Baranyai: Use of AI should be mindful

Most of this resource consumption, however, is merely obscene

Last weekend, amid record-shattering heat and humidity, The Beaverton published a satirical headline verging on perfection: “Officials announce new AI data centres to calculate why heat dome happening.” As satire goes, it hits all the right notes: pithy, topical and ultimately enraging. It plumbs the absurdity of outsourcing critical thinking, while winking at the evidence under our noses.

“Canada will analyze climate data using advanced AI, which is provided by data centres that raise surrounding temperatures by an average of two degrees Celsius,” the faux report continues. The context, of course, is real. Those two degrees refer to “heat islands” enveloping the hyperscale data centres powering AI.

Monitoring areas within 10 kilometres of these data centres shows localized temperatures rise about two degrees, and as much as nine. At this point, data centres are about as popular with potential host communities as the ICE detention centres proposed by U.S. Immigration, Customs and Enforcement. Some 96 facilities are planned or under construction across Canada, according to researchers at York University.

Protests in Vancouver, Regina and Hamilton reflect growing opposition to their rapid expansion. Manitoba premier Wab Kinew looked at the environmental toll and short-lived economic benefits of a data centre in Ile des Chênes and said: Not for us, thanks. More than 90 per cent of planned capacity is concentrated in Alberta; unfortunately, it also has the most emissions-intensive electrical grid in the country.

Globally, the picture is even worse. According to the International Energy Agency, right now, 30 per cent of the electricity consumed by data centres is fuelled by coal. AI does have tremendous potential for solving big problems, including climate modelling and planning to protect ecosystems.

It’s also environmentally catastrophic in its own right, measured in energy use, land use and water consumption. A fuller picture of the carbon, water and land footprint of AI has been produced by the UN University Institute for Water, Environment and Health (UNU-INWEH). Its report on the environmental cost of artificial intelligence describes an energy footprint that rivals some of the biggest countries in the world.

It also expects that to double in four years. “The cleanest form of AI use is no use,” university director Kaveh Madani summed up bluntly. “When you could avoid using AI, don’t use it.”

In the absence of regulatory leadership by nations rushing headlong toward AI’s economic potential, the United Nations has stepped up to address its repercussions, for good and for ill, which are global in scale. Some of AI’s environmental costs may be a reasonable trade-off for curing disease, or optimizing food distribution to 8.3 billion people. Most of this resource consumption, however, is merely obscene, measured against the ways we actually use AI: to offload homework assignments; create fake content and video slop; seek therapeutic advice it cannot safely deliver; to flirt with something compliant.

To encourage more mindful use of the technology, UN Secretary-General Antonio Gutteres has launched an environmental transparency initiative for AI, calling on tech companies to fully disclose environmental impacts, and commit to powering data centres with renewable energy by 2030. “No more hidden costs,” he appealed during London Climate Action Week. “It is time to come clean.”

The UN also launched a global dialogue on AI governance this week — things like guardrails, accountability and human oversight — in other words, the cautious considerations that preoccupied AI developers before they started racing for market share. Almost certainly, there are areas where AI can accomplish miracles. When it comes to keeping the world from burning, however, the technology is accelerating the very problem we might be counting on it to solve.