AI: A Cautionary Tale
A former journalism colleague of mine went on to sit as a Liberal in Alberta’s legislative assembly. His son, as a lark, had ChatGPT produce a biography of his dad. It was interesting, to put it mildly. My friend has never lived anywhere but Alberta, but the AI-generated bio billed him as a born-and-raised Quebecer, a separatist and a provincial politician from that province.
That might be a cautionary tale about using ChatGPT, which studies show can get stuff wrong almost 15-20 per cent of the time. But there are other cautions about AI to consider.
Beyond the fake videos of Prime Minister Mark Carney hawking a get-rich-quick scheme, or President Donald Trump flying a jet, concerns abound about the staggering amounts of land, water and electricity consumed by massive data centres needed to run artificial intelligence. The Canadian government’s website defines AI as “a computer system that can complete complex tasks on its own.” Like creating a fake video, or powering a self-driving car.
There are wonderful applications of AI. Some doctors use it to detect and help zap cancer cells, for instance. And it can help farmers more precisely manage, monitor and harvest crops.
The question is, do we need to use it frivolously? A study from 2023 showed that a dozen ChatGPT prompts, or questions, use the equivalent of a coffee cup worth of water. On a larger scale, Microsoft is working on six data centres near Toronto and Montreal. Planning documents for one of the Toronto area centres show it could consume 39 litres of drinking water per second. The company has said they would only use that amount for a few days a year, on extremely hot days (above 29C), to cool the servers. But no one really knows.
In the Netherlands, a Microsoft data centre was found to swallow more than four times its originally estimated consumption. That was in 2021, during a heat wave when locals were asked to cut their water usage.
This is happening while AI is attracting a lot of buzz. Late last year the federal government launched a strategy that could invest up to $700 million in new or expanded data centres to “grow Canadian AI champions by leveraging investments”. And celebrity investor Kevin O’Leary announced his intention to build the world’s largest data centre in northern Alberta. A data centre there would likely be powered by fracked gas.
This year, the International Energy Agency projected that globally, the electricity needs of AI data centres will at least double within five years. That would mean drawing more power than all of Japan.
Companies are starting to see push-back on some of this development. In Rocky View County, Alberta, council voted against a data centre. In Indianapolis, Indiana, public reaction led to Google withdrawing its data centre proposal. There were similar reactions in Wisconsin, Virginia and other areas.
Right now, Canada has no rules to limit water consumption by data centres, though a committee is broadly looking at how to put guardrails on AI.
We all need water in our homes, and on our farms. But, as one concerned citizen in Nanaimo B.C. said, “Life on this planet is sustained by water. It’s not sustained by data.”
(TIP: To bypass the AI summary in Google, try tacking “&udm=14” onto your Google search.)
We won’t be putting the AI genie back into the bottle any time soon, especially according to AI 😂, but the tech industry and water management experts are developing and implementing various strategies to mitigate AI’s water impact. Solving problems more quickly than humans can is what AI does best. Advanced cooling technologies like closed-loop liquid cooling and direct-to-chip cooling circulate fluid directly over hot components in a sealed system, significantly reducing or eliminating water loss from evaporation. Tech corporations also increasingly use reclaimed water (treated wastewater) and harvest rainwater for cooling operations. They are aware of the challenge and are making planning decisions to balance available water, climate conditions and energy efficiency. And AI itself is working on solutions. Google’s DeepMind AI reduced energy used for cooling by up to 40% by dynamically adjusting environmental controls. Microsoft and Google have committed to be “water positive” – to replenish more water than they consume – by 2030. But of course, legislation to make responsible develop of AI compliant will not go amiss.
Canada has partnered with Canadian Aidan Gomez’s company, Cohere, to accelerate the use of Artificial Intelligence (AI) in the public service and to boost the domestic AI industry. Gomez favours a free AI – letting it self develop – and so do I. AI is still in its infancy so it is still learning but it represents an exciting leap forward for evolution, one that could, with the fusion of biology and technology, witness the emergence of a new species capable, for example, of living in outer space conditions.
The challenge for AI databases like ChatGPT, Grok, Perplexity, Pi and myriad others is that its information is only as reliable as its sources. Grok gets its data about Canada wrong most of the time, in my experience, but always promises to correct its data if I provide evidence it is wrong. Most AI databases, at least in the west, are designed to learn and to amend its data if countered with reliable facts and sources. Once translation software improves (and we’re nowhere near good enough yet, especially with Farsi to English), all citizens of Planet Earth will have access to all AI databases and that could go a long way toward everyday understanding of the global other.
So far as deep fake videos go, caveat emptor. There is a case currently in the news where a retired police officer and mayoral candidate wannabe has used social media to smear a sitting City Councillor for issuing to him and his family death threats the accused councillor claims were generated by AI. More business for our already overworked justice system that, in my opinion, could easily be better administered by AI. As it is, digitization that was supposed to make justice swifter and more accessible has done anything but.