Regulating AI to protect our democracy

The possibilities of AI are both exciting and unsettling. Harnessing the positive potential of new AI technologies will firstly require robust policy frameworks, says Daniel Tsai, an expert in artificial intelligence and technology law.

In June 2022, the Government of Canada tabled the Artificial Intelligence and Data Act (AIDA) as part of Bill C-27, the Digital Charter Implementation Act, 2022, anticipating the significant impact that Artificial intelligence (AI) systems may have on Canadians and the country’s economy. The act takes a three-pronged approach: building on existing Canadian consumer protection and human rights law; ensuring that policy and enforcement stay relevant as the technology evolves; and prohibiting reckless and malicious uses of AI.

But the regulatory framework and proposed legislation is still too narrow in mandate and will come too little too late, warns Tsai, who is a lecturer at Ted Rogers School of Management with expertise in artificial intelligence and technology law. AIDA is not expected to come into effect until at least 2025.

Tsai says AIDA, as it stands now, is limited to privacy protections and the use of private data as opposed to artificial intelligence, and argues it should address the need for transparency, accountability and the protection of fundamental human rights.

“Government of Canada officials haven’t called on any of the leaders of big tech, like OpenAI's, Chairman and CEO, Sam Altman, or the leaders of Facebook, Meta, Google and Microsoft – companies all deeply immersed in AI – to talk about how they're going to protect consumers and citizens, jobs and livelihoods as well as democratic institutions, from the implications of AI,” says Tsai.

Tsai is a former senior Canadian government policy advisor with experience drafting new Canadian laws and advising senior government officials on policy changes. Given the speed and sophistication in which AI is evolving and the exponential rate in which new uses of the technologies are being applied, he urges that it is critical that laws and policies keep pace.

Increasingly, AI is being developed with a theory of mind, which means it can recognize human emotions and thought processes, says Tsai. This advancement in AI capabilities uses neural networks modeled on the human brain and allows the technology to make inferences, predictions and even reason. However, this also increases the possibility of persuasion and manipulation.

“People are susceptible to emotional manipulation, such as online love scams and phishing attempts, which exploit people's emotions and trust,” says Tsai. “As AI systems become more sophisticated, they may be able to fake empathy and manipulate emotions to a dangerous extent, potentially influencing individuals to commit actions that are unethical, harmful and illegal.”

The destabilizing threat of AI

The proposed AIDA policy framework is inadequate because it doesn’t address the destabilizing risks AI poses to our democracy, says Tsai. Even with conventional social media, the harmful effects of the proliferation of disinformation – due largely to AI algorithms programmed to attract the most views and interactions – can be stark, he says. A powerful example is the U.S. Capitol attack on January 6, 2021. 

“Simple AI found that content that was false but inflammatory had high engagement value, which is that Donald Trump won the election, that the election was stolen and that there was a mass conspiracy preventing his re-election,” says Tsai. “This is called heating; when you have algorithms that take content that's controversial but untrue and amplify it by boosting it to others.”

Tsai believes Generative AI is going to have farther-reaching impacts as it becomes more sophisticated, such as the ‘deepfake (external link) ’ computer-animated video and audio content that is seemingly authentic. 

He was lead contributor on AI policy recommendations in the recently released  (PDF file) McGill University Law School and Desautels Faculty of Management Report (external link) . The impact paper argues that social media platforms serve as digital public spheres where political opinions are shaped, making them crucial for social, economic and political stability. As a result, the report proposes four recommendations around accountability, transparency, responsibility-by-design (the concept of minimizing or preventing potential negative societal and ethical impacts when developing new technologies) and enforcement. 

The report also recommends regulating the use of AI by large online platforms in curating, amplifying and moderating divisive content to encourage corporate social responsibility in the development of algorithms and AI systems. Finally, it recommends a government regulatory authority to oversee the use of AI systems on large online platforms that can conduct technical audits to assess reliability and bias and release publicly accessible evaluation reports.

The Black Box effect

The development of generative AI that can autonomously code and debug itself raises concerns about unexpected decisions and actions made by AI systems, says Tsai.  “This phenomenon, known as the Black Box effect, occurs when AI operates in ways that programmers cannot fully understand or control.”

And as AI gains the capacity to independently code, these impacts could snowball. For example, the use of AI and autonomous weapons by the military raises concerns about transparency and the potential for misuse. Instances of successful cyberattacks on North American electrical grids, the compromising of personal information and the hacking into banks to steal credit card details highlight the growing risks associated with the misuse of these technologies. 

“With the rapid use of AI for consumer purposes, but also potentially military purposes, and by malicious actors that want to do us harm, we have gone from the world of science fiction into the world of science fact,” says Tsai.

Understanding AI’s transformative impacts

For democracy to function, citizens need access to trustworthy and unbiased news to make informed decisions. AI algorithim’s tendency to favour content that's controversial over accurate may pose great threats to this, warns Tsai.  “The horses are already out of the barn. Now you have to catch them.”