  telling us what to do next and then giving us [7:45] options on how to do it that are reliable, that  are trustworthy, that are well-vetted that are [7:52] going to give outcomes that are well-defined. And  really this is where AI becomes operational value. [8:01] Frankly, this is where AI becomes  operational. Why does that matter? [8:08] Well, it matters really simply because  we need to transform our enterprises [8:16] to account for this massive influx of IT  systems that are becoming operationalized. [8:25] When people ask me, "Where does AI fit  in the future of the modern utility?" [8:30] You know, I really answer very simply,  "It's everywhere." When you think about it, [8:35] AI is We are using AI and you can be  using AI to connect all of your systems, [8:42] to find relationships, to create content,  to explain why things are happening, [8:47] not just what is happening. Why is it happening?  Because that's when you start making smart [8:53] decisions about the data, the information, the  context that you've collected. It can produce [9:00] outputs that help you predict consequences.  It can recommend options. But you notice [9:09] the whole point here on the far right of this  slide is the operator is the accountable one. [9:15] We're not turning the grid over to AI to  run autonomously. I do not recommend that. [9:21] What I recommend is giving the operator  enough information to make smart decisions [9:28] so they can save lives, so they can  produce reliable, resilient energy, [9:35] and service the customer. AI cannot replace  judgment. It strengthens judgment. So, with AI, [9:44] you take those screens that I showed you  earlier, and you transform it into a new screen. [9:52] No, that's that's not decision intelligence.  A new map of with all the fused data, [9:59] that's not the goal here because that's the same  thing. You just created another dashboard. But, [10:03] how many of you have seen products where AI  has done this? Fused pieces together. Yeah, [10:09] quite a few. That's not Visualization isn't the  end. The map isn't the product. The dashboard [10:16] isn't the product. It's just simply how  humans experience decision intelligence. [10:22] We need something more, right? All this is is  a fusion of weather, of customers, of assets, [10:28] of cyber, of DER, of crews, and their  locations. It's a version of the truth. [10:36] It's one operational view with a little bit of  operational context. Well, what if we add this? [10:44] What if we add the ability to have courses of  action that are well defined on all of that [10:53] reliable, secure data that we're  able to mine in the enterprise? [11:01] See, I'm not interested in talking to you about  an AI pilot. I'm interested in talking to you [11:05] about operational AI, and that's what this is.  AI that operators can trust. AI that they can [11:12] actually depend on. That improves thousands of  decisions that they can make every day. It's [11:19] you you can look at decision intelligence not as  another application, not something that you're [11:24] going to shoehorn into the enterprise, but  an operational capability you bring into your [11:29] systems operation center, or your power supply  operation center or any other operation center, [11:35] even if you're not in the energy industry,  decision intelligence is applicable. So, [11:40] what does it look like? Operational decision  intelligence looks like this. These are nine [11:46] examples of operational workflows that you can  fuse together to create a decision intelligence [11:54] output. We're bringing material impacts of these  workflows to light within the utility and then [12:02] making decisions on each of those operational  areas in real time. So, storm or sprawl response, [12:11] crew dispatch, cyber risk management,  DER, wildfire, all the stuff I was talking [12:16] about earlier, all can be molded into this  decision intelligence framework quite easily. [12:25] Maybe not so easily. So, we know we've digitized  the utility. We're using AI now to connect those [12:36] digital systems. To basically plug them into a  context framework. And then to take that context [12:43] framework and run it through a fabric that is  going to take the context and apply an overlay [12:50] that helps the output become intelligent.  That helps the output become additive [12:57] to the decision-making process. We're giving the  human data in a way that the human can digest. [13:06] An organizational where operational  processes are continually improving [13:11] through the fusion of AI and human expertise  is an organization that is efficient. [13:18] That is reliable. That is trustworthy. That is  resilient. That has high customer service ratings. [13:26] Think about that. That's That's the enabler  that AI can bring to the table. So, you think [13:33] about it for a second. When the lights go out,  nobody asks, "How many agents did you deploy?" [13:44] They ask, "When are you going to restore  the power?" "How's that going to happen?" [13:49] Nobody asks, "How many dashboards do your people  look at?" My commission never asked me that. [13:56] What they want to know is, "How are you efficient?  How are you bringing value to the customer?" [14:04] My operations folks ask me, "What should we do  next? You're giving me data, what do I do next?" [14:10] AI can help with that. I believe AI will transform  utilities as we know them. I believe that [14:18] it's not going to replace people. I believe  it's going to help people make better decisions. [14:24] The organization that leads in the AI era won't  be the ones that are deploying the most AI. [14:32] They'll be the ones that are  making the best decisions using AI. [14:37] So, that's decision intelligence in a nutshell.  I hope you enjoyed that. Thank you for your time.