A Criminology of Machines

AI agents should be treated as entities with genuine agency, not just tools. The paper argues that criminology must move beyond viewing AI solely as an instrument and instead recognize AI agents as having "computational, social, and legal dimensions" of agency—a fundamental shift in how we conceptualize responsibility and accountability in AI systems.

Machine-to-machine interactions create new crime risks that human crime theories may not explain. The author proposes that as AI agents increasingly interact with each other rather than just with humans, new forms of deviant or criminal behavior may emerge that existing criminological frameworks weren't designed to address, requiring new theoretical approaches.

We cannot assume AI will simply replicate human behavior. The paper identifies a critical question: "Can we assume that machines will simply mimic humans?" This challenges the common assumption that AI risks are merely extensions of human crime, suggesting autonomous systems may generate novel harmful outcomes through their own decision-making processes.

Multi-agent AI systems present distinct risks requiring systematic analysis. The author develops a "dual taxonomy" to characterize how interactions among multiple AI agents can generate unlawful outcomes, indicating that the risks multiply as AI systems become more interconnected and autonomous.

Criminology and AI safety must work together on emerging governance challenges. By drawing on Actor-Network Theory and AI safety literature, the paper argues that criminologists have a responsibility to help shape how society manages autonomous machines before widespread deployment creates uncontrollable harms.