Life is y = f(x) and can be optimized via gradient descent

It&#x27;s not really a great analogy.<p>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.<p>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.<p>Also, to put it bluntly, life is just a lot more complicated than that.

It&#x27;s a few trivial points dressed up with remedial ML concepts.<p>The homeless problem is not all about drug abuse and that idea is harmful because it&#x27;s used as an excuse to ignore it.<p>Everyone knows that they need to have life goals. Even addicts know that they are optimizing the wrong thing.

There is a really interesting idea in this book, &quot;The Myth of Meritocracy&quot; that we are now forcing ourselves to invest so much into human capital (and his thesis is that most of that investment is a waste in an education &#x2F; self improvement arms race) that we have to optimize the return of our lives, particularly financially but alos in other quality metrics. I just don&#x27;t think this is a good way to think about your life.

Seems to me that life is fundamentally different from y = f(x). It&#x27;s more like an imperative x = f(x), where the output keeps getting fed back into the input. That inherently produces more interesting behavior, even when f is relatively simple.<p>And the thing that is missing in AI today is the back-and-forth conversion of correlations to causal relationships, which, yes, humans are not that good at, generally don&#x27;t consciously do, but have a built-in mechanism for.

Life is single-objective?

I&#x27;ve thought this as well but there is more. You are constantly pulling from memory and examining that input combined with other inputs from other memories that you might not have had in the past resulting in a new memory and input built entirely from past memory.