Engineering Responsible AI: foundations for environmentally sustainable AI

You might well have seen news on the National Engineering Policy Centre’s report: ‘Engineering Responsible AI: foundations for environmentally sustainable AI’

The report is an excellent one. There is so much misinformation in this area, so it’s refreshing to see a measured report, focusing on opportunities. The report is pretty readable, other than the dreaded 2 page multicolumn PDF layout. Somewhat ironically the only way I could comfortably read it was to put it into NotebookLM to remove the formatting.

The report proposes five foundational steps to promote environmentally sustainable AI, focusing on expanding environmental reporting, addressing information gaps, setting sustainability requirements for data centres, reconsidering data management practices, and increasing government investment in sustainable AI.

All of these are sensible. We are constantly asked for ways to provide accurate information about AI environmental impact, and the reality is, we can’t. We can estimate, based on what we know about model training, the hardware used, the time a query takes etc, but these are just that – informed estimates. Actual data would be so much better, and hopefully drive both consumer and provider behaviour.

Quite a few of the URLs I’ve seen for the report seem to be wrong. This one works for me.