Wasn't up for a while, but its there now.
Cockshott is a terf, and I hate that he is one of the best in the field of economic planning, but this book was interesting. Now you can read it without paying him lol
Wasn't up for a while, but its there now.
Cockshott is a terf, and I hate that he is one of the best in the field of economic planning, but this book was interesting. Now you can read it without paying him lol
Wait, what's the use of machine learning there? I read Towards a New Socialism and from what Cockshott outlined there, you can make an input output table represented as a matrix to encode the amount of each other commodity that goes into any given commodity. Then you figure out the vector of all the final products you need (this part is decided by the planners on a democratic basis), and run Gaussian elimination to solve the matrix and get a vector that shows how much of each commodity to produce. All you'd need is some way to enter the data for all the technical coefficients for each process from the factory, and some way to coordinate the logistics of delivering the materials. I don't really see where ML tech fits in.
ML tech is basically just gaussian elimination. It's dealing with tensors and matrices and could easily be adapted to do exactly what he lays out in TANS. Which is why I think China is spending a lot of money funding research into building ML systems.
The hardest part as always is getting those input weights.
That being said, using a ML model for this would allow you to scale it to run at different speeds. So you could simplify the input models and get a faster result then use that to more efficiently allocate computing resources in the more detailed planning models.
I saw you mentioned Towards a New Socialism elsewhere in the thread, do you remember the section where he talks about using an iterative algorithm to approximate gaussian elimination?
It's also interesting to think of ML as a way of circumventing Gaussian elimination, I guess it makes some sense but I've always thought of it as a minimization algorithm for some loss function and nothing else.
He's mentioned tensor flows and machine learning in his work before.
He also has an implementation called New Harmony written in Julia you can mess with.
Oh yeah, and the ML stuff would never replace raw gaussian elimination. It's just more time efficient if you know what end state you're looking for. It would just be a tool for planners, not a planning system. Kinda like when you have render previews in 3D modeling software to get a general idea of what you need to add to make a scene lol right.