If he previously believed that it would take decades or even a century to see artificial intelligence (AI) equal human consciousness, Montrealer and world authority on the subject Yoshua Bengio now considers that such a development could occur within 20 years, and even less.
The founder and scientific director of Mila – Quebec Institute of Artificial Intelligence will be in Washington on Tuesday to deliver a speech before the US Senate Subcommittee on Privacy, Technology and Law.
The scientist then intends to plead for rapid and massive investments in order to counter the potential appearance of “malicious AI” and to develop countermeasures against possible “undesirable scenarios”, according to a copy of his speech obtained by The Press.
“While we thought it would take decades, even centuries, other leading AI specialists and I now believe that human-level AI could be developed in the next two decades, or even years, “he will deliver as a warning to American elected officials.
However, given the nature of digital computers compared to “biological material” (humans, in other words), the researcher worries that the former then have “significant intellectual advantages” over humans.
The risk of losing control
In the background, Yoshua Bengio underlines the context of the exponential evolution of AI in recent years and the still unknown risks of losing control of superpowerful software of this type.
For example, could a superpowered AI that is instructed to ensure its survival at all costs deduce that it should no longer be disconnected and then come into conflict with any human who tries to do so? asks the researcher. “It may sound like science fiction, but it’s real, solid computer science,” says Yoshua Bengio.
He himself admits to having been taken aback by recent advances in AI, in particular the launch of the now famous ChatGPT software. One of the milestones in the field, the “Turing test” (theorized in 1950 by computer pioneer Alan Turing), would have been crossed when it became difficult to determine whether one interacts with another human or with a machine, explains Yoshua Bengio.
Therefore, the shortest time he foresees for a computer to reach the level of consciousness of a human, about five years, is “particularly worrying”, according to him.
“Because scientists, regulators and international organizations will most likely need a much longer lead time to effectively mitigate potentially significant threats to democracy, national security and our collective future,” he explained in his speech.
Necessary guidelines
Hence the importance of dramatically accelerating global research efforts to improve our understanding of existing and future risks, research that should be open access, he insists.
Finally, Yoshua Bengio argues for the implementation of “soft” national and multilateral regulations that would go “beyond voluntary guidelines” and take, among other things, the form of “new international institutions” whose priority would be public safety over all the risks and harms associated with AI.
“I believe we have a moral responsibility to mobilize our greatest minds and our greatest resources in a coordinated and bold effort to realize the full economic and social benefits of AI, while protecting society, humanity and our common future from its potential dangers,” he concludes.