Tis a trick question. The actual complexity experienced by humans depends on the job they are trying to do with the object. If you are a botanist/arborist driving to work, then clearly it is the apple. If you are a mechanic on your lunch break, then clearly it is the engine.
It's true that abstractions can lead to a lot of trouble when they are failing to work correctly. But when they are good they add a lot of good value. When did you last have to debug a system call, because you couldn't open a file? There are really good, old abstractions that no one thinks about anymore, since they're just really good.
A framework in web development is of course another matter, where complexity can sometimes be staggering. I believe that the lifecycle of React components is really nice. Instead of initialising X things you can think about how a single component will change over time.
There are endless examples of good and bad abstractions, I guess. But at the end of the day, we have to choose abstractions when we need them to reason about what we're doing, despite adding more points of failures.
After all, if an abstraction just enables us to think faster and more clearly about a problem, AND allows us to write less code, that means that less bugs will be in the final product, assuming that the density of bugs is constant. ;)