Neural networks have been around for a much longer time than they have been popular/practical/commercially-viable. It just so happens that they can be accelerat

Neural networks have been around for a much longer time than they have been popular/practical/commercially-viable. It just so happens that they can be accelerated using dedicated floating point computing hardware --something GPUs are very good at. I often think about symbolic AI, and how it relates to Boolean satisfiability. This is an integer problem. We don't seem to have a similar technology to GPUs that would be transferable to the problem domain. If we had that, maybe things would be different. I looked into this a bit, and Microsoft seems to have put some resources into a SAT computing ASIC. To get the same kind of progress in symbolic AI, perhaps we need massively parallel/scalable SAT solving hardware. The gaming industry gave us the initial floating point hardware; maybe the cryptocurrency industry will gift us with analogous integer hardware that could push symbolic AI further.