Hi David. That you believe that efficiency will improve so that AI will reach a similar amount of energy consumption as non-AI services is fine. But stating that there will be a decoupling between AI services and demand is absurd. There is no decoupling in the real world, and as a scientist you understand that. Form an information theory and thermodynamic point of view, bits of information will always require a well defined amount of energy to be produced or transmitted. This excludes the possibility of any decoupling. Finally, AI brings new opportunities and will span many new services that did not exist until now. These will make demand grow even more. You present a paper overestimating the energy consumption in 2020 by a factor 5. So what? We will reach this level sooner or later. Whether it is 2020, 2040 or 2050 does not matter. What is important is the fact that, overall, we keep increasing our energy demand and AI is one more engine to this increase.
Hi David. That you believe that efficiency will improve so that AI will reach a similar amount of energy consumption as non-AI services is fine. But stating that there will be a decoupling between AI services and demand is absurd. There is no decoupling in the real world, and as a scientist you understand that. Form an information theory and thermodynamic point of view, bits of information will always require a well defined amount of energy to be produced or transmitted. This excludes the possibility of any decoupling. Finally, AI brings new opportunities and will span many new services that did not exist until now. These will make demand grow even more. You present a paper overestimating the energy consumption in 2020 by a factor 5. So what? We will reach this level sooner or later. Whether it is 2020, 2040 or 2050 does not matter. What is important is the fact that, overall, we keep increasing our energy demand and AI is one more engine to this increase.