Towards “growing software”, with RxInfer and automated free energy minimization

Bert De Vries — the man behind RxInfer, a toolkit for Active Inference modeling — recently gave a wonderful interview to the folks at Machine Learning Street Talk.

I highly recommend the entire interview, but one of the most salient points is when he discusses his goals with the RxInfer toolkit — link below to the part of the discussion about RxInfer.

In essence, Bert’s mission with RxInfer is to provide companies with an off the shelf way to specify models that minimize free energy in their own domain. In this approach, all software engineering essentially becomes free energy minimization — and you go from designing software to “growing software”.

Here’s a partial excerpt of the discussion:

“In the end, what we want to do is make a toolbox where the inference is automated because in non-trivial models you cannot derive it anymore by hand. So we want to have people come up with a generative model. Well that’s half a page of code. That’s not much. And then you say, ‘Hook it up to your sensors and actuators, and you push a button and you say, optimize for speed or optimize for time or for accuracy.’