Advanced artificial intelligence systems such as OpenAI’s GPT could destabilize democracies unless governments act quickly and “protect the public”, an artificial intelligence pioneer has warned.
Yoshua Bengio, who won the Turing Award in 2018 alongside Geoffrey Hinton and Yann LeCun, said the recent rush to launch AI products from tech giants has become “unhealthy,” adding that he sees “a danger to the political system, to democracy, and to society as a whole.” “. the nature of truth”.
Bengio is the latest in a growing number of AI experts sounding the alarm about the rapid rollout of powerful, large-scale language models. His colleague and friend Hinton resigned from Google this month to speak more freely about the risks artificial intelligence poses to humanity.
In an interview with the Financial Times, Bengio pointed to society’s increasingly indiscriminate use of large language models as a serious problem, noting the current lack of scrutiny of the technology.
“Companies can rent ChatGPT . . . for example,” he said. “It’s important to monitor this access closely so we know who is using these systems and so we can track down potentially illegal or dangerous uses.”
Over the past six months, technology groups have raced to unveil groundbreaking generative AI products — computer software that can write text and create images — sounding the alarm. Google last week unveiled a revamped search engine powered by PaLM 2, its new artificial intelligence system.
Released in March, GPT-4, Microsoft-backed OpenAI’s state-of-the-art software, is rapidly being adopted by banks, consultancies, start-ups, educational institutions and governments to improve productivity.
“The dynamic of the game . . . it’s a vicious cycle,” said Bengio, a professor of computer science at the University of Montreal and founder of Mila, a Quebec artificial intelligence institute.
Regulators and governments around the world are stepping up scrutiny of the technology amid concerns that it could be misused. OpenAI’s chief executive, Sam Altman, called on U.S. lawmakers on Tuesday to regulate the field, outlining his concerns that it could provide “interactive disinformation” ahead of next year’s U.S. election.
Best known for his work on deep learning — the field that has driven the field’s recent giant advances — Bengio is one of the few senior AI researchers remaining in academia following the mass exodus of tech companies in 2019 one. Last year.
His academic peer LeCun joined Facebook owner Meta in 2013, while Bengio’s brother Samy was a senior research scientist in Google’s artificial intelligence division for nearly 15 years before resigning to join Apple in 2021.
Bengio is one of more than 1,000 AI experts and tech executives who signed an open letter last month calling for a six-month moratorium on the development of advanced AI systems.
While he doesn’t believe the letter will work, he said he supports the idea of a moratorium. “As far as I know, OpenAI, Microsoft, and even Google have different views on this,” Bengio said. “The whole community is divided on this issue.”
He pointed to the divide among AI experts as an important signal to the public that science doesn’t yet have the answers. “If we don’t agree, it means we don’t know.” . . if it could be dangerous. If we don’t know, that means we have to act to protect ourselves,” Bengio said.
“If you want humanity and society to survive these challenges, we can’t have competition between people, companies, countries — and very weak international coordination,” he added.
Near-term dangers of the technology include misinformation, insidious bias, and discrimination against marginalized and minority groups. These societal hazards are the focus of nonprofit research organizations such as the Distributed AI Institute, founded by former Google AI researcher Timnit Gebru.
But the long-term threat that also worries people like Bengio and Hinton isn’t today’s chatbots — which provide human-like answers to questions. Instead, they worry that research will rapidly advance within this decade to make AI systems more autonomous.
Bengio said the most urgent thing for regulators to do is to make AI systems more transparent, including auditing the data used to train them and their output.
He is also, along with others, proposing an international consortium to fund AI research in areas of public importance, such as climate and healthcare.
“It’s like investing in CERN or the space program — that’s the scale of public investment in AI today to really bring the benefits of AI to everyone, not just make a lot of money,” he said.
Bengio added that he is currently discussing ideas with Hinton for future collaborations on these challenges.
“There’s a lot of emotion in the broader AI community right now, a lot of shouting. But we need more investigation and more thinking about how we’re going to fit in with what’s coming,” Bengio said. “That’s the scientific method.”