The idea off using evolving, self-replicating robots as exploration devices has been around for a while, (it even has a name, a von Neumann probe). A spacecraft arrives somewhere, uses local resources to make copies of itself, using its own internal blueprints as well as what it's learned along the way, and the daughter probes then take off to continue (hopefully) their original mission. Now what could possibly go wrong with that, eh?

I remember reading somewhere some wag suggesting that the very act of releasing von Neumann robots on an unsuspecting galaxy would be the one crime all sentient species could agree on qualified the perpetrator for immediate and total genocide. Besides, why bother designing and building von Neumanns and taking chances on them working as planned? All living things are von Neumann machines, even us. We already have mastered a von Neumann technology: sex!

However, I do agree some form of artificial intelligence will be necessary for automated interstellar spacecraft. Enterprise probes will need to conduct self-maintenance, to repair unanticipated damage, handle changing circumstances, carry out alternative mission plans, navigate unexpected environments, and to operate for centuries: perhaps after time, accidents, wear and tear and breakdowns had severely compromised their original capabilities.

A truly robust interstellar probe should be able to continue functioning (albeit with reduced efficiency and capability) even after having suffered considerable damage. I also like the idea of fleets of ships, giving us both the capacity for redundancy and specialization of function.

Most important, our probe should be able to make mission-changing decisions if it encounters great dangers or opportunities for its builders. This implies some form of independent decision-making capability.