[tibiahurried]

My background is in automation and robotics; I studied system identification: a discipline where you would use mathematical means to identify a dynamic system model by observing input/output.

You treat the system as a black box and estimate a set of parameters that can describe it (e.g., Kalman filter).

I struggle to understand what's the fundamental difference between system identification and ML/AI. Anyone?

You ultimately have a bunch of data and try to estimate/fit a model that can describe a particular behavior.

It all comes down to a big optimization/interpolation problem. Isn't what they call "Learning" just really "estimating" ?

Then the more CPU/memory/storage you have, the more parameters/data you can estimate/process, the more accurate and sophisticated the model can be.

[fasterbynight]

As someone with a similar background, I believe some of the confusion is because there is a lot of overlap. System identification is very similar to supervised learning, however there are other learning "methods" that still fall under the umbrella of ML/AI. For example, unsupervised learning doesn't really have a good controls analog (as far as I know). Reinforcement learning on the other hand is somewhat analogous to model predictive control.

A better way of phrasing your point is that ML/AI is "just" optimization.

[tommiegannert]

> It all comes down to a big optimization/interpolation problem. Isn't what they call "Learning" just really "estimating" ?

Yes.

A mathematician will say "why do you call this 'back propagation', isn't it just matrix multiplication?" Many disciplines have different names for the same process.