Jun 29, 2023 · 52m · green-blueprint
Will poor data hold back the potential of AI on the grid?
Elizabeth Cook · 19m spoken Jess Melanson · 7m spoken Titian Palazzi · 6m spoken Stephen Lacey · 5m spoken Pamela Isom · 5m spoken David Groarke · 2m spoken Joy Buolamwini · 30s spoken Sam Altman · 13s spoken
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Recorded at the Transition AI conference, this episode examines how data quality, ethical governance, and edge computing will determine the success of artificial intelligence in managing the modern clean electric grid.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 3.8 Guest teaching 4.9 Guest disagreement 1.5 The hosts pushing back 1.3
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 26:09 Vendor lock-in critique
Jess Melanson forcefully criticizes sensor vendors who try to upsell utilities their own data or create closed, non-interoperable data silos.
Hardest push from the hosts ▶ 10:04 Challenging regulatory progressStephen Lacey directly challenges Pamela Isom on whether government regulators are lagging significantly behind the industry's technological pace.
Biggest teaching moment ▶ 19:04 Physics and modeling distinction between T and DElizabeth Cook systematically educates the room on why transmission and distribution systems operate on completely different mathematical equations and data models.
The host holds their own ▶ 1:02 Framing dual-track AI ethics risksStephen Lacey establishes strong technical framing early on, contrasting existential risk narratives from Big Tech CEOs with immediate, systemic algorithmic bias in clean energy deployment.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Pamela Isom on Ethical Frameworks and De-Risking AI | 3 | 4 | 1 | 1 | Host Stephen Lacey asks standard open-ended questions about how companies can build ethical AI frameworks and avoid unintended harms. Guest Pamela Isom explains the 'happy path' concept and details structural governance requirements like independent testing, cybersecurity, and ethics boards. | |
| Government AI Regulation and Department of Energy Insights | 4 | 4 | 2 | 3 | Lacey presses on regulatory lag behind rapid AI innovation, nudging Isom on whether the government is falling behind. Isom defends the government's AI risk management framework while gently moderating Lacey's characterization of how far behind US policy is relative to Europe. | |
| Panel Introduction: Defining Operational AI in Utilities | 4 | 4 | 1 | 1 | Moderator David Groarke opens the panel by cataloging industry use cases and asking Elizabeth Cook how Duquesne Light defines AI. Cook explains why she avoids AI buzzwords internally, focusing instead on foundational data literacy, AMI meter voltage data, and cultural change across utility leadership. | |
| Grid Architecture, Forecasting, and Distributed Edge Intelligence | 4 | 6 | 2 | 1 | Titian Palazzi explains that most AI development time is spent data wrangling for price and supply forecasting. Elizabeth Cook educates the panel on separating transmission and distribution, noting that transmission relies on balanced equations from the 1970s while edge distribution requires completely new modeling approaches. | |
| Data Quality, Interoperability, and Multi-Stakeholder Utility Platforms | 4 | 4 | 2 | 1 | Groarke prompts the panel on cross-industry data maturity. Jess Melanson and Titian Palazzi critique closed utility architectures and emphasize the need for cross-stakeholder data sharing among EV manufacturers, solar providers, and cloud data platforms. | |
| Data Maturity: Transmission Standards Versus Distribution Realities | 3 | 7 | 1 | 1 | Cook delivers an extensive, highly technical breakdown of data maturity across AEIC member utilities, contrasting strict federal transmission compliance burdens with unregulated, fragmented distribution GIS and LiDAR modeling. She also highlights customer sensitivities around utilities detecting behind-the-meter assets like EV chargers. | |
| Real-World AI Case Studies Across Utility Operations | 4 | 5 | 1 | 1 | The panelists share real-world use cases: Palazzi details California CCA load forecasting during heatwaves, Melanson explains premise-level load learning algorithms on smart chips, and Cook shares Duquesne Light deployments of dynamic line ratings, secondary topology mapping, and AI storm outage forecasting. | |
| Future Outlook: Rate Cases, Text-to-SQL, and Workforce Evolution | 4 | 5 | 2 | 1 | The panel explores the future outlook, where Melanson critiques traditional utility rate case incentives that favor piecemeal hardware over forward-looking software investments. Palazzi and Cook highlight transformative productivity gains from natural language text-to-SQL interfaces and recruiting data engineering talent into the utility sector. |
Statements from this episode (15)
Isom: US government is slightly behind Europe on AI regulation
“So I think the U.S. Government has done a pretty good job, but we're a little bit behind some of the other countries like Europe.”
Groarke: Grid AI market is worth billions across 50-plus use cases
“It's worth billions. There's fifty-plus use cases depending on how you define a use case. There's hundreds of vendors in this space, from startups to conglomerates, the existing grid giants, Siemens, Schneider Electric, GE, etc.”
Cook: Grid transmission analytics still rely on 1970s computational models
“Transmission is three-phase, balanced power flow that requires a very different set of equations to run those analytics, and we're really tied to, Really computational systems that were built in the seventies by some generator experts that really built out the…”
Melanson: Grid activity is inverting from bulk generation to edge infrastructure
“Historically, the action and the money and the competition was at the bulk part of the system, and so the last mile was sort of the sleepy last mile infrastructure. That's about to invert, where you have all this activity, you have mobile batteries, you have e…”
Melanson: Grid AI requires orders of magnitude more data and compute
“So we need to now take the next step in investing, and that's not 10% more data capture processing power. That is orders of magnitude more data and processing power.”
Palazzi: Five of top 10 US utilities store data on Snowflake
“Five of the top 10 US IOUs store their data on Snowflake, and that's also, that's data from CRMs, but it's also data from AMI meters.”
Cook: Utility mapped grid in GIS, modeled 63% of power flow
“We did a whole LiDAR distribution system and uploaded seven million points, and now we have an asset associated database in GIS, which means we have a whole distribution system modeled in GIS. Our next step is making those models power flow models. We have abo…”
Cook: Only 30% of utility distribution systems are currently fully modeled
“Now I would say 10 years ago, there's probably like 10%. I think we're probably maybe, you know, up to 30% of the systems that are fully built in the distribution.”
Palazzi: Smart meter data cuts CCA load forecast errors over 50%
“And by aggregating that data, as well as a different weather data, we were able to reduce some of the load forecast error by more than 50% compared to other CCAs who didn't have that.”
Cook: Duquesne Light mapped secondary topology across all 600,000 meters
“So he, and with the collaboration of his whole team, successfully rolled out our secondary topology identification of our meter-to-meter transformers across all 600,000 meters.”
Cook: Most electric utilities lack secondary grid topology mapping
“Not many utilities have their secondary modeled. Not many utilities know exactly which customers are connected to which transformer and which transformer is connected To which circuit, and which circuit is connected to which substation.”
Melanson: Selling distributed AI to utilities is harder than executing the technology
“Selling distributed AI to utilities is harder than actually executing the AI because they, and everybody who sells and works with utilities knows that their historic incentives are not pro-innovation, not pro-software, not pro-future-proofing investments, and …”
Melanson: Hardware-defined grid investments are slower and more expensive than software updates
“What you end up doing is stacking a lot of hardware defined limited investments on top of each other. And you end up with a grid that's not Modernizing quick enough. It's expensive because ultimately in the bigger picture, that's a more expensive way to do thi…”
Melanson: Deploying non-AI-ready assets on the power grid will become a non-starter
“Once we do, I think it's going to really be an exponential change, because this will just become the industry standard, and the paradigm will shift, and then everyone will say, well, why would I put an asset out on the grid that's not AI ready and smart? It'll…”