Baranyai: AI may not be worth the environmental cost

Article content

The author Mary Shelley is said to have kept the calcified heart of her dead husband, Percy Bysshe Shelley, tucked away in a drawer, wrapped in his poetry. The arresting image befits a writer whose most famous novel, Frankenstein, revolved around a creature cobbled together with organs and tissue nicked from dead bodies.

Long before the digital age, Shelley pioneered the new literary genre later known as science fiction. Modern sci-fi writers often explore artificial intelligence as a wrathful adversary its human creators can no longer control, or a tragic figure longing for human connection. Yet, of all the cautionary tales about generative AI, Shelley’s 200-year-old creation may be the most apt: stitched together with disreputably sourced parts, and shocked to life with an electric jolt, with no real certainty what might happen next.

Article content

Generative AI has taken the world by storm. Like Dr. Frankenstein’s monster, AI large language models are self-taught. They learn to mimic human language based on vast amounts of data, chiefly scraped from the internet: the good, the bad and the ugly. Whereas Frankenstein’s creation was shunned by society, generative AI has been enthusiastically embraced. ChatGPT became the fastest-growing consumer software app in history.

However, AI technology carries a hidden cost. It consumes a vast and growing amount of energy. Each generation consumes exponentially more power than the last.

The proliferation of data centres requiring more and more electricity are putting power grids at risk. Forbes reports the graphics processing units ordered for just one platform, Nvidia’s Blackwell, could consume as much as 14.4 terawatt hours, or enough energy to power more than 1.3 million households for a year.

This surging demand for power has outpaced clean electricity generation. In May, Microsoft announced its emissions were up nearly one-third since 2020. At Google, emissions have risen 48 per cent in five years, the Financial Times reports.

Article content

This is an environmental travesty. Globally, the energy sector is responsible for three-quarters of greenhouse gas emissions. The Canada Energy Regulator projects Canadian electricity use could more than double by 2050, a trajectory starkly at odds with the federal plan to reach net-zero emissions by the same year.

Roughly two-thirds of Canada’s power generation comes from renewable sources. Most of this is hydroelectric, with a low carbon footprint relative to oil and gas. The impacts of increased demand are worse south of the border, where 60 percent of their electricity is generated from fossil fuels. A whopping 16.2 percent of American power still comes from asthma-spewing coal. To meet the energy demands of the tech sector, both countries may need to expand their nuclear power generation capacity.

Transitioning to less-polluting energy sources has proven a slow ascent up a steep hill, against fierce headwinds from the head-in-the-tar-sands oil sector. Of course, there’s a faster way to reduce emissions: Use less energy.

It’s worth asking how urgently we need apps that can churn out draft reports through widely sourced plagiarism, or create images of celebrities in the style of My Little Pony. We might enjoy using these new tools, but until we can cleanly generate the power they need, they may not be the best use of our energy resources during a climate emergency.

AI has tremendous potential to improve our lives with advances such as faster medical diagnoses. It also has a ruinous capacity to generate fake photos and news stories, which are increasingly difficult to distinguish from reality. Time will tell if we have tapped into untold benefits for humanity, or created a monster, but one thing is already clear: The jolt of electricity sparking this creation isn’t free.

write.robin@baranyai.ca

Article content