The explosive demand of AI has pushed electric power systems to their tipping point by outpacing the construction of renewable energy infrastructure. As with many large-scale technology-induced shifts, the current trajectory of (Gen-)AI, with every 1000x increase in its size leading to an improvement of model quality merely by 3 percent, neglects consideration of negative environmental effects alongside its perceived benefits. Limited regulatory oversight further aggravates this crisis.

Going beyond carbon-aware demand management, this seminar will introduce in-network computing performed sustainably, where we will leverage neuromorphic computing principles executed at each edge on a large-scale power systems network. By doing so, we can offload model development and training of AI to be carried out naturally in power systems circuits instead of digital executions that mandate dedicated electric sourcing, cabling and network infrastructure.

This seminar will start with a pedagogic illustration of the fundamentals (in layman terms), key concepts and design theory from the field of computational neuroscience and power systems. Finally, the key theories of in-network computing with its direct and indirect computing-related impact will be discussed in detail.