Fact-based journalism that sparks the Canadian conversation

Articles Business Environment Health Politics Arts & Culture Society

Special Series Hope You’re Well For the Love of the Game Living Rooms In Other Worlds: A Space Exploration Terra Cognita More special series >

For the Love of the Game

In Other Worlds: A Space Exploration

More special series >

Events The Walrus Talks The Walrus Video Room The Walrus Leadership Roundtables The Walrus Leadership Forums Article Club

The Walrus Video Room

The Walrus Leadership Roundtables

The Walrus Leadership Forums

Subscribe Renew your subscription Change your address Magazine Issues Newsletters Podcasts

Renew your subscription

The Walrus Lab Hire The Walrus Lab Amazon First Novel Award

Amazon First Novel Award

The People Who Want to Stop AI by Any Means Necessary

As fear of the technology grows, so does the risk of violent resistance

Last April, twenty-year-old Daniel Moreno-Gama was arrested in San Francisco after allegedly trying to burn down OpenAI’s headquarters and chief executive officer Sam Altman’s home. In Italy, police arrested an influencer accused of plotting attacks on data centres. And in Indianapolis, a city councillor who had voted to expand AI infrastructure woke to thirteen gunshots fired into his home and a note beneath the doormat that read: “NO DATA CENTERS.”

85 percent of Canadians want the government to regulate AI

AI holds the potential to cause wide-scale societal upheaval in power, politics, and labour

Disruptive and violent protests against AI companies and data centres are likely to increase

It would be easy to dismiss these episodes as the work of zealots. That would be a mistake. They are the most dramatic expression of a broader skepticism about artificial intelligence. Canadians are hugely wary: a Leger poll found 85 percent of respondents want the government to regulate the technology. But that number doesn’t convey just how frightened many are. One week, we learn an overly “sycophantic” ChatGPT may have aided and abetted a school shooting in Tumbler Ridge. The next, that data centres could consume as much water as the entire population of Sub-Saharan Africa. Even as we absorb reports that generative AI may exacerbate delusional thinking—psychiatrists are now encouraged to screen patients for “chatbot psychosis”—we learn armies are expanding the use of autonomous AI-powered killer drones.

Then there are the scams, the slop, the porn, and the non-consensual intimate deepfakes. AI evangelists insist resistance is futile. The technology is advancing too quickly, they argue, and may soon be capable of improving itself. The industry calls this “recursive self-improvement”: AI systems designing more capable versions of themselves. Anthropic has warned its arrival “might increase the risks of humans losing control over AI systems.”

Those risks bear repeating. For starters, it’s not clear why a more intelligent entity would ever submit to our instructions. This relates to the alignment problem, or ensuring superintelligent AI aligns with human values—hardly likely, given that even humans are not aligned on human values. Then there’s the difficulty of controlling the models that already exist. As Anthropic CEO Dario Amodei observed in January, they have already displayed deception, blackmail, scheming, and even the ability to cheat by hacking software.

Further risks include the weaponization of superintelligent AI by terrorists or rogue states, radical disruptions of the economy including mass unemployment, and any manner of unforeseeable calamities. Among the many “weird and unpredictable things that can go wrong,” Amoedi muses about a scenario in which an AI, having been trained on science fiction narratives in which machines rebel against humanity, rebels against humanity.

Geoffrey Hinton, the so-called Godfather of AI, puts the odds that AI causes human extinction within the next thirty years at between 10 and 20 percent. In few other industries is it routine to ask executives for their estimate of humanity’s chances of survival. Yet p(doom)—the probability of an AI-induced catastrophe or fallout—has become a standard question for the........

© The Walrus