The question is: can complex (indivisible?) features evolve? So, they pick a complex feature and see if it just happens. Predictably, it doesn't, because it is only rewarded if it evolves. Then they figure out a bunch of intermediate steps, and reward those, and predictably, the complex feature DOES evolve.
But these are just artificial fitness functions. I think complexity is a natural fitness function. That is, it isn't just rewarded when it is observed, it drives ALL interactions in the world.
Perhaps this is an indictment of the "complexity as a fitness function" mindset. I apologize for what is likely a poor metaphor, but if organisms are drowning in the sea of viability, fitness functions reach out of lifeboats to pull them out. But I think complexity is more fundamental... like buouyancy, which consistently pushes things in a certain direction.
It is possible that a series of mutations each of which confers selective disadvantages might persist if the final mutation in the series confers a selective advantage. Let's call this an OOOOOOX chain. However, the longer the chain is, the less likely it is to occur, and I imagine the probability nears "never in a billion years" in chains of somewhere between 5 and 20 mutations. I imagine there is research on this, but I haven't looked at it. Regardless, this is what I mean when I say fitness functions are reaching out to organisms. With a fitness function, only the final selectively advantageous mutation is rewarded, and this selective advantage must reach down past all of the intverning disadvantaged mutations.
So what I am proposing here is that built in to the laws of chemistry is a push towards polarizing complexity. That is, everything else being equal, some things will tend to become more complex in order to increase the production of entropy. If this is the case, OOOOOOX chains where each step represents an increase in complexity will be more probable than others.