The philosophical and historical context the documentary puts each of these discoveries into is absolutely fascinating, yes. It isn't as fascinating as basically being shown evolution in progress, but then again I'm not a programmer.
But it's just micro-evolution, and personally there's micro-evolution examples that I find a lot more interesting, like dog breeding. (Besides, I've seen all that footage before; it's a middleware ad).
Show me a video where macroevolution happens before my eyes and then that'll be relevant to me. 'Til then, "Whatever still awesome."
Ehh, I thought the clock simulation linked here a while back was a pretty good example:
http://www.youtube.com/watch?v=mcAq9bmCeR0It shows long periods of stability, then jumping to something completely different with a rapid population shift.
By contrast, machine learning when you're teaching something to walk over 10 generations is just refinement--there's no logical jump. In fact, usually they have to start out the gene pool with lots of legs that already look like they're walking--if they start with completely random input, you'll get very different locomotion (like the guy who belly flops across the ground as the most energy-efficient way to travel). Machine learning is generally aiming for something very specific, and uses all sorts of artificial limits in an effort to keep the process short and directed. It contrasts from the clock example which started out with completely random clocks, 98% of which did nothing.