Nature Podcast · Springer Nature Limited

‘Malicious use is already happening’: machine-learning pioneer on making AI safer

·15 min·1 clip

Yoshua Bengio warns that malicious AI use is already here, driven by deepfakes and cyber attacks.

Nature Podcast journalist Benjamin travels to London to interview Yoshua Bengio, who works at the University of Montreal and is considered one of the godfathers of AI, having won the Turing Award alongside Geoffrey Hinton and Yann LeCun. Bengio describes a significant recent shift in his own position: he has largely redirected his research from advancing AI capabilities to working on AI safety, and spends much of his time explaining to governments and the public why the technology he helped create represents a genuine risk to humanity. The interview was occasioned by Bengio's role chairing an international panel of AI advisors that produced the International AI Safety Report, which identified three main risk categories: unintended risk from malfunctions, malicious use by bad actors, and systemic risk including the displacement of livelihoods. Bengio's colleague Davide Castelvecchi asks him which risk is most likely to have short-term impact, and Bengio identifies malicious use as already happening in the form of deepfakes and AI-enhanced cyber attacks, though he describes these as only the first shadows of what AI-enabled malicious use will become. What keeps him awake at night is the longer-term possibility: the extreme malfunction scenario, or what he calls human extinction through AI that is not designed to be safe by default. This has led him to pivot his research toward the question of how to build AI that will not harm humans by design. Bengio criticizes the tendency of the US and China to frame AI safety primarily as a bilateral competition problem, arguing this framing misses threats from third parties including terrorist organizations and the possibility of rogue AI that threatens everyone regardless of nationality. The International AI Safety Report, he says, has begun having real impact on government policy, establishing shared scientific ground for policymakers. The tone is serious and carefully measured, with Bengio speaking as a scientist who has accepted an uncomfortable public responsibility. The audience is anyone trying to understand AI safety concerns from someone with the deepest possible technical credentials.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Nature Podcast