Editorial of October 10, 2024

The Extravagance of AI

In early 2023, it was reported in these pages that a new form of artificial intelligence, in the guise of a newly-developed chatbot, had been introduced by Microsoft’s OpenAI Artificial Intelligence lab in San Francisco and was fast becoming the major go-to for information searches the world over. This chatbot—ChatGPT (Generative Pre-trained Transformer)—provides answers, conversations, comments, and suggestions in viable human-like text, for the most part both grammatically and, we always hope, factually, correct. It, along with its recently conceived competitors at Google, Amazon, and Meta, is now on the leading edge of a fast-growing market, solving problems in a variety of fields from automotive to climate to research and education, with many interested companies and organizations using it for their own benefit and, it follows, their own financial success. Indeed, ChatGPT holds a massive corpus of high-quality text data—in the range of hundreds of billions of words—which allow it to generate coherent responses to a wide range of topics and in a wide range of languages and styles.

But for these generative AI chatbots to find and retain this immense amount of information, they must use a disproportionate amount of energy, requiring supercomputers in data centers of immense proportion that suck up enormous amounts of electricity to crunch the data that “trains” the AI systems. Every question asked of ChatGPT—even those that produce non-answers or trivial facts—uses as much electricity as does one lightbulb shining for 20 minutes. The training process for a single AI model, such as a large language model, can consume thousands of megawatt hours of electricity and emit hundreds of tons of carbon.

Our widespread use of AI is not only burning up energy, it is also treading on other precious environmental resources. The data centers need water—fresh water—to cool their power sources, and they are consuming millions of gallons of it. This spent water does not return to its source. Rather, it evaporates, not to come back to Earth for at least a year, leaving us in the midst of our growing global wildfires and increasing global freshwater scarcity. The tech giants have significantly increased their freshwater needs for cooling data centers because of the increased demand for online services and generative AI products.

And there seems to be no end to the pressure. Just last week, Microsoft announced plans to open the undamaged section of the nuclear power plant at Three Mile Island, near Harrisburg, Pennsylvania, that was not devastated in the nuclear disaster there in 1979, changing its name to Crane Clean Energy Center, to supply Microsoft’s data centers with the astonishing amount of necessary energy. Talen Industry has agreed to sell Amazon electricity from the Susquehanna Steam Electric Station, a nuclear plant near Berwick, Pennsylvania; other inactive nuclear plants, in Michigan and in Iowa, are being considered for reactivation, as are old coal-fired plants. This immediate demand is outstripping the ability of energy producers to build and operate cleaner wind and solar power generators, adding substantially to air pollution and carbon emissions—climate change at its most egregious. Google no longer considers itself to be carbon neutral; Microsoft has given up on its sustainability goals.

As of now, there are no rules, and no regulations, for AI. The European Union’s AI Act, recently passed, which requires high-risk AI systems to report their energy consumption, won’t go into effect until next year; the International Organization for Standardization could issue criteria for sustainable AI later this year. It is now, with no need to ask permission, on all our devices, milking our information for its own. Because of AI’s inherently important benefit to the future, answering questions, proposing new ideas, proffering salient solutions and raising stock value, no company or technology will curtail its oxymoronic and ridiculously expensive charge into our future.

Every question we ask an AI chatbot is routed to a data center, and it is there that massive amounts of energy and freshwater are consumed. So, it’s up to us. Ask yourselves whether your question is essential, and go easy on our challenged resources.