While the risks of artificial intelligence (AI) for our democracies are increasing, professor at the University of Montreal and world authority in the field, Yoshua Bengio, maintains that our elected officials are not well equipped enough to face the challenges of the phenomenon.
“Politicians are not sufficiently aware of the speed at which things are moving and the importance of the collective decisions that will be taken through our parliaments in the coming years,” the professor immediately maintains in an interview with La Presse. Canadian.
The Mila Institute, which Yoshua Bengio founded, offers training on AI, particularly to elected officials. “We want to ensure that politicians from all our jurisdictions and all parties can better understand the issues,” explains the man who received the Medal of Honor from the National Assembly of Quebec on Tuesday.
“We started this type of thing with the federal government. We would be entirely interested in doing it with our Quebec Parliament,” he adds.
The National Assembly indicates that it has not organized training on AI until now, but that activities on the subject are sometimes held.
Interference, disinformation, hyperfaking
The dangers linked to AI for democracies are increasingly present and numerous: interference in elections, disinformation or even hyper-rigging (deepfake). We can in particular think of a false interview between Prime Minister Justin Trudeau and the controversial American host Joe Rogan. Even though the voices were very realistic, they were generated by AI.
Like a dystopian novel, the professor at the University of Montreal imagines a scenario where political parties could train AIs which would dialogue on a social network with voters with the aim of influencing their political opinion and their vote .
“Imagine that politicians who are trying to convince citizens could enter everyone’s living room and talk for months and months. Obviously with human beings it is not possible, but with AI it is,” argues Mr. Bengio, adding that voters would not know that they are discussing with artificial intelligence.
And the AI’s speech would be personalized. “Each person could be targeted differently,” adds the professor.
We already know that politicians adapt their speech according to the political clientele they are courting, but the scenario imagined by Yoshua Bengio could lead to significant deviations.
“We must frame this type of thing with an ethical perspective and return to the primary meaning of democracy, that is to say a fair sharing of power between everyone, that we can all participate in collective decisions and that these decisions are the result of an informed, rational and inclusive debate,” says the man who was named among the 100 most influential people in the world by the prestigious “Time” magazine.
And if we push the anticipation scenario even further, is it possible to imagine that our elected officials could be replaced by AI?
“Even if we had AIs more intelligent than us, it is not clear that they would faithfully represent the general will of a population. There is no question of important decisions being made by AIs. For me, it’s something that is dangerous and that we must avoid,” he says.
The benefits of AI for democracy
Yoshua Bengio still wants to point out that AI could bring positive things to public debate and democracy.
“At the speed at which things are moving, we could envisage AI helping us identify whether a speech or response is truly a contribution to the discussion and not a repetition of things that are not answers to questions, for example,” explains the researcher.
“It could force a certain discipline of debate which is oriented towards something constructive rather than a game hockey,” he adds.
AI can also be used to do more effective research so that MPs are well prepared when they go to parliamentary committees.
We could also imagine that it synthesizes a legal text in order to make it clearer, or to allow it to be compared with other laws. “The ability to analyze textual content is already very advanced,” says the professor, while reminding that there is still work to be done.
AI tools like ChatGPT can be used for research, but they should be used sparingly as they can produce incorrect answers.
“Robots coming to kill us”
The researcher says polls show citizens generally favor greater control over AI.
However, this issue is often very low on their list of priorities. “And so, it is not a priority for governments,” adds Yoshua Bengio.
He draws a parallel between AI and climate change on the fact that the most concrete impacts will be seen in the future. “When we look out in the street, we don’t see robots coming to kill us. We also don’t see the climate which is completely demolished,” he illustrates.
So, although this is a major issue, the professor affirms that things are not moving quickly enough to regulate the use of AI.