(Montreal) It is urgent for governments to act to regulate the development of artificial intelligence (AI) because the speed at which it is currently being deployed represents a risk for democracy, for many jobs and could aggravate inequality.
This is the cry of alarm launched Monday in a joint work by Mila, the Quebec Institute of Artificial Intelligence and UNESCO entitled Artificial Intelligence Governance Blind Spots.
One of the 18 authors of the book, Yoshua Bengio, founder and scientific director of Mila, underlines for example that the famous ChatGPT tool from the firm Open AI “should be a warning signal for the population, for governments. Yes, very positive things can come out of all this, but the power of AI has reached a level where we need to focus our attention on it to avoid negative things, to minimize the negative effects. »
Risk to democracy and jobs
He places at the forefront of his concerns “the risk of using these tools in a way that would be dangerous for democracy. We have heard, for example, of “deep fakes”, images, videos, sounds which are synthesized by computer with these same technologies and which can be used to manipulate people, because people can be manipulated” , he recalls, referring to the increasingly present misinformation on social networks. “It’s something that can really endanger the foundation of our society. »
PHOTO HUGO-SÉBASTIEN AUBERT, LA PRESSE ARCHIVES
Yoshua Bengio
“It just takes an election and we elect a dictator who takes all the tools, who takes control. We have seen in recent years that there are dangers of skidding that exist, ”he adds.
On the socio-economic side, he says a significant impact on the job market is to be expected, with AI being a fast track to automation. “Do we have the social safety net we need to deal with these rapid transformations, for example? To put in place the protections we need, it will take time, it will take years and we see the speed at which technology is developing. You also have to get there pretty quickly. »
A necessary inclusion
Another worry is that AI tools are mainly developed in Western countries, points out Benjamin Prud’homme, executive director of AI for humanity at Mila. “We have to make sure to develop AI in an inclusive way. We have to ensure that all populations are represented. So it has to be true of white men, but it also has to be true of women, Indigenous peoples, racialized people. It is still a serious risk. »
Indeed, as Yoshua Bengio explains, the architecture of these systems and their algorithms are built on masses of data drawn from innumerable texts and images and, as a result, “will thus perpetuate the biases, the discriminations that they find in our culture”.
Also, adds Benjamin Prud’homme, AI “has the potential to aggravate North-South inequalities, inequalities within States between different populations. These risks must be taken seriously and regulated”.
Investing in the sociology of AI
Because one thing is certain, it is not the companies themselves that will come to self-regulate. These companies have a history of developing technologies at breakneck speed without worrying too much about the consequences. According to Professor Bengio, this is changing with employees becoming more aware of the impacts of their tools, but it is not enough. “We need to invest, not just in AI computing, but in the sociology of AI, in how we have to rethink our rules at the government level, how we have to rethink society. »
AI, he says, has the potential to generate huge profits for a very small number of players, with the risks that this entails. “It’s mostly power, the concentration of power, the concentration of wealth. AI must be at the service of all. This means that the wealth that is created must be redistributed, we must give ourselves the means to redistribute it. »
Local and international governance
Guest speaker at the launch and co-author of the book, Professor Kate Crawford of the University of Southern California Annenberg and principal investigator at Microsoft Research, argued that the AI framing effort must not only be domestic, but must also transcend states and take place on an international scale, as was the case when the International Atomic Energy Agency was created.
“I don’t think we have the tools of international governance with enough teeth today, but we have to get down to it,” adds Yoshua Bengio, who recognizes that, in the nuclear field, it took a dozen years before getting there, which would be unthinkable in this case.
It combines the management effort required for AI with that of the fight against climate change, two areas where it is absolutely necessary to hurry because the urgency is already there.
“It took years for nuclear, but we have to get started.
“As long as there is life there is hope, but it’s super important and it’s the same problem with the climate, if we give up and say we can’t do anything , that’s the worst thing. »
For Benjamin Prud’homme, Canada has a crucial role to play in this area. Bill C-27 on artificial intelligence and data, in the course of adoption, can serve as a trigger. “One, it takes international leadership; two, it takes leadership from certain major states and I think that Canada has a fairly ambitious bill in this area. Honestly, at the pace of the parliamentary process, I think Canada may be the first country to regulate. »