| HN Mirror

My personal approach has been trying to find a general Markov decision process solution to any feeback loop process I can identify.

I first encountered this as the Scientific method where there is an experimental hypothesis testing loop the defines are epistemology

That was intersectional with computers for me trying to learn path finding algorithms in early video game development that I was working on as a early teenager

I read about John Boyd and the OODA loop Which got me interested in process control systems - defined in a mechanical way While still in high school. This led me to the Air Force and studying economics and trying to apply the concept of a processing loop to solve generalized decision-making

Because at the time economics had just turned into a statistical science and cognitive science was new, my economics study really became behavioral economics and so I began to study how humans actually do the “loop” as experimental data production via actuating our effectors and sensors into a prediction network. So my research started becoming around how do we model infant learning in computing architecture and prediction for human action - Frank Guerin and Ben Goertzel became my loose mentors for this direction

It was at that point that I really started to understand the Markov decision process and I got very deep into starting to learn reinforcement learning in the tradition of Richard Sutton starting around 2008.

If you go deep enough into generalizable reinforcement learning agent structures, you wind up in cybernetics because you have an agent, environment, predictive modeling, and physical systems requirement that is best defined in the way that cybernetics allows us to define them.