Experts debate how to make AI's power use more efficient

A new episode of the video series Conversations in Action features Vijay Gadepally, a senior scientist at MIT Lincoln Laboratory Supercomputing Center, discussing the physical infrastructure needs of artificial intelligence with hosts John Werner and Alex Wissner-Gross.

The conversation covers AI's rising electricity demands, the concept of measuring intelligence gained per unit of energy, and specialized computing approaches like reversible computing that could reduce power consumption. The group also discusses the global expansion of data centers and whether AI systems could eventually help solve the energy challenges created by their own growth.

Topics include recursive self-improvement in AI systems and efforts at MIT to build more energy-efficient computing methods for future AI development.