Scholars advocate for an approach to regulating artificial intelligence, centered around mental health.
A study conducted by the American Psychological Association (APA) found that 41% of teenagers reporting excessive social media use suffer from poor mental health compared to 23% of teens who use social media more frequently. These data highlight growing concerns among academics and policymakers about the intersection of mental health and social media technologies driven by AI.
In a recent book chapter, Przemysław Pałka, a law professor at Jagiellonian University, analyzes the European Union’s Artificial Intelligence Law (AIA), which provides the EU regulatory framework for AI, and analyzes its potential impact on mental health. Citing empirical research that can negatively affect users’ mental health by offering a range of AI-powered products, such as social media, chatbots, and video games, Pauka highlights the urgency to consider the psychological harms in AI law.
Pauka cites two important motivations for focusing on AIA’s mental health. First, he points out that AIA refers to “psychological harm” when condemning certain uses of AI, informing the drafters that they are trying to focus or at least consider the mental health of the law. Second, Pauka points out that AIA, the first regulation of AI across multiple sectors and industries, could serve as a template for other governments seeking to regulate technology. He urges lawmakers and academics to scrutinize the provisions in detail to ensure that the framework is covered in iron.
To provide context, Pauka first describes the AIA as a component of the EU’s broader product safety law. According to Pałka, AIA employs a scaled regulatory framework that classifies AI usage based on four levels of risk.
He explains that most regulatory attention is dedicated to high-risk systems such as biometric authentication and law enforcement tools. However, Pałka says many AI systems that have a major impact on consumers, such as content moderation, advertising and price discrimination, are excluded from these high-risk categories and are not regulated. Also, even if AI services are classified above the “high-risk” category, Pauka warns that AIA will only intervene in tight situations.
Prohibited use of AI is outlined in Article 5, explains Pauka. He notes that the law uses subliminal techniques to limit AI programs that distort or exploit vulnerabilities in users related to age and disability. However, Power emphasizes that service providers are only responsible for any actual psychological harm that can be demonstrated to users. Pauka argues that penalties for fines such as up to 30 million euros or annual sales fines and 6% of annual sales are of little use.
Worse, Pauka argues there is little clinical consensus on the definition of psychological harm for AIA drafts to refer to. According to Pauka, some scholars equate psychological harm with serious emotional distress or trauma. Others argue that psychological harm includes feelings and conditions such as fear, sadness, and addiction. Without clear guidance from the AIA, Power warns that the burden defining psychological harm could shift to private companies with financial incentives to underestimate its importance.
Pałka suggests that policymakers should instead focus on preventive mental health protection rather than having trouble defining the scope and meaning of later psychological damage. In other words, policymakers must adopt standards of “good mental health” and regulate technologies that can cause harm or contribute to the development of psychological issues.
Power adds that mainstream psychiatrists provide a clear clinical definition of “good mental health.” He cites the widely accepted World Health Organization definition of “good mental health” as “the ability to cope with stress, do fruitful work and contribute to the community.” According to Pałka, mental health risks include all interactions that increase the risk of developing a disorder. Decreased ability to cope with stress, reduced work productivity, or interference with community contribution are all examples of indicating a decline in mental health or potential development of a disability.
If policymakers adopt this alternative standard, Pałka argues that companies that produce algorithms that have been proven to reduce cognitive function will be punished or forbidden from the distribution of consumers, such as those with addictive designs aimed at keeping users engaged infinitely before the application causes real harm. In short, Pauka argues that “mental health” standards provide AIA sharp teeth to better protect users early.
However, Power acknowledges that transforming the current AIA framework is extremely difficult. He alternatively offers a solution that allows policymakers to adopt what remains within current standards.
First, Pałka explains that AIA can simply expand the category of high-risk AI systems to include applications such as content moderation, advertising, price discrimination, and more.
Alternatively, Pauka suggests that policymakers can adjust AI limits to the severity of potential psychological harm caused by AI systems. For example, Pałka argues that AI systems that contribute to eating disorders and self-harm should be subject to stricter regulations than those that can cause internet addiction, which can be alleviated through age restrictions or mandatory warnings used in tobacco and alcohol.
Finally, Pałka questions whether AI-specific regulations are the best way to make policy. He argues that perhaps common tort and consumer laws are more effective. However, Pauka warns that there are limits with proven legal precedents. According to Pałka, existing CaseLaw offers a wealth of analysis of physical harm, but recently the mental harm has been treated less severity and suffered from social taboos.
While AIA represents an important step towards regulating AI, Pauka argues that current provisions are lacking in addressing important risks to consumer mental health. By failing to define psychological harm, excluding key AI applications from high-risk classifications and relying on ambiguous definitions of key regulations, Pauka argues that AIA leaves a major gap in its regulatory framework. He asks policymakers to seize the opportunity to create more effective and comprehensive solutions at moments when public demands for AI regulations are loud and clear.
