[throwaway894345]

I think this is a general argument that you can make about any risky enterprise (e.g., GMOs, vaccines, etc). There's always the possibility that your testing is inadequate, you can never drive risk to 0, etc. But by starting small and scaling up very gradually (while studying carefully the whole time) you maximize the risk that you will find and mitigate problems early and "cheaply" (i.e., without committing to applying the technique to the entire world).

Anyway, the idea is to have an ace in the hole in case we run out of other options to reduce climate change--at that point, the question is "which is less risky, geo-engineering or runaway climate change?".