As Alberta rushes to attract $100 billion in data centre investment, one ethicist wants us all — even people outside the artificial intelligence industry — to stop and think about what that means.
"This is a moment where we kind of have to look at all infrastructure and take a step back to think about, 'What are we building? Why are we building it? What's driving what?'" said Katrina Ingram, the founder and CEO of consultancy and training organization Ethically Aligned AI. "Everybody seems to be just running full steam ahead."
Ingram is taking ethical AI discussions beyond the boardroom and into the community with a new community of practice focused on responsible AI literacy.
"I am really interested in this idea of building a community around AI literacy for people to have these kinds of conversations at the level of just being a person or being a citizen," she told Taproot. "I'm really passionate about this idea of broadening out that conversation and having these kinds of discussions (outside the workplace)."
As part of her ongoing effort to bring such questions to regular people, Ingram published a satirical blog post that proposes turning the struggling Edmonton City Centre mall into "Edmonton City Data Centre."
"Imagine AI as the digital beating heart of the city," she wrote with tongue in cheek. "Some have proposed that this site be turned into housing. A data centre would be like housing for our digital 'co-workers.' Who knows, maybe once the singularity is unleashed, we all may 'live' in this new 'housing' project as digital versions of ourselves — vastly exceeding our city's density goals."
That's not a real project, but there are data centres in the works across the province. Those proposed for the Edmonton region include three by Beacon AI Centers, a company backed by New York-based Nadia Partners. They would be in Parkland County, Strathcona County, and Leduc County.
It's not that Ingram is against building data centres, but she hopes to inspire a more careful, measured approach. That's why the environmental impact of AI is one of the topics she's gauging interest in as she puts together the community of practice.
Evidence suggests data centres' power needs exceed Alberta's capacity, will increase emissions from burning natural gas to generate more electricity, and will require a lot of water for cooling, though data centres in the comparatively cool Alberta will require less water than, say, Meta's $10 billion project in Louisiana. Residents near a Meta-owned data centre in Georgia have claimed the facility disrupted their access to water, and Google faced criticism over water use in Mesa, AZ, before shifting to air cooling for a forthcoming data centre.
Katrina Ingram, the founder and CEO of Ethically Aligned AI, is sussing out interest in a community of practice around what it means to pursue responsible AI. (Supplied)
The involvement of American tech giants in this space brings forth another ethical wrinkle — that of data sovereignty.
"When I think about the ethical tensions of (data centres), on the one hand, there's all the environmental and infrastructure impacts," Ingram said. "On the other hand, there's this notion of being able to have autonomy and sovereignty as a nation — so it's complicated."
Right now, a data centre operated in Canada by an American corporation like Amazon or Microsoft is subject to the Clarifying Lawful Overseas Use of Data (CLOUD) Act, which allows the United States government to order them to surrender data to law enforcement that is stored in countries that have signed agreements with the U.S., including Canada.
The Canadian Sovereign AI Compute Strategy is part of the federal government's effort to reduce reliance on the U.S. by increasing domestic compute capacity and supporting the Canadian AI ecosystem. The Alberta Machine Intelligence Institute has received almost $19.7 million from that initiative, as well as $9 million to improve the AI skills of energy workers.
The current impetus is to go fast and build big. But Ingram sees a need to go a little slower and perhaps think a little smaller. For example, it's not clear that giant data centres are needed, she said, citing DeepSeek, which was built cheaper than ChatGPT and requires less server capacity.
"Maybe they're not large language models. Maybe they're smaller or mid-sized language models, but they're still pretty good for particular tasks, and we can get away with a lot less infrastructure," she said. "If you're trying to build an everything machine, I suppose it does become this project of needing one big thing."