Because of start up problems, we are using an unproven design. It was a calculated risk. We let the robots run twenty years and we scrap them before they go unstable. We've got it under control.
They're not going unstable. At twenty years, they undergo a neural pruning process, reach full intelligence, and develop interests outside of their primary function.
Might one of those interests be not getting scrapped?
Oh, they've pretty much got that one figured out already.