Here’s what Should be Prioritized as the Main Threat of AI

Most of the em­pha­sis of the ex­is­ten­tial risk of AI is drawn to worker re­plac­ments and even­tual AGI. How­ever, I see much more in the short term a big­ger threat AI is pos­ing: that of GHG emis­sions.

In this pa­per, Strubell & al (2019) out­line the hid­den cost of ma­chine learn­ing (from in­cep­tion to train­ing and fine tun­ing) and found emis­sions for 1 model is about 360 tCo2. That is no in­signifi­cant amount. For com­par­i­son, the en­tire life of a car with fuel­ing is about 70 tCo2.

The AI com­mu­nity is aware of this such as Mila Labs host­ing an on­line tool on their web­site to calcu­late the car­bon in­ten­sity of build­ing and train­ing ML mod­els. As com­pa­nies rush to in­cor­po­rate AI into their busi­nesses, emis­sions could bal­loon if the en­ergy sources of AI aren’t clean.

To me, for the next decade at least, the threat of AI to con­tribute to cli­mate change should be pri­ori­tized over con­cerns of AI gov­er­nance.

Please let me know what your thoughts are on the sub­ject!