Artificial intelligence has the power to eliminate mundane jobs and create tremendous wealth, but it could also lead to widespread unemployment and reinforce existing inequalities.
That was the message at the forum on the socially responsible development of artificial intelligence, a two-day conference that ended on Friday.
Montreal has an opportunity to take a leading role in lead in ensuring AI technology is used responsibly, said Marie-Josée Hébert, the vice-rector of Research, Discovery, Creation and Innovation at the Université de Montréal and one of the organizers of the conference.
The forum was intended to bring academic researchers together with industry and government to exchange ideas about the responsible development and use of AI, she said. That’s important because the technology has the power to change the foundations of our society, she said.
Issues raised at the forum ranged from who is legally liable if a selfdriving car gets in an accident, to the possibility AI will lead to widespread unemployment.
Some of the issues are already here — like the ability for the creators of AI systems and the producers of data to introduce their own biases into AI systems.
“If you have an algorithm that sees biased data, gender-biased, racially biased, biased based on economic status and so on, the algorithm is going to ingest that and result in a biased model,” Doina Precup, a professor of computer science at McGill and the head of Google-affiliated DeepMind’s Montreal research lab, said in a session at the conference.
While AI systems are currently tools used by people, in the future, there could be general AI systems that act on their own, she said, which raises further questions of responsibility.
“Are AIs ever going to get to a stage where they’re sufficiently complex to be responsible for what they do? That I think is an open question,” Precup said.
There are also issues related to monopoly power, Yoshua Bengio, a U de M professor and one of the founders of Element AI, which sells AI services to businesses, said in a speech at the event.
A handful of large companies could control the data required for AI systems to “learn” and hire the majority of top researchers, creating a situation where other businesses couldn’t catch up, Bengio said. He also raised the issue of autonomous robots with the ability to kill, calling on the Canadian government to take a similar approach to this technology as it took with landmines.
“We are really at the cusp of something that it is important,” Hébert said. “It’s important to initiate these conversations before it’s too late, but it’s going to be as important to maintain these conversations as we go.”
As part of that process, the forum has created what it calls the Montreal Declaration for a Responsible Development of Artificial Intelligence. Over the next few months, that document will be developed through a process of “co-creation” and consultation with the general public.
The goal is to “establish a consensus on basic principles that are representative of our values,” Hébert said, “that should all guide us to how we are going to live through this phase of innovation and transformation.”
We are at the cusp of something important. It’s important to initiate these conversations before it’s too late.