S. A. Shelley: Over the last few weeks, OWOE staff have been asked, “How much power (energy) does AI consume”?” This is a doozy of a question. As we noted in a prior blog, estimates of the energy consumed just by ordinary data centers, those massive server facilities on which Meta, YouTube, X and the like store videos, photos and the musings of the mad, famous or not, already consume about 3% of all electrical power generated.

Now come the AI Data centres. According to the International Energy Agency (IEA), one moderate AI data centre will consume as much power as 100,000 homes. The IEA goes on to forecast that about 945 TWH of energy will be consumed just by data centres by 2030, or about 10% of total electrical energy demand, globally. By 2030, AI Data centres will consume almost as much electrical energy as the nation of Japan. More locally, it gets worse. New data centres to be located in Wyoming will consume more than 5 times the electricity that all Wyoming state residents use: Every time that a Wyoming resident boils a kettle to make one cup of coffee, the AI Data centres will make five cups by equivalent power.

Some of this additional electrical energy will come from renewables, but not all of it. Referring to the IEA studies, the IEA projects that at best only half of all new energy required by AI data centres will come from renewables. The rest of course, will come from carbon based energy, along with the associated greenhouse gas (GHG) emissions. This increase in GHG is already being seen at such large firms as Google, where GHG emissions have already increased by 48% over the last 5 years.

Think of it another way: All the additional electrical energy generated for AI data centres will not go to improving the lighting, heating or cooling of homes, nor to support any industrial activities or transport technologies (EVs).

If AI could be shown, with certainty, that it will transform and greatly improve our lives, then this energy sacrifice is worthwhile.  However, industry, such as airlines, those masters of reducing comfort and service while increasing fees, are already using AI to initiate dynamic pricing. This will not improve on-time performance, nor passenger experience, but it will boost airline revenues. It’s not just airlines, but I leave the reader to ponder other AI applications in which the goal is revenue maximization while not improving the consumer experience. Heaven help us all if governments begin to blindly herald the advantages of AI in running their operations.

There’s another thing about AI that is quite troubling. It’s not always correct. A few days ago, I embarked upon a chat with MS Co-pilot about Canada’s recent political history regarding LNG exports. The conversation follows.

Question: When did Justin Trudeau say that there is no business case for LNG sales from Canada to Germany?

Fig. 1: AI Answers First Question

Follow-up Question:  Mark Carney says that there is a business case. Who is correct, Mark Carney or Justin Trudeau?

Fig. 2A (above) and 2B (below): Co-pilot asserts that Justin Trudeau was correct

Comment: I believe that you are incorrect with respect to the 2022 assessment of uncertain demand. There has been steady, global demand for natural gas for quite a few years.

Fig. 3A (above) and 3B (below): Co-Pilot agrees that there has been steady NG demand and thus contradicts its assertion in Fig. 2B.

Conclusion: AI misinforms, either intentionally or out of gross ignorance. Note how AI summarized that there was demand uncertainty (Fig. 2B), in support of Justin Trudeau’s position, but then immediately agreed that natural demand has been steady for quite some time when I challenged the initial response (Fig. 3A). Then note how quickly AI changed its tune to claim insufficient infrastructure (Fig. 3B) in Germany. OWOE disproved that lack of infrastructure in a previous blog:  it took Germany only 9 months to bring online a new FLNG (Floating Liquid Natural Gas) import facility.

Anything that humans touch introduces bias. The only truly unbiased data set in the universe belongs to the universe (star masses for example). But anything else built by humans will be biased. We need to be more wary of the AI revolution. AI can be useful, but it can also be stultifying and dangerous. We have already seen the unfortunate social impact of smart phones on society, and now we need to be careful of the impact of AI on society, both on energy demand and on social expectations.

Vive l’Alberta Libre

Fermer la ligne 5!