New Study Urges Complex Systems Approach to AI Risk Assessment

Experts advocate for a complex systems approach to AI risk assessment, emphasising long-term, systemic effects alongside immediate concerns.

As artificial intelligence (AI) becomes ever more embedded in daily life, researchers are urging a fresh approach to evaluating the risks associated with these technologies. The latest findings, published in Philosophical Transactions A by a team from the Complexity Science Hub (CSH) and TU Graz, argue that a comprehensive, complex systems’ perspective is necessary to manage the intricate, evolving relationship between AI and society.

Fariba Karimi, a leading researcher in algorithmic fairness at CSH and professor of Social Data Science at TU Graz, notes, “Understanding the risks of A.I. requires recognising the intricate interplay between technology and society. It’s about navigating the complex, co-evolving systems that shape our decisions and behaviours.”

The researchers underscore the need to address long-term risks, which they argue are currently overlooked in traditional AI risk frameworks. These risks, they suggest, can escalate due to the widespread adoption of AI technologies and their broad societal impact.

Short- and Long-Term Risks in AI

Traditional AI risk assessment has focused primarily on immediate harms, such as algorithmic bias and safety concerns. However, according to study co-author Dániel Kondor, these frameworks often fail to account for broader, long-term systemic risks. “These frameworks often overlook broader, long-term systemic risks that could emerge from the widespread deployment of A.I. technologies and their interaction with the social context they are used,” he explains.

In balancing short-term and long-term perspectives, the authors aim to better understand the effects AI technologies can have on society over time. This distinction is crucial, Kondor says, as “long-term risks are not the linear combination of short-term risks. They can escalate exponentially over time.”

To illustrate the potential pitfalls of focusing solely on immediate effects, the researchers examine the example of a predictive grading algorithm deployed during the Covid-19 pandemic in the UK. Although intended to reduce bias in grading, the system ultimately resulted in disparities that unfairly impacted students from disadvantaged backgrounds. Valerie Hafez, an independent researcher and study co-author, observed that “once the grading algorithm was applied, inequities became glaringly obvious,” with nearly 40% of students receiving lower marks than expected.

Computational Models and Public Engagement

The authors advocate for using computational models to explore these long-term effects in greater detail. By simulating potential outcomes, such models can offer insights that go beyond traditional assessment methods. As Kondor notes, “computational models—like those assessing the effect of A.I. on minority representation in social networks—can demonstrate how biases in A.I. systems lead to feedback loops that reinforce societal inequalities.” This approach, they argue, enables more informed decision-making by revealing complex patterns that emerge over time.

The team also emphasises the importance of public participation in assessing AI risks. They propose forming small, diverse “competency groups” to bring varied perspectives into the risk assessment process. According to Kondor, these groups can be instrumental in ensuring that AI risk management reflects the interests of those most affected by these technologies. Engaging with people from diverse backgrounds not only enriches the debate but also promotes social resilience, a quality Kondor considers vital for effective AI-related decision-making.

“A more general issue is the promotion of social resilience, which will help A.I.-related debates and decision-making function better and avoid pitfalls,” Kondor suggests. This concept of resilience, he adds, extends beyond technology, underscoring the value of participatory forms of governance.

Toward a Sociotechnical Understanding of AI

In the study’s final recommendations, the authors highlight the need for a more holistic approach to AI systems that considers both technological and human factors. Hafez, who is also an AI policy officer at the Austrian Federal Chancellery, stresses that separating the technical aspects of AI from the people it affects undermines the public’s role in shaping these systems. “I think that once you begin to see A.I. systems as sociotechnical, you cannot separate the people affected by the A.I. systems from the ‘technical’ aspects,” she says. Recognising this sociotechnical interplay, she argues, will empower affected communities to influence the development of AI systems that are responsive to their needs.

This call for an integrated perspective represents a growing consensus among researchers that AI technologies cannot be divorced from the societies in which they operate. The study’s findings reflect an urgent need for broader conversations about the societal implications of AI, with a focus on engaging not just policymakers and tech experts, but the wider public as well.