Send message to Authors

Paper short abstract

Although systemic risks of AI are already subject to regulation, there is a lack of understanding their systemic nature, characterized emergent effects at a societal or global level. Besides a complementary role, their analysis may contribute to an analytical connection to existential risks of AI.

Paper long abstract

The regulatory framework of the EU AI Act addresses systemic risks of AI, including risks that are often discussed as existential risks, such as CBRN attacks, loss of control, or large-scale cyberattacks. Risks are understood here as caused by individual AI providers or users. However, we are already observing feedback and amplification dynamics, rebound effects, emergence of damages by data-based inferences and transfers of proxies, non-linear increases in the severity of violations of human rights, structural dominance and dependencies, and failures cascading along more integrated value chains. These systemic risks arise from complex interactions between different actors or AI systems, threatening societal goals as emergent effects at the societal or global level. However, their ‘systemic nature’ is not adequately addressed by current risk assessment and governance.

Besides being a complementary approach, we share the view that the analysis of systemic risks may also serve as an analytical connection towards existential risks. While systemic risks to safety, security, and environmental protection may be connected to existential risks, this is open for fundamental rights and democracy. The latter should not be given less attention in research to avoid trade-offs or lacks in governance. They too should be investigated as potentially catastrophic rather through processes of accumulation and erosion. Overall, their assessing and governing necessitate a better understanding of complex processes of interacting causes, pathways of cascading, compound or accumulated risks, and erosion or collapse types of damages, including complex collective action problems or externalities, to which this paper would like to contribute.