Agreed! That’s why I also mentioned AI agents as being computationally bound as well. Any process capable of “observing” or modeling another process has limits and those limits shape the “rules of nature” that it can perceive.

You can see this in action if you train a small neural net for example. If you take a 1-hidden-layer neural network and try to fit it to the movement of the planets around the sun you won’t get the loss below some really high threshold. It can’t accurately learn that behavior. However you can take an LLM and teach it the equations of motion and how to use a calculator and it can predict the movements just fine.