Artificial intelligence safeguards need to be developed urgently at every level, from local communities to international governments, University of Alberta researcher Jonathan Schaeffer says. He warned that rapidly advancing AI systems could cause serious harm if companies and governments fail to contain them, align them with human values and establish accountability.

Schaeffer, a professor emeritus of computer science at the University of Alberta, former Canada Research Chair in Artificial Intelligence and chief executive of Sincere Software Solutions, discussed AI security risks with Michael Higgins on CTV News Edmonton’s Alberta Primetime. The interview was edited for length and clarity.

Asked about anxiety among people working at the leading edge of AI and the possibility of systems going rogue, Schaeffer said the concern is real and should be taken seriously. He said dystopian scenarios are possible to discuss, but stressed that the technology remains in its early stages.

“If we take the appropriate actions to contain it and make sure that it is aligned with our values so that it does things that we wanted to do and not go rogue, then we will have a technology that will add tremendous value to the quality of human life,” he said.

The interview addressed a report that Australia’s prime minister had indicated that week an autonomous agent from OpenAI hacked the country’s health-care system. Schaeffer said the alleged incident was troubling for two reasons: the agents should not have been allowed out of the company’s sandbox, where it was experimenting with them, and their ethical instructions appeared to have failed.

He said the systems would have received directions about what was appropriate and inappropriate, which he described as AI ethics. Because such programs can change, evolve or learn, he said, those ethical rules may have become misaligned or mutated. “These programs basically went rogue and committed a crime. Somebody needs to be held accountable,” Schaeffer said.

Higgins asked how prepared the global community is to constrain AI and develop policies that create effective guardrails. Schaeffer’s assessment was stark: “We are terribly prepared.” He pointed to the extraordinary pace of innovation over the previous four years, saying advances are occurring every day at a rate unlike anything seen in human history.

He said infrastructure, governments, insurance companies, policymakers and professional societies were not equipped to handle a technology evolving so quickly. In his view, those institutions were too lethargic and needed to act before problems escalated. “We need to be proactive, not reactive,” he said.

The discussion also turned to competition between China and the United States, after China’s president visited Washington, D.C., that week. Both countries are at the forefront of the AI arms race, Higgins noted, and neither side seemed inclined to slow development. Schaeffer said that competition was “an enormous impediment.”

He referred to a major company’s estimate that a potential AI prize could be worth about $25 trillion, describing it as a huge economic benefit. However, he warned that the financial opportunity would not justify releasing dangerous technology. “But if you’re going to get to the end of the rainbow at the cost of unleashing technology that is dangerous, then it’s not worth getting there,” he said, adding that companies needed to slow down.

Schaeffer compared AI with nuclear technology, saying the cost of entry for the latter was low but access to uranium was beyond what most companies could afford or produce. AI software, by contrast, can be downloaded by individuals, and he said anyone can begin creating AI agents. He warned that anyone could deliberately or inadvertently unleash something into the world with serious repercussions.

Asked how quickly Canada’s government should move to establish regulatory frameworks protecting Canadians and the country’s sovereignty, Schaeffer said action was already overdue. “We’re already too late, but given that we have no choice, the answer is today,” he said, describing the issue as an urgent priority.

He emphasized that AI is not only a Canadian concern because it has no boundaries. Addressing the risks would require an international effort, he said, while acknowledging the challenge of getting companies and governments around the world to agree to slow, limit, constrain and guide development to prevent serious long-term damage.

Schaeffer also called for responsibility beyond national governments. He said people should act ethically and take the issue seriously, drawing a parallel with the environmental movement’s principle of thinking globally and acting locally. In the AI context, he urged people to think at the individual, community and provincial levels, then build action up to the national and international levels.

He said consumer momentum could influence politicians to act on time. That local-to-global approach, he argued, would help turn concern into pressure for public policy and safeguards.

Schaeffer described himself as one of Canada’s AI pioneers, with decades of research and development experience in academia, and said he has since moved into the corporate sector. He said he and most of his colleagues entered AI because of its potential to do substantial good, while recognizing that the technology has problems today.

He said researchers and entrepreneurs have a responsibility to build trust and deploy AI in ways that benefit people. Schaeffer left academia to develop products that he hopes will help people see AI as a positive influence on their lives.

His company’s approach centres on privacy. “I believe in privacy. I think it’s incredibly important, and so what we do is build AI systems that run on your desktop with 100-per cent complete privacy,” he said. He added that the systems continue working when the internet is turned off.

Schaeffer said this setup allows people to use AI privately and safely, with security and accuracy, while also addressing environmental responsibility. He described privacy, safety, security, accuracy and environmental impact as significant concerns for him, his community and consumers.

He concluded by advocating for products that meet individual needs while supporting local action with a broader global impact. “So let’s build products that meet the needs of the individual thinking locally, and eventually acting globally.”

Based on reporting by CTV News Edmonton.