AI, Data Centres, And How To Build Trust

(My good friend Nathaniel Arfin has a lot of thoughts on AI, public trust, and how to navigate it all, so I’m happy to publish them.)

Dear Minister Solomon,

Canada has a productivity problem, and we are out of time to treat it as abstract. Businesses feel it. Workers feel it in wages that don’t keep up and services that can’t keep pace. When Washington cut non-Americans off frontier U.S. models, it should have horrified us, and more than that it should have clarified everything: our access to the tools that will decide who stays competitive is not guaranteed. That moment is an opportunity to illuminate what’s at stake. AI for All is right. Our economy needs to adopt AI, and we need to build the environment and the social conditions that make that possible. But if Canadians don’t believe this is a net benefit, the strategy stays a slogan, and we will lag behind our peers.

For most people I talk to, AI isn’t a productivity story yet. It’s the slop in their feeds. It’s deepfakes of real people, from a Markham pharmacist running one of the worst sites on the internet to Grok generating sexualized images of public figures. Hell, this site’s usual author is one of the most distrustful people of AI I’ve ever met. He is unpersuaded that any of this is necessary, important, or even worth discussion, in large part because we have not done the work of explaining the actual utility, and the opportunities, of the technology. His views are not a fringe minority, but are shared by a considerable number of Canadians, and until we convince people like him, we will be nowhere.

This is something my home town is grappling with right now. A data centre application on South Service Road - right off the QEW, that, under the Planning Act, goes to staff for site plan approval, not to council. What should have been an unremarkable industrial file is turning into a municipal ballot issue anyway, with mayoral candidates and residents arguing about ground rules council can’t even set. Nearly half of Canadians say we need domestic AI infrastructure for digital sovereignty, but two-thirds would oppose a large data centre a few blocks from their home. People will nod along to the national project until the warehouse shows up on their street.

And it gets tighter from here. Washington is weighing curbs on Chinese open-weight models that American firms have been grabbing because they’re cheaper and nearly as good. Global compute is already constrained. Whether the next chill is Entity List pressure or a harder ban, Canadian businesses will feel it in cost, in what tools they can still reach, and in whether they get to compete only when someone else allows it. You’ve said it yourself: sovereignty isn’t a slogan, it’s compute, cloud, data, energy, connectivity, and talent. Only about one in three Canadian businesses uses AI today, and the strategy wants roughly three in five by 2034. That jump is how we close the productivity gap. Tell people what gets faster, what gets cheaper, what gets built here, before the only story they know is the one that scares them.

You won’t win that argument while government looks careless with the same technology. Ontario’s Auditor General found that only three percent of public servants had finished the province’s AI training, and that thousands were hitting unsecured tools in ways that put Ontarians’ data at risk. That’s bad enough. The harder question is what we do when a system can rewrite parts of the social contract.

When a tool can decide who gets confined how tightly, who gets benefits, or who gets believed, you don’t start by letting it decide. You start by building it as a mirror. Use it first to show us what we already do: our own biases, and the biases baked into the institutions and data we inherited. Name them. Confront them. Only then decide whether, and how, the system gets to act on people’s lives. Ontario’s corrections system uses SAFER for security classification, and it’s now facing a Charter challenge over racial disparities in who lands in harsher conditions. SAFER was treated as a finished product - one that would help Ontario deal with some of the most difficult problems in corrections. It should have been the mirror. Deploy first, discover the bias later, and you permanently alter lives with someone else’s patterns written into the code.

So build trust. Say publicly where AI is being used in government communications and operations, and why. Make training mandatory, not optional. Meet communities where the infrastructure lands with honest accounting on water, power, and noise, so sovereign capacity doesn’t arrive as a warehouse nobody consented to. Communicate your ideals, goals, and values clearly. Tell people what alignment looks like. And push your provincial counterparts to do their part out loud. The levers that decide whether a hyperscaler connects to the grid or clears local environmental rules are entirely in their hands. Here in Ontario we have Bill 40 for connecting high-demand loads like data centres, and the Ontario Energy Board’s Distribution System Code and Transmission System Code protecting the grid and ratepayers. Regions and municipalities can set environmental requirements on water and land use. Those protections only earn trust if people know they exist and are being used. A National AI Literacy Initiative that stops at kits and courses, and never presses provinces to explain their own energy, grid, and environmental safeguards, leaves the hardest part of public buy-in untouched. Trust isn’t a communications campaign after the fact. Canadians will not accept the infrastructure, the adoption, or the productivity gains you’re promising unless they are informed.

Canada can still catalyze investment, bring together leading people in AI and machine learning, and give Canadian companies and labs room to build sovereign capability here: inference, compute, data infrastructure, and the applications that close the productivity gap in the real economy. The government is right to push AI for All, but right now it will not be a success. We need to push harder on trust first. We need to give people a reason to believe AI will benefit them. It’s time to build.

Or we will see the pain of our failures for years to come.