Research program for a career in Academic Economics: Non-systems Connectionist Prediction Machines

Economics is all about models that predict things. Explanatory models. Math that describes why people do stuff through money and value.

The big recent insights of the last decade’s Nobel Prize winners came from rethinking one actor in the systems – humans – and the psychological and behavioral realities formerly ignored. Biases. People don’t actually behave rationally. But they do behave in predictable ways. Models have been getting updated.

But here is a huge breakthrough in Computer Science in the last decade. AI is a name for it. But here is what it really is. The CS community finally made Deep Learning work.

In the 1990s philosophers noticed that the brain what designed differently than computers. Hierarchical and top down, computers (“Turing machines”) implemented problem solving mechanisms that literally mimicked philosophers theories of thought (“logic”). Computer chips were basically zillions of tiny “logic gates”.

But brains have no apparent logic gates and looks super messy compared to computers. Non-binary. Not strictly functionally modularized. Neurons connecting all over the place. Neurons changing jobs.

These philosophers were called connectionists. They said the top-down computing approach to thought was totally wrong and the right one, based on how the brain physically works, means we don’t understand intelligence at all, etc. AI will never succeed. And so forth.

Around 2005 some CS folks actually did succeed in building computer systems that resemble the brain. Crazy lattices of interconnected values. Stacks of them. A set of inputs, many layers of lattices with ultra minimal rules for behavior, and then an outputs layer whose goals are also ultra minimal.

Early successes were in vision processing. Computers like this got good at labeling pictures.

Feed in a billion rounds of the game Go, and on the other end the system has better moves than the best human ever.

Inside the Google DeepMind AlphaGo system, there is an indescribable approach to playing go. There is no “list of rules” to hand a human to play better. (Indescribable…? Well not simple.)

Which means there isn’t a classical economic model for playing Go.

So: why not the same for Economics? Down with models.