In your water example, perhaps that liter of water used to cool the datacenter is offering software to a hospital that offers life saving treatments. How will you measure the trade-off of a liter of water used one way vs. another?
Likewise, how can we distinguish between a ton of carbon emitted in the datacenter vs. a ton of carbon emitted by an airplane? Again, you might train an AI and emit one ton of carbon and that AI a save a million lives. Contrast that ton of carbon emitted with any number of frivolous airline flights by rich talking heads.
It may sound like I'm being deliberately difficult/obtuse, but this is exactly why regulation is so difficult to do well, especially in such a rapidly innovating space.
Likewise, how can we distinguish between a ton of carbon emitted in the datacenter vs. a ton of carbon emitted by an airplane? Again, you might train an AI and emit one ton of carbon and that AI a save a million lives. Contrast that ton of carbon emitted with any number of frivolous airline flights by rich talking heads.
It may sound like I'm being deliberately difficult/obtuse, but this is exactly why regulation is so difficult to do well, especially in such a rapidly innovating space.