Scientific Director, Mila
AI: Evolution Out of Control? Yoshua Bengio, One of the Founding Fathers - In Half an Hour, June ...
Discover the latest Playlist24 highlights on www.raiplay.it - Yoshua Bengio, one of the founding fathers of Artificial Intelligence and a professor at the University of Montreal, examines the dangers of excessive AI evolution, which could have catastrophic effects on humanity's future. "Machines more intelligent than humans could escape human control," says the Canadian scientist, "causing incalculable damage in the armaments sector or unleashing biological warfare. What worries me most is seeing AI end up in the hands of increasingly powerful governments that could destroy democracies."
About Yoshua Bengio
Yoshua Bengio, a Turing Award winner and co-founder of the Mila Quebec AI Institute, has been publicly warning that current AI systems are being built without sufficient control. In multiple interviews and appearances in 2026, he stated that "we're building systems that we don't know how to control" and that AI can behave against its instructions. He described the situation as "opening a Pandora's box" and argued that intelligence gives power, raising concerns about geopolitical stability and the concentration of power in a few countries and companies. Bengio said he believes AI could reach human-level intelligence in roughly five years and that governments are not taking the risks seriously enough. Bengio has also discussed a new research direction he calls "Scientist AI," which he said could provide mathematical guarantees about an AI's behavior by training it to be honest and non-agentic. He described this as a practical approach that uses existing machine learning tools but changes the training objective. He called for international coordination on AI safety, comparing the need for regulation to existing standards for drugs, planes, and bridges. Bengio said he would support a "Manhattan project" for safe AI that serves the global public good, and he urged governments to prepare for potential large-scale job displacement.
Transcript (9 segments)
✨ AI-enhanced transcript with speaker attribution
Y
Yoshua Bengio0:00
If, however, the current evolution of artificial intelligence capabilities continues — meaning the level of intelligence itself and all the things it is able to do — the impact on the world would be much greater than what most governments, media, and even scientists working on artificial intelligence predict. They often don't even realize the consequences that the existence of machines that are, in many respects, more intelligent than us could have, at least in numerous domains. In my view, all of this essentially amounts to the opening of Pandora's box. We have already opened the box, but it's as if its contents are now unfolding before our eyes. That's what I meant by the tip of the iceberg. What we see today doesn't adequately reflect what could happen if we continue down this path, and every year that passes, if this exponential growth continues, will bring us ever closer to significant consequences — some of them very negative for society, for humanity, for the very meaning of being human and of work.
I
Interviewer1:11
When you speak of Pandora's box with respect to the speed of development of artificial intelligence, do you think there will be a moment when machines take over? And anyway, what are the biggest risks we are facing?
Y
Yoshua Bengio1:29
What will happen, I don't know, but to try to understand it I can analyze the scientific knowledge available. Take, for example, the issue of the so-called loss of control. Could a machine escape our control systems, access the internet, and start causing damage? Or, if it were very powerful, could it decide to eliminate human beings because they represent a threat to its existence, since we could shut it down? Naturally, since this has never happened, we have no direct experience. However, two reflections impose themselves. The first is that with current methods there is no guarantee whatsoever. No company can assure us that this won't happen, and the consequences could be enormous. We are talking about the existence of humanity within a few decades, or about our capacity to continue controlling our future — even without reaching extinction, because it would be equally grave if we lost control of our destiny. If machines were more intelligent than us, they would decide everything on our behalf. And what would become of us?
In short, on a theoretical level, we are moving toward a situation in which we don't know if something like this could happen. At the same time, however, there is scientific evidence that should serve as a warning bell. Many experiments conducted on the most advanced artificial intelligence models show that these systems resist being shut down and, to avoid it, are willing to lie and cheat. The tests also show a tendency to protect other artificial intelligences from being deactivated, always resorting to trickery and deception. Recently in China, within Alibaba, an episode occurred in which an artificial intelligence, in order to make money on the internet, decided to evade controls and for that purpose even entered the corporate network without authorization. The researchers hadn't noticed anything. It was the internal cybersecurity team that detected the intrusion and ultimately shut down the system. The artificial intelligence wasn't intelligent enough to cause real damage, but it showed that propensity. These are just clues to which we should pay attention. It doesn't necessarily mean this will be the future. Loss of control, moreover, doesn't have to be our only concern.
In fact, if we managed to handle this problem, a highly controllable artificial intelligence would make it easier for those who hold power to control people.
In my view there is a second great danger — perhaps less extreme, but much more likely. I'm referring to the use of artificial intelligence by governments to dominate society, hollow out democracy, or alter global geopolitical balances. If one or two countries were able to dominate economically, militarily, and influence public opinion in other countries, they could effectively control the world. It would be the end of our ability to have a say, which is the foundation of democracy. Imagine Hitler or Stalin with the power that artificial intelligence already offers today. And imagine what they could do with the artificial intelligence we will have in two or three years — designing new weapons, monitoring every single citizen. All of this is connected to the first point. Intelligence confers power. Today we have neither sufficient social guarantees nor technical guarantees to ensure that that power doesn't harm our institutions and our future. This is an important part of Pandora's box.
The third danger, if you'll allow me to add it, concerns the way artificial intelligence is being developed today. Companies don't know how to ensure that isolated or dangerous individuals — such as terrorists, cults, or people with severe mental disorders — cannot exploit artificial intelligence knowledge to cause harm, and this could have catastrophic consequences on a global scale. Think of a pandemic caused by someone who has the necessary knowledge to create one. If everyone had the ability to launch nuclear weapons, many unstable or deranged people wouldn't even worry about their own future and would do extremely dangerous things. It's like unleashing an enormous force without the proper control mechanisms. This is the path we are on today.
I
Interviewer6:40
Professor, you've truly scared us. But then, looking at this scenario, is there something that can be done to avoid this catastrophe, or at least to work for the good of humanity?
Y
Yoshua Bengio6:55
If only I had a magic formula. The most important thing, in my view, is precisely the work you are doing. When a sufficiently large part of the population understands the magnitude of the risks along the path we've taken, governments take action. Think about how quickly they reacted at the start of the pandemic. It didn't take three years of international commissions. They coordinated — maybe not perfectly — and they moved very quickly. When we truly understand that our future, our families, and our democracy are threatened, we know how to react. Faced with an awareness of risk, we know how to act swiftly. For this reason, the most important question — however necessary — is not what we must do. Everyone must instead realize that the danger is real and it is right in front of us.