A frantic race by tech giants to develop ever more powerful artificial intelligence (AI) could have “harmful” effects, leading researcher Yoshua Bengio warned Thursday as scientists marked the start of the Paris global summit on the technology.
The computer scientist and 2018 winner of the prestigious Turing Prize was presenting the first-ever International AI Safety Report, compiled under his supervision, at Paris’s ENS university.
Bengio pointed to AI risks that are already widely acknowledged, such as its use to create fake or misleading content online.
But the University of Montreal professor added that “proof is steadily appearing of additional risks, like biological attacks or cyberattacks”.
In the longer term, he fears a possible “loss of control” by humans over AI systems, potentially motivated by “their own will to survive”.
The emergence last month of the low-cost, high-performance Chinese AI model DeepSeek had “sped up the race, which isn’t good for safety”, Bengio added.
He called for heavier international regulation and more extensive research on AI safety, which for now accounts for “a tiny fraction” of the massive investments being made in the sector.
“Without government intervention, I don’t know how we’re going to get through this,” Bengio said.
Still the world’s leading AI nation for now, the United States has dropped former president Joe Biden’s attempt to impose some rules on development since the inauguration of Donald Trump.
When OpenAI’s ChatGPT burst onto the public scene two years ago, “I felt the urgency of thinking about this question of safety,” Bengio said as he presented the report — designed to match now-familiar documents from the Intergovernmental Panel on Climate Change (IPCC).
“What scares me the most is the possibility that humanity could disappear within 10 years. It’s terrifying. I don’t know why more people don’t realise it,” he said.
Around 100 experts from 30 countries, the United Nations, the European Union and the OECD contributed to the Bengio-led report, published at the end of January.
Work was kicked off at a previous summit on AI safety in Britain in November 2023.
AFP
