Superficially examining the Clos Network article, if Factorio is amenable to creating such networks, then it seems like it would be possible to create some of the simpler neural network designs, such as shown here: [1] Maybe having something like a weighting of the resource density arriving at a certain ___location to change the output? From the mechanics I can see here [2] it looks like Mergers / Un-Mergers and three belt speeds could probably do density weighted decision making.
Factorio's "circuits" mechanic could implement neural networks pretty directly since the components can do 32-bit logic and arithmetic (integer only). It'd be pretty straightforward to implement a multi-layered neural network with a ReLU activation function.
Agree. Just based on the previous comments, it sounded like the goals was a "belt only" type of design. Figured there might be interesting functionality with variable belt object density arriving at a ___location being a determinant.
[1] https://www.asimovinstitute.org/wp-content/uploads/2019/04/N...
[2] https://wiki.factorio.com/Belt_transport_system#Splitters