Jun 22, 2023 · 1h 4m · green-blueprint
How AI is being used in energy, right now
Priya Donti · 23m spoken Savannah Goodman · 20m spoken Stephen Lacey · 13m spoken Eric Brynjolfsson · 1m spoken Anna Radovanovic · 35s spoken Sundar Pichai · 17s spoken Ryan Reynolds · 5s spoken Content Creator · 3s 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 of The Carbon Copy investigates the practical applications and operational challenges of artificial intelligence across power grids and corporate decarbonization, featuring insights from MIT's Dr. Priya Donti and Google's Savannah Goodman.
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 3.3 Guest teaching 3.0 Guest disagreement 0.7 The hosts pushing back 0.9
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 23:24 Donti challenges the AI P-Doom premise
When asked about the statistical likelihood of an AI-induced grid catastrophe, Donti firmly pushes back, asserting that the grid is already under severe threat from climate extremes and poor planning independent of AI.
Hardest push from the hosts ▶ 16:10 Lacey probes black-box opacity risks in critical infrastructureLacey directly challenges the theoretical framing to ask whether uninterpretable deep neural network models represent an immediate, tangible operational danger for real-world grid operations.
Biggest teaching moment ▶ 11:10 Donti educates on edge-case physics versus average ML predictionsDonti explains why traditional machine learning models trained on internet text or images break down on power systems, where high average accuracy is useless if an extremal edge case violates physical line limits and causes a blackout.
The host holds their own ▶ 50:15 Lacey contextualizes data center energy growth with historical PUE dataLacey demonstrates command of the subject by contrasting sensational AI energy forecasts with the historical precedent of the 1990s, where internet energy demand was projected at 10% but stabilized at 2% due to data center PUE improvements.
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 |
|---|---|---|---|---|---|---|
| AI as the New Electricity: Historical Context and Parallels | 0 | 0 | 0 | 0 | Introductory monologue where Stephen Lacey frames the commercial AI wave using historical parallels to electrification from economist Eric Brynjolfsson. As this is a solo host intro with archival clips, host interaction metrics are zeroed. | |
| Priya Donti on Machine Learning Applications and Physical Grid Constraints | 4 | 5 | 1 | 1 | Lacey asks foundational questions about the taxonomy of AI and its application to physical electrical grids. Priya Donti explains why typical pattern-matching algorithms struggle with physical power constraints where average-case accuracy is insufficient to prevent blackouts. | |
| Grid Safety, Algorithmic Bias, and Security Challenges | 5 | 5 | 2 | 2 | Lacey presses on regulatory auditing, black-box deep learning risks, and grid cyber vulnerabilities. Donti reframes regulatory interpretability needs, addresses algorithmic bias in building retrofits due to historic redlining, and gives a grounded take on grid P-Doom. | |
| AI Team Structuring and Acceleration Scenarios in Clean Energy | 4 | 4 | 1 | 1 | Lacey explores corporate organizational structuring and macro scenarios for AI adoption in the energy sector. Donti outlines specific engineering skillsets, highlights cultural frictions between agile software engineers and conservative grid operators, and notes utility capex incentive misalignments. | |
| Archival Perspective: The Genesis of Google's Carbon-Aware Computing | 0 | 0 | 0 | 0 | Short narrative interlude by Lacey detailing the history of Google's carbon-aware computing initiative and research scientist Anna Radovanovic. As a monologue segment, interactive scores are scored at zero. | |
| Savannah Goodman on Google's 24/7 Energy Goal and ML Solutions | 4 | 4 | 0 | 1 | Lacey engages Google's Savannah Goodman on operationalizing 24/7 carbon-free energy. Goodman explains load-shifting mechanics across global data centers and Google's externalized ML initiatives like Grid Intelligence and Tapestry. | |
| Data Standardization, Energy Efficiency, and Industry Partnerships | 6 | 3 | 1 | 1 | Lacey demonstrates domain expertise regarding historical internet electricity demand projections from the 1990s versus actual PUE efficiency data. Goodman expands on Google's TPU hardware efficiencies, hourly data standards (T-EACs), and partner criteria. |
Statements from this episode (13)
Pichai: AI is more profound than fire or electricity
“You know, I've always thought of AI as the most profound technology humanity is working on. More profound than fire or electricity or anything that we have done in the past.”
Brynjolfsson: AI will transform all industries and displace major corporate titans
“And as it ripples through the economy, every industry is going to be transformed. And just as with electricity, some of the great titans went out of business and new companies arose, the same thing is happening today where some of the big successful companies …”
Brynjolfsson: Both optimists and pessimists wrongly assume technology acts without human agency
“Both the optimist and the pessimist make the same mistake, and that is that they think of the world as something that gets decided for us. They think that the technology is going to do something to us. The reality is that these technologies are tools. In fact,…”
Donti: Generative AI Hype Risks Triggering Another AI Winter
“And with technologies like generative AI, there's a lot of, you know, flashiness and sort of jumping on the hype train, but I do sincerely worry that this will lead to an AI winter if a lot of those, you know, if we see a lot of, you know, hype and unscrupulou…”
Donti: Average-Case ML Fails on Grids Without Physical Boundary Guardrails
“The fact that you can get a machine learning algorithm that learns to do something nuanced from data and does it right most of the time doesn't help you in those times when it does something wrong that one time that really blacks out your grid. So this is wher…”
Donti: Conditioned on AI or not, 'P-Doom' is high for power grids
“Conditioned or not conditioned on AI, P-Doom is high for the grid, right? We are dealing with climate change and extreme events. We are dealing with a lack of, you know, planning for these things as we saw in Texas, for example. So I don't know that AI increas…”
Donti: Utilities lack regulated rate of return incentives for software innovation
“Utilities and system operators can receive a regulated rate of return, some percentage of the money back on kind of investments in things like, you know, wires in the ground, but we don't currently see that for software innovation, and we don't currently see t…”
Radovanovic: Google 24/7 Clean Energy Target Requires Data Center Load Shaping
“This is practically impossible if we don't shape the load.”
Radovanovic: Carbon-Aware Computing Required No Fundamental Changes to Google Data Centers
“There is nothing that we fundamentally changed in how we run our data centers.”
Goodman: Google reached 66% carbon-free energy across global data centers in 2021
“And so in twenty-twenty-one, we reached 66% carbon-free energy across our global data centers, and that actually includes five data centers that are operating at or near 90% carbon-free.”
Goodman: Google TPU v4 is twice as energy-efficient as TPU v3
“So just one example TPU version four is actually twice as efficient as TPU version three. So that means, you know, the amount of machine learning compute that we can run is substantially more for less energy.”
Goodman: Google delivers 5x more compute per unit of electricity versus 2018
“And just to add another kind of metric there compared to, Five years ago, we are now delivering the same amount of compute, or sorry, five times the amount of compute for the same amount of electricity.”
Goodman: Immediate 10x moonshots fail for core decarbonization use cases
“I think one of the things we've learned at Google is jumping right to the, like, 10 X solution doesn't always work. We have X for that. That's kind of the moonshot factory, but for a lot of the sort of core use cases, it's really important to define a bite-siz…”