Founder of VC-backed startup building optical computers for AGI I'm endeavoring to do the most good I can. I co-founded Fathom as a public benefit corporation whose mission is to advance machine intelligence for the benefit of all. I've investigated alternative computing methods like superconducting single quantum flux logic and done hands-on Si and Nb microfabrication. I also co-founded an electron microscopy DNA sequencing company--Halcyon Molecular which aimed to build tools to understand the genetic basis of health and disease. I developed a method for high-volume nano-manipulation of single DNA strands as well as helped grow and lead Halcyon...
In this talk, Michael Andregg discusses the compute requirements for running a human brain emulation. He emphasizes the importance of model building, which is often overlooked in brain emulation. Andregg’s background is primarily in nanotech and optical computing, but he aspires to delve into large-scale supercomputing and computational neuroscience. The main question he addresses is how much compute power is needed to simulate a human brain. Andregg suggests that it may not be as challenging as assumed, especially if only a circuit-level simulation is required. He breaks down the compute requirements into memory, interconnects, and processors. Connectivity is a significant challenge in brain emulation, and the cost of networking gear is currently ten dollars per gigabit per second. Future computational requirements for brain emulation are expected to decrease in the next five to ten years. Andregg also highlights the communication challenges in the brain, which are often overlooked in estimates for brain emulation. Understanding the communication aspect is crucial for developing accurate estimations.