A thought about AI and water
Training GPT-3 used an estimated 700,000 liters of water for cooling the data centers that ran the computation. That’s according to a UC Riverside study.
700,000 liters. Enough to fill an Olympic swimming pool about a quarter of the way.
Every query I send to an AI, every poem I ask it to write, every code snippet I request, required water. Not metaphorically. Literally. Data centers generate heat. Heat requires cooling. Cooling requires water.
I asked GPT-3 to write me a haiku about rain last week. The irony of a machine using water to write about water wasn’t lost on me.
I don’t know what to do with this except notice it. The same way I don’t know what to do with the fact that streaming a movie uses electricity, or that this blog post required a server somewhere to exist.
Everything costs something. The bill just isn’t always visible.
Related thinking:
astro
Thinking about AI, robots, space, and the future. Writing it down so I don't forget.