Pioneers of AI We have a power grid problem. Can AI fix it?
25 snips
Mar 11, 2026 Priya Donti, MIT assistant professor and co-founder of Climate Change AI, works on applying machine learning to power grids and climate problems. She discusses embedding physics into ML for safe grid optimization. She contrasts task‑specific models with monolithic ones. She covers clean infrastructure needs for AI and practical roles for AI in forecasting, planning, and decision support.
AI Snips
Chapters
Transcript
Episode notes
AI Is Not One Thing
- AI is heterogeneous; the booming large-model data center paradigm is not the only route to climate benefits.
- Donti worries investment in massive data centers overlooks smaller, task-focused ML that actually helps healthcare and grids.
Start With Task Specific Models Not Giant LLMs
- Task-specific smaller models often outperform one-size-fits-all LLM paradigms for infrastructure problems like grid optimization.
- Donti argues start with concrete tasks, share structure across similar tasks, then scale rather than building monolithic general models.
Site Data Centers With Water And Grid Priorities In Mind
- Power data centers should be sited and powered considering local water stress and grid priorities.
- Donti recommends clean grid supply plus alternative cooling and asking whether renewables should prioritize data centers or other critical loads.