Canadian artificial intelligence experts and industry chief executives are urging Ottawa to pass legislation to regulate AI before the summer to deal with the potential harms, saying that any delay would be “out-of-sync” with the speed at which the technology is developing.
Around 75 researchers and startup CEOs, including deep-learning pioneer Yoshua Bengio and the leaders of Canvass AI and AltaML, have signed an open letter calling for MPs to pass the Artificial Intelligence and Data Act (AIDA) as quickly as possible. The act was introduced by the federal Liberals in June, 2022, as part of Bill C-27. It would establish a framework for developing and deploying powerful AI systems and set out financial penalties for individuals and businesses that are offside.
“We ask our political representatives to strongly support AIDA,” reads the letter, which was co-ordinated by Mila, the machine-learning institute in Montreal. “Unless parties work collaboratively to move AIDA forward before the summer, we are looking at significant delays.”
The government has said AIDA would not be in force until 2025 at the earliest.
The letter notes AI has the potential to boost economic growth and improve society, but also reels off a number of possible harms, including discrimination, misinformation, labour market turmoil and impacts to mental health. “And at the pace at which powerful AI systems continue being developed, other critical risks will most likely arise, while current ones may significantly increase,” according to the letter.
Mr. Bengio, the founder and scientific director at Mila, said in an interview there has been an “unexpected acceleration” with AI in the past few months owing to language-generating technology such as OpenAI’s GPT-4, along with text-to-image programs. “We really need public protection. These things are moving quickly, so we can’t wait another six months,” he said.
In March, some AI experts and business leaders signed a different open letter calling for a six-month pause on developing systems more powerful than GPT-4, which is capable of producing text, computer code and conversing fluently. The signatories argued that a temporary halt allows AI labs and other experts to create shared safety protocols.
AIDA is Canada’s attempt to enforce responsible AI development and deployment, and in introducing the act last year, the country became an early mover.
But some experts have criticized the act for being too vague. It’s not even clear which instances of AI will be regulated, for example. The act will apply only to “high-impact” AI systems, but the term is still undefined, creating uncertainty for businesses. The government intends to write more specific regulations after the act passes into law.
Mr. Bengio said that establishing a legal framework followed by regulations allows the government to more easily adapt to new dangers posed by AI, which is crucial given the speed at which the technology moves. “Having a law that leaves some responsibility to the government to react to problems as they go protects us and is going to be better for our businesses,” he said.
A rigidly prescriptive law, on the other hand, could create unnecessary burdens for industry or quickly be rendered ineffective.
Even though many crucial details still have to be determined, the framework AIDA provides is an important starting point, said Sam Ramadori, the CEO of BrainBox AI in Montreal, which uses artificial intelligence to improve energy efficiency in buildings.
“The biggest concern we would have is the absence of law while the technology is taking off,” said Mr. Ramadori, who signed the letter. “We want to be involved and have a further say, but we do feel like it needs to get moving.”
An email sent by Mila to solicit signatures and obtained by The Globe and Mail said that Geoffrey Hinton, who is often called a “Godfather” of AI, would be signing the letter, along with CEOs from companies such as National Bank of Canada. But they ended up not doing so.
“Professor Hinton finally decided not to go forward with the signing of the letter. As for other businesses, it was decided to publish a short list of signatories from the immediate AI ecosystem,” said Benjamin Prud’homme, executive director, AI for Humanity at Mila.
“We are supportive of the letter and agreed to sign it. However, we understand that MILA has prioritized companies from the AI ecosystem at this time,” said National Bank spokesperson Jean-Francois Cadieux in an email.
Mr. Hinton did not immediately respond to a request for comment.
The government has allotted almost two years for consultations and drafting regulations. Mr. Bengio, who is also co-chair of Canada’s federal advisory panel on AI, met with Minister of Innovation, Science and Industry François-Philippe Champagne last week and urged him to speed up the timeline. (The minister was engaged but non-committal, he said.)
Some regulations could be drafted and implemented relatively quickly, he added. As an example, companies should be obligated to disclose when users are interacting with AI as opposed to a human, and any text or other media generated by AI models should be watermarked so as not to deceive the public.
AIDA has drawn other criticisms, too. Some experts are concerned about the lack of extensive public consultation so far, and questioned whether there will be adequate oversight of AI under the law. The act would establish a commissioner that resides within Industry, Science and Economic Development – the federal department that is also charged with promoting the AI sector.
AIDA is also tied up with the rest of Bill C-27, which deals with consumer privacy and data protection. Some parliamentarians and experts have questioned whether it would make more sense to split AIDA into a stand-alone bill.