Artificial Intelligence | Towards a law assisted by algorithms

Artificial intelligence will revolutionize justice and law. His unimaginable abilities pose a threat to fundamental rights, but they also enable fairer access to justice. Researchers looked into the question from May 5 to 7 at the Time World 2022 congress, at the University of Montreal.

Mathieu Perreault
The Press

Forgery and use of forgery

In 2020, a mother introduced into evidence, in a child custody case in Britain, an audio recording of her ex-husband uttering threats. Lawyers for the father, who lived in Dubai, examined the recording and managed to prove that it was manipulation by artificial intelligence (AI) software. “This is the first example I’ve seen of the use of a deepfake “, explains Benoît Dupont, criminologist at the University of Montreal, whose conference was entitled “Can AI threaten the integrity of justice? “.

“Challenging the reliability of evidence in court is one of the three big challenges of AI for the justice system. Others are the risk of AI algorithms inadvertently threatening justice through unpredictable biases, and the possibility of a cyberattack modifying AI systems in ways that introduce bias. »

Mediation

The first application of AI systems in justice will be mediation in consumer cases, statements of offense and small claims, according to Karim Benyekhlef, professor of law at the University of Montreal. “I think we’ll see that by 2025,” says M.^and Benyekhlef, who has worked on online mediation platforms since the 1990s, at the time in internet domain name cases.

“There is already an AI system for the Administrative Housing Tribunal that offers case law. It would be enough to modify it a little so that proposals of mediation are added. I work with Saskatchewan for AI platforms for consumer disputes, ticket disputes, and labor standards cases. And one of my collaborators in Utah is piloting AI mediation proposals in four counties. » M^and Benyekhlef has also made computer mediation platforms, without AI, for the Consumer Protection Office in Canada and for an administrative tribunal that deals with conflicts in the world of co-ownership in Ontario.

Augmented justice

According to M^and Benyekhlef, we will not arrive soon, and perhaps never, at definitive judgments established by the AI. “I prefer to speak of augmented justice: we increase the power of the litigant, says the Montreal lawyer. We will direct the quantitative decisions, for example what sum in damages has been awarded in other cases for a defective fridge, for the loss of a leg, without binding the judge. The AI ​​for example could never have generated the judgment legalizing same-sex marriage, because the analysis is based on the past. A judge can assess the evolution of the socio-political context. This constant evolution of justice sets it apart from other areas where AI is currently used, such as marketing. “It’s fantastic technology for stable and predictable environments, but in court there are a lot of individual circumstances, inferences to be made from cases that aren’t totally similar,” adds Dupont.

Probation

Artificial intelligence will also be used for probation or parole evaluations, where currently algorithms are already causing controversy. “We use manual algorithms based on actuarial calculations, which could be automated,” says Mr. Dupont. But there are already concerns that biases are introduced by the algorithms. The a priori and a posteriori control of the biases introduced by AI algorithms will require a lot of manpower to analyze the reports and detect the biases, according to Vincent Gautrais, professor of law at the University of Montreal. “We can, for example, draw inspiration from the equivalent of the Access to Information Commission in Great Britain, which is well funded by a royalty system,” says Ms.^and Gautish. Big bugs are discovered by accident, but you need a monitoring system. »

This does not exclude a prior system for analyzing possible biases. UNESCO, for example, is currently working on a system of preliminary analysis of the impacts of AI on human rights, notes Catherine Régis, professor of law at the University of Montreal. “The human rights impact analysis of a project is already used, it would suffice to apply it to AI projects,” says Ms.^and Regis. It’s a bit like an environmental impact analysis. »

Surveillance

The regulation of facial recognition will be a major issue for the law, according to Céline Castets-Renard, professor of law at the University of Ottawa. “Canada’s privacy commissioners have said so [la semaine dernière]we need to better regulate this application of AI, says M^and Castets-Renard. One could imagine for example that it is possible to use it in the perimeter around the scene of a crime, but not everywhere in the city in terms of this same crime. The use of commercial databases of faces also poses a problem. The European Union is currently working on a text framing the use of AI by companies, whose major points of contention are facial recognition and the use of “social scores” giving access to certain products. and services, according to the Ottawa lawyer.