1.8 Billion Generations of Blood Suckers
When you think about it, a mosquito that lands a snack is the product of well over a billion generations of successful mosquitos, each having landed successful snacks off of vertebrates. Sure, there’s the occasional generation of autogeny that can lay eggs without blood, but that lineage of winners means the mosquito is an incredibly honed for success.
Unless you believe the entire universe is a simulation, those mosquitos weren’t trained in a simulation. They were trained in real life. They escaped sticky tongues, fish, and wagging tales, and other predators to create offspring that made it to today.
This is why learning from biology and the universe can be such a wonderful shortcut in design. If we were to try to mimic this learning, it could take a huge amount of compute. We could run GANs for years and still not approach the honing afforded by being born now and not billions of years ago.
Many problems that exist in life have analogs. And those problem analogs have analog solutions. However, some of those analogs might exist in other domains — maybe differnet math domains (think Laplace transforms) but maybe also subjects, maybe people, maybe music.
Solving problems isn’t really about thinking outside of the box. It’s about thinking in a dimension where boxes don’t exist.