A potential way to solving AI is via whole brain simulation. Currently we have the algorithm to model a human brain (albeit far from perfectly): https://thenextweb.com/news/theres-an-algorithm-to-simulate-our-brains-too-bad-no-computer-can-run-it

It is estimated that we might need 100 petaflops to 1 exaflop of computing power to run a brain simulation in real time. Google's Tensor Processing Unit pods, however, have already achieved 1 exaflop of computing power a while back: https://spectrum.ieee.org/heres-how-googles-tpu-v4-ai-chip-stacked-up-in-training-tests

Since these brain simulations are basically giant spiking neural networks, can they be run on Tensor Processing Units (TPUs), which are specifically designed for neural networks? Since TPU pods can do an exaflop, they might pack enough power to finally run a whole brain simulation?

  • $\begingroup$ Worth noting from the article - the pod that reached this milestone is a distrubued system consisting of 4096 devices. $\endgroup$ Aug 23 at 5:50
  • $\begingroup$ Note that anything you can do with an exaflop you can also do with a teraflop, but it takes a million times longer. My computer at home has a couple of teraflops. $\endgroup$
    – user253751
    Aug 23 at 15:07
  • $\begingroup$ And probably. Maybe in a smooth mode instead of spiking mode. IMHO the real challenges with brain simulation are probably going to be (1) collecting the data and (2) making sure the simulation matches reality. Turns out brain cells are affected by all kinds of hormones in their local vicinity, not just direct connections like an ANN. $\endgroup$
    – user253751
    Aug 23 at 15:07
  • $\begingroup$ Are you asking about Google's TPUs or TPUs in general? Are you asking about this particular algorithm or brain simulation in general? $\endgroup$
    – user253751
    Aug 24 at 14:55
  • $\begingroup$ @user253751 I'm asking about this particular algorithm; but I thought that all spiking neural networks are more or less similar in general. And, yes I'm asking about Google's TPU system. $\endgroup$ Aug 26 at 5:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.