[lexpar]

It's not really a great analogy.

In machine learning the cost function is static and is defined by a smart person working hard to come up with something sensible. In life, the cost function changes constantly and is developed consciously as well as subconsciously. Indeed, the training examples (previous experiences) will influence the dynamically evolving cost function.

Moving along the gradient in a ML framework means that every refinement of the framework will influence cost non-positively, while in life we often act in ways that decrease utility.

Also, to put it bluntly, life is just a lot more complicated than that.

[joker3]

Or to put it more succinctly, life is path-dependent.