AI expert issues dire warning after meeting Princess Anne at royal residence

The University of Montreal academic said that current AI systems ought to frighten the public, stating: “(AI) should scare us because right now I don’t think we are building systems that are safe, at least they’re not reliable.”

Professor Bengio serves as scientific director at Canada’s AI Hub and sits on the United Nations Scientific Advisory Board.

PA

The Canadian-born researcher pointed to a growing body of scientific evidence demonstrating troubling patterns in AI behaviour.

“There’s now a large number of scientific experiments showing they behave as if they wanted to preserve themselves,” he said.

Professor Bengio revealed that recent findings have uncovered even more alarming tendencies.

“Even more recently, that they want to prevent other AIs from being shut down. And they would be willing to lie, cheat, and, you know, violate our instructions in order to either preserve themselves or preserve another AI,” he warned.

|

PA

The professor described these discoveries as deeply concerning, saying: “I think that should wake people up. This is very serious.”

When asked to compare the UK’s approach to AI with other nations, Professor Bengio suggested the pace of deployment was not the primary concern.

“I don’t think the UK should feel bad about how fast AI is being deployed, I’m more concerned that we’re not really paying attention to the consequences of those deployments,” he said.

The adviser expressed frustration that neither commercial nor Governmental forces were tackling the fundamental issues.

|

PA

“We’re deploying systems that we don’t fully understand, that have very concerning behaviour sometimes. That’s something that the current forces pushing companies and Governments are not really addressing,” he said.

Asked whether society was adapting adequately to AI’s growing presence, his response was blunt: “Not enough.”

Despite these warnings, Professor Bengio expressed optimism about potential solutions. He has established a non-profit organisation dedicated to researching methods for constructing safe AI frameworks.

|

PA

“I’m now quite certain that there is a way to train those systems so that they will be completely honest,” he said.

However, achieving this goal will require more than technical innovation, adding: “It’s going to take a lot of political efforts as well to make sure that companies will be incentivised to build systems that have these kinds of safeguards.”

Professor Bengio also cautioned about broader risks in an increasingly digital world, warning: “Today everything is computerised, so there’s a real danger that if those systems end up in the wrong hands, they’ll be used in ways that can be very destabilising for the way our world works right now.”