On Friday, July 15, 2022, at 5:09 AM, stefan.reich.maker.of.eye wrote: > I am also searching for a replacement for neural network based learning. The > bar is high, but on the other hand, neural networks have fundamental > limitations, and I think they will one day be completely abandoned when > something better has been discovered. We don't know what that thing is yet > though. > > The general point I am making is that neural networks are basically just ONE > algorithm. At best they're ONE FAMILY of algorithms. > > If we one day figure out *general* algorithm synthesis, we can move beyond > NN's problems and weaknesses. Neural networks are basically the devil's > shortcut to a AI - a local maximum in intelligence if you will -, but > unfortunately mainstream AI doesn't seem to see this far.
You need neural networks to create AGI. Intelligent machines need a lot of memories / big brains. Intelligence is trying to store and compute the universe's items (particles, humans, worlds) without actually doing so using Brute Force to do so, but instead by combining many memories to get probable answers to unseen future hard problems. Without neural networks, you'll have a very hard time storing such a machine in RAM, including doing the computation in parallel. As a simple example, take this neural network: It stores multiple memories efficiently plus computes them together. Say it stores multiple phrases like 'we walked around' and 'we walked too fast' and 'we read books', then it would look and work like this: > around > walked ROOT > we > too fast > read books ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T9019734479a51fd8-Mb63494daffbbef1e98740104 Delivery options: https://agi.topicbox.com/groups/agi/subscription
