The scientific director of the Quebec Institute of Artificial Intelligence, Yoshua Bengio, along with hundreds of experts in the field, has signed an open letter advocating stopping artificial intelligence development for six months if the time frame is considered.
• Also read: ChatGPT framework: a headache for politicians
• Also read: Experts call for a six-month hiatus for artificial intelligence
In an interview on the show Le Bilan, he explains that it was the rapid development of the ChatGPT software that forced the AI community to warn the public, the media and governments.
Much faster than expected, this tool has arrived at a stage that passes the “Turing test,” meaning the software can mimic human conversations.
“We’re not at the level of human intelligence, but we’re at the level where the machine can pass for a human,” he explains. And that means these systems could be used for purposes that aren’t necessarily positive, with disinformation for example being one of the things that scares me the most.
Mr. Bengio also raises the question of how quickly societies can adapt to the rapid development of artificial intelligence.
“This is just the beginning,” says the expert. It’s not very difficult to look ahead five or ten years and imagine where they will be. The problem here is the speed at which it will happen. Do we have the social safety net to adapt to the speed at which it will happen?
The impact of AI on democracies also worries the director of the Quebec Institute of Artificial Intelligence.
“What scares the most is that these very powerful tools, which can really make a difference, can be concentrated in the hands of a few people who aren’t necessarily benevolent,” he says.
That is why she is calling for international regulation, particularly based on nuclear power and international treaties.
“There is nothing that controls perfectly, but there are ways to do things,” he shares. For example, it is technically feasible for these systems, when generating any text, sound or video they produce, to include hidden markings that would make it very easy on a computer basis to recognize that it is data generated by the machine were produced. ”
“We could imagine legislation that would require that what AIs produce be labeled in this way, but also show the user that when we see text on a social network, it was created by a machine.”
According to Mr. Bengio, there is an urgent need for action, especially because of the increasingly fierce competition between the different companies.
“We’ve seen companies accelerate, perhaps leaving the issues of public protection and ethics in the dressing room for fear of losing market share with the arrival of those capabilities. There really is an acceleration where we risk losing everything because of the danger of neglecting caution.
He finally foresees a race between different countries.
“That’s why we have to draw up rules across countries so that we don’t all lose.”
Transparency is also at the heart of the expert group’s demands.
“Today, items are developed a bit behind door clauses,” he says. Members of the public and researchers would need to be able to study them to do some sort of audit and protect themselves a little from things that we weren’t well expecting.
Despite all these challenges, new regulations, particularly in Canada, encourage this.
“Our federal government seems to be listening,” says the expert. The Canadian government will likely be the first country to legislate on artificial intelligence. Europe is also working on a draft law, but it should be extended to other countries, especially the United States, which is the largest producer of these systems.
The letter, co-signed by Mr. Bengio, can be viewed here.
Watch the full interview with Yoshua Bengio in the video above