Demystifying Artificial Intelligence for a Carbon-Free Energy System January 8, 2024 | By Ann Collier & Mac Keller SEPA’s Ann Collier, Senior Manager of Emerging Technology: “The last time I wrote a blog – in June 2023 – I remember talking with a friend about how she had started using ChatGPT for her writing. She encouraged me that it was saving her time and generating great results. It all sounded quite dazzling. I tried it out and, while impressed, I did not use the generated text. One big reason was that I was writing about the Inflation Reduction Act of 2022, and the tool could not yet access information past 2021. As technologies like these continue to emerge and mature, I have come to wonder: what would make AI a good-sense investment for SEPA’s community of utilities, businesses, and government and regulatory institutions?” To answer Ann’s question, SEPA engaged our members and stakeholders to learn about their uses, barriers, and goals for AI as well as their awareness of the technology. While there is more to come, one thing is already clear: it is best to explore AI adoption in the context of our biggest, true north ambitions. For SEPA’s network, that means reviewing AI’s value and risks in terms of accelerating the transition to an electric power system that is carbon free, as well as safe, reliable, affordable, resilient, and equitable. In this view, AI is no longer the question– it is a potential answer. The work now is to explore where AI might be the right answer and where it might not be. After all, as authors of the 2021 Climate Change & AI report noted, “AI should only be employed in places where it is actually needed and truly impactful” (Climate Change and AI: Recommendations for Government Action). So let’s dig in: Where might AI be a realistic, impactful tool for accelerating carbon reduction work? (Don’t have a lot of time? Click Here for the bottom line) The Basics: What is AI? Here is a simple take offered by SEPA member Bidgely: artificial intelligence is “the capacity of machines and computers to mimic human behavior” (Bidgely, AI in the Utility Industry, 2023). Software programmers develop AI by training computer programs to observe patterns in data and perform actions based on that understanding. Whether the application is classifying a transformer as damaged based on a photo, or autonomously recommending a microgrid optimization as a heat wave rolls in, today’s real-world uses of AI generally relate to automated classification and prediction. The technology components of AI include a computer program (in other words: a model or an algorithm), massive amounts of high-quality digitized data to “train” the program, and a high-powered computer system to complete the calculations. The technical opportunity for AI has widened following years of effort to improve models, the proliferation of digital data (images and text as well as databases), advances in computer power, and the emergence of lower-cost remote (cloud) computing. The human elements of AI include people and processes. People, as the new saying goes, “stay in the loop” by determining what data to use for training an AI tool, deciding how to use the outputs, assessing quality and uncertainty, and re-calibrating models. Data literacy and data ethics are key. As a process, we can use AI to augment human operations to save time and money, perform more complex tasks than we could reasonably complete otherwise, or automate simpler steps of our work to free up resources for creative and collaborative efforts. In the energy system, we have heard four issues emerge as key mediators to AI adoption within the utility and energy industry so far: Access (e.g., utility digitization, sensors on utility assets and consumer technology; who can get the data) Trust (e.g., cybersecurity, consumer protection, grid operator risk tolerance) Accountability (e.g., explainability, legal and regulatory consequences of wrong or misleading results) Innovation management (e.g., IT/OT integration, innovation expenditures, pilot-to-scale, continuous improvement and recalibration) Equity is a vital fifth issue to the broader adoption of AI. This includes work to ensure benefits and risks are spread equally across society. There is opportunity to link the energy sector’s equity efforts with efforts from technology spheres, given the ongoing focus on equity at the federal and state level with initiatives such as Justice40. What Are the Biggest Uses and Opportunities for AI in the Power Sector? The AI adoption opportunity is the widest where there is access to high-quality data on which to train an AI model, a solution that is tightly matched to an impactful and complex problem, and real-world conditions supportive of using credible AI-generated insight. Here are three use case-technology matchups that SEPA is tracking. These illustrate three paths to carbon reduction and how three forms of AI are being used to advance them. The energy transition, and digitization (and AI specifically) are complex processes to begin with– and are changing month by month. We expect these relationships, and the mix of use cases, to continue to evolve for years to come. Carbon-free Generation + Machine Learning: Machine learning makes intelligent decisions based on patterns detected in prior data, with some forms requiring human input to define the input data and problem statement. Machine learning has been used to plan and operate regional energy markets for years. Supply and demand dynamics are evolving to include more intermittent renewable generation and incorporate increased electrification and grid-connected distributed energy resources (DERs). With traditional data and analytics, it can be difficult to know which devices are being used behind the meter, when they are being used, and how much energy they are using. Using machine learning tools, software is now available to detect customer DERs and interpret trends based on meter data. For those charting paths to carbon reduction, this new information provides many benefits: it can be incorporated in demand forecasts, used to refine customer programs, or to more nimbly manage operations. Machine learning involves “…computer systems that are able to learn and adapt without following explicit instructions, by using algorithms and statistical models to analyze and draw inferences from patterns in data” (Generative Artificial Intelligence (AI), Washington University in St. Louis). Resilience + Deep Learning: Deep learning is a subset of machine learning that is trained to interpret multi-level trends, level by level. For example, by processing reams of text, a large language model will independently “learn” that grammar is composed of letters, then words, then sentences, then paragraphs – each layer with rules and a certain style. Deep learning also has significant potential to help with problems of prediction and identification on the grid. Utilities with transmission and distribution infrastructure are testing deep learning tools to interpret pictures to understand how well assets are performing (transformers, power lines, solar panels, wind turbine blades, etc.) and how conditions around them are changing (temperature, vegetation, etc.). AI-supported issue identification frees up resources to focus on the most high-risk cases. One outcome? AI-enabled predictive maintenance has the potential to reduce truck rolls, reduce outage risks, increase resilience, and enable operations staff to focus time and budget more efficiently. Deep learning is a form of machine learning “…that teaches computers to process data in a way that is inspired by the human brain. Deep learning models can recognize complex patterns in pictures, text, sounds, and other data to produce accurate insights and predictions.” (What is Deep Learning?, Amazon Web Services). Workforce Innovation + Generative AI: Knowledge workers, field staff, customer service representatives, and others may benefit from large language models, image generation, and other generative AI tools. Off-the-shelf large language models like ChatGPT use prediction algorithms to generate a narrative based on patterns seen in other digitized documents. Custom products also exist to reference niche resources from a specific domain and provide a cited summary of trends detected within them; examples include regulatory dockets, reports, and energy databases. Generative AI in these domains can help save time, automate simple tasks, and support creativity. Generative AI are “…deep-learning models that can generate high-quality text, images, and other content based on the data they were trained on.” (What is generative AI?, 2023, IBM). What are the Risks of Using AI for the Energy System? AI brings a variety of risks and uncertainties. Thinking about the domain-specific AI we are seeing in the energy industry today and expect in the near future, much of the risk is because AI’s speed and other properties intensify risks we are already tracking. These include physical security of the grid, cybersecurity, or explainability. Other risk elements are novel, including algorithmic discrimination and the intentional or unintentional spread of misinformation into new domains. Realistically, AI could help or harm the energy system and our work to decarbonize it– so risk management is needed to steer it towards benefit. Three risks stand out for their importance in our work to pursue an equitable and decarbonized energy system: Safety, security, and reliability of a digitized utility system Load impacts and indirect emissions of computer systems to train and run AI tools Energy equity, transparency, privacy, and ethics of AI-enabled decision making To build trust and accountability, our industry needs a path to assess these risks and make plans to mitigate them. Early indications suggest that AI risk management in the energy sector will require braiding together current frameworks for electric power system management, governance frameworks from other industries, and from societal discussion about AI generally. Ultimately, we must prioritize transparency, governance, security, and equity for safe AI implementation. Data transparency ensures that individuals understand how patterns within datasets lead to the development of new insights and predictions. Data governance recognizes that organizations that implement AI have a shared responsibility to maintain the accountability needed to mitigate its risks. Data security dictates that customer and utility data is protected particularly with the widespread use of open-source models. Data equity centers the social, economic, racial, and cultural context of data and algorithms and underscores that AI, if used, should be designed inclusively, provided accessibly, and used for the benefit of all people. The White House Blueprint for an AI Bill of Rights is a resource to learn more. Bottom Line AI tools are computer algorithms that mimic human behavior. AI is most valuable when there is access to plentiful digital data and a need for automation or for wringing out insight from complexity. AI advancements come at a time when our energy system is also changing. Finding the most impactful use cases of AI for energy is a moving target. New use cases may emerge as we see parallel changes in digitization, technology, policy, and climate. Real-world applications of AI in energy include machine learning for load forecasting and grid operations, automated analysis of image data for outage and wildfire risk assessment, and generative AI to beneficially support the workforce. Tradeoffs in AI are complex and require us to assess a wide range of benefits against a wide range of risks. The energy industry can look to data transparency, data governance, and data security standards in the utility industry, as well as other high-trust, high-consequence industries and data ethicists for guidance. How can I get involved? Stay tuned for more education and engagement on the roles of digitization, artificial intelligence tools, and advanced analytics to support carbon reduction in the electric power system. We have case studies, landscape intelligence, working groups, and other activities lined up for 2024! If you are grappling with what to make of AI or want to learn what your peers are working on, this is the place for you. We would also love to hear from those already driving change and with the expertise to contribute. All perspectives and skill levels are welcome. Send us a complimentary research request for resources and information (SEPA members only) – email [email protected]. Sign up below for updates on our events, publications, and insights. Share Share on TwitterShare on FacebookShare on LinkedIn About the Authors Ann Collier Senior Manager, Emerging Technology Ann joined SEPA in December 2021. She leads research and education on a range of emerging technologies for carbon reduction. She collaborates across SEPA’s team and member network to build awareness, encourage partnerships, and support business and policy action to accelerate the power system’s transition toward a net zero carbon future. Previously, Ann contributed to SEPA’s Utility Transformation Challenge and member advisory services. She led energy efficiency and electrification market research and program evaluation projects at Opinion Dynamics and performed economic analysis of federal environmental policy at Abt Associates. She holds an M.S. in Resource Economics & Policy from the University of Maine and a B.S. in Biology from Bates College. Based in Massachusetts, Ann spends her free time exploring the great outdoors with her family. Follow Ann LinkedIn Mac Keller Senior Analyst, Research and Industry Strategy Mac joined SEPA in December 2018 after having interned and worked as a research assistant on SEPA’s Utility Market Snapshots. In 2020, he supported the ideation, surveying, and analysis of results from SEPA’s Utility Transformation Challenge in which we surveyed utilities on their progress towards clean and modern. He subsequently co-authored our Utility Transformation Challenge Profile based on those results. In addition to his work on SEPA’s Utility Market Snapshots and Utility Transformation Challenge, Mac is the co-lead of SEPA’s Microgrids Working group, which convenes members for monthly calls to discuss topics related to microgrid deployments and develop deliverables aimed at overcoming barriers to deployments. Mac holds a bachelor’s degree in Economics with a minor in sustainability from the University of Maryland, College Park. In his free time, he enjoys cooking up a storm and playing ultimate frisbee.