It's kind-of trivial - you can make goal-directed systems pursue any goal you like.

The correct reply is not to argue for impossibility - but implausibility: our machine descendants will probably contine to maximise entropy - like the current biosphere does - and not some negentropic state - like gold atoms or something.

Yes, that leaves things open for a "but there's still a chance: right?" reply. That's OK - after all, there is a small chance.