World leaders in artificial intelligence (AI), including Montreal researcher Joshua Bengio, issued a new warning this week about the risk that AI poses to the future of humanity. In a remarkably concise effort, their twenty-word missive captured the imagination. The 350 signatories, at the forefront of the development of AI, warn that their creature poses a risk of extinction of the human species.
The list of signatories includes Mr. Bengio and Geoffrey Hinton, portrayed as the godfathers of modern AI, as well as Sam Altman (the man behind ChatGPT at OpenAI), Demis Hassabis (Google DeepMind) and Dario Amodei (Anthropic).
The letter is more of a call to action than an apocalyptic pamphlet. “Mitigation of AI-related extinction risks should be a global priority alongside other society-wide risks, such as pandemics and nuclear wars,” they write. This is the second major outing by AI thinkers, following the call for a six-month moratorium on the deployment of AI tools made last March.
These experts place us in front of a whole paradox. They demand that democratic states take charge of the inherent security risks that they have unwittingly implanted at the heart of their deregulated systems. In the race for AI, they have found the shortest path to functionality with prodigious success. The ethical dimensions and the security of the systems for individuals and democratic institutions have certainly not found their fair weight in the reflection.
Their cry of alarm echoes Robert Oppenheimer, the father of the nuclear bomb, who drew on the sacred texts of Hinduism to portray himself as the “destroyer of worlds” during the hideous formation of the first mushroom cloud, the July 16, 1945. Oppenheimer who will spend part of his career futilely fighting against nuclear proliferation and the arms race. History is not lacking in these cruel lessons. By the grace of his superior intelligence, the human has released a force that he does not control.
Let’s give Joshua Bengio the intelligence and the humility to recognize that there is a public policy problem that researchers cannot solve on their own. In an interview with the BBC, the founder of Mila said he was losing his bearings and personally tested by the direction that AI is taking. In his defense, no one could have foreseen that the genie would come out of the lamp with such velocity. It took Netflix ten years to reach the mythical mark of 100 million users, about four and a half years for Facebook and two and a half years for Instagram. For ChatGPT? It was a matter of two months.
During the recent edition of C2 in Montreal, Joshua Bengio spoke with Yuval Noah Harari, the author of the best-selling Sapiens. An extraordinary popularizer, Mr. Harari used a striking metaphor. The evolution of AI is as if we went from amoeba to tyrannosaurus in a few years in the evolution of species. Humanity is prodigiously adaptable, he said, but it takes a few generations to get there. The first industrial revolutions were measured in centuries, that of AI is a matter of months. And it has nothing comparable with the previous ones. It’s Harari, again, who points out that no technological advance before AI had the power to make autonomous decisions and generate ideas on its own.
It is not the AI as such that is the problem, but the use that will be made of it by ill-intentioned humans. Disinformation, propaganda, the replacement of jobs by the thousands are among the immediate risks. Addressing biases and the discriminatory potential of AI tools is an urgent task that developers need to tackle quickly. Research ethics should precede regulation, because the latter will be uneven and imperfect.
Some countries, including those in the European Union (EU) and Canada (with Bill C-27), are at the forefront of efforts to regulate AI. It is on the side of authoritarian and despotic regimes and ill-intentioned non-state actors that we must fear the worst… And in democracies where merchants of lies invite themselves into the political debate.
In these malevolent hands, AI has the potential to replace democracy with demagoguery. The first line of defense therefore involves strengthening and expanding the democratic square everywhere in the world, which will not be an easy task. The Altmans, Hassabis and Amodei who call for the United States to regulate AI come from a school of thought that has kept the state away from the digital revolution for 30 years in the name of net neutrality, a principle recently confirmed by the American Supreme Court. Thus, framing AI will require a paradigm shift within democracies, threatened in their foundations both by AI and by the philosophy of non-interventionism in the digital economy.