[Radim]

Focused human attention is still by far the most powerful optimization device we have.

AI hype notwithstanding, no machine learning comes even close -- it's generally a parameter search in a space that's too pedestrian, leaving the real hard work (defining the problem context, goals / objective function, viable tradeoffs and shortcuts) to the human.

Cute AI demos aside, when reliability comes knocking on the door, you end up looking for ways to simplify or avoid the whole bloody mess.

If you spend the time to understand the problem well enough (as you must with space tech), the number of degrees of freedom aka model parameters shrinks. Then suddenly computing power and large-scale parameter searches don't buy you as much; their trade-offs against increased complexity aren't as appealing.

"With four parameters I can fit an elephant, and with five I can make him wiggle his trunk." -- John von Neumann