AI and sustainability

I picked up a few threads from a conversation in December 2023, where Naveen Balani, Chris McClean and others were talking about where ‘green’ AI could be headed. I noted some great points, a lot of them to do with the application of models in enterprise environments:

• Caching and timeshifting as key tools in the development workflow, to flex how much energy is used depending on its production status
• Prompt strategy - how they’re developed, stored, backed up
• Recognising the environmental harm of AI as a known entity, a known cost, rather than a risk to be mitigated or explored further, therefore measuring the cost from internal AI initiatives as you would any such programme that impacts existing internal CSR or green policies

Chris McClean made a good point that organisations ‘aren’t sitting around waiting for regulations’, that a need to meet their own goals is independent of any national decree.

15 months later, and things feel far less optimistic here. It may be a very glib point, but I get the sense that organisations tend to push in a direction that they feel they’re going to get pushed in if they don’t take the initiative. Are we feeling that necessity now? It’s unlikely that regulation can ever truly be a nimble support or guide - has it ever been? - and the incentives to just get on are as strong as they ever were, but without that strong focus on the environment, is it liable to slip into the bucket of things we really do care about, but jeez, you know we have to keep up and we can’t really spend the time trying to calculate this stuff.