I guess it's tongue-in-cheek. But you made me imagine the scenario for a movie : once humans start going down that rabbit hole, code becomes more and more like nature : no over-arching "design" that can be reasoned about, just a sprawling mess of stochastically created spagetti that has been progressively patched.
>code becomes more and more like nature : no over-arching "design" that can be reasoned about, just a sprawling mess of stochastically created spagetti that has been progressively patched
If one were to "decompile" an existing artificial neural network model, is this basically what it'd look like inside? Or is it too crude of an analogy / a category mistake?
What happens when one all-coder needs to understand or debug another all-coder's code? The definition of "readable" may change, but I imagine there would still be ways of organizing the code that would make it easier or harder. The halting problem would seem to imply that, for any given all-coder, it's possible to obfuscate hard enough to frustrate that all-coder.
Science. The universe has no source code for us to read, so we tinker and investigate, take our best guess, see if it holds up under the test cases we're capable of running, and update that guess whenever we encounter edge cases the last guess can't explain. We've gotten quite far without perfect understanding.
I have a feeling that the distribution of complexity in the laws of the universe is likely to be very, very different from the distribution of complexity in code created by an extremely intelligent being (with machinelike memory) that isn't optimizing for simplicity.
Verifiability. I have no source for this, but on a basic level it makes sense that a clean implementation will be easier to verify over an ad hoc spaghetti code doing the same thing.