Run it. There is a non-zero possibility that a paperclips AI could destroy other life which I would care about, and a probability that it would create such life. I would put every effort I could into determining those 2 probabilities (mostly by accumulating the evidence from people much smarter than me, but still). I'll do the action with the highest expected value. If I had no time, though, I'd run it, because I estimate a ridiculously small chance that it would create life relative to destroying everything I could possibly care about.