Artificial Intelligence as a Tool or an Agent of Crime? A Comparative Examination in Contemporary Criminal Law

Abstract

Artificial intelligence is increasingly participating in activities that, if carried out by a natural person or even a legal person such as a corporation, would constitute criminal conduct; these activities range from triggering sudden crashes in financial markets to purchasing illegal narcotics and even running over pedestrians. We argue that criminal law faces serious deficiencies in cases where an artificial intelligence system effectively commits an offense while no identifiable human actor situated upstream can, as a practical or legal matter, be held responsible for that conduct. This article examines possible approaches to addressing this problem, with particular emphasis on the possibility of imposing direct criminal liability on artificial intelligence where it operates independently, autonomously, and in a manner that cannot be reduced to the conduct of human actors. The prevailing view is that punishing artificial intelligence is incompatible with fundamental principles of criminal law, including the attribution of culpability and the requirement of a criminal mental element (mens rea). Drawing on analogies with corporate criminal liability, strict criminal liability, and established principles of attribution, the authors demonstrate that the possibility of punishing artificial intelligence cannot be categorically rejected on the basis of cursory theoretical arguments alone. Punishing artificial intelligence may produce general deterrent effects and serve symbolic and expressive functions, and it is not necessarily inconsistent with the limiting principles of criminal law, such as the prohibition against imposing punishment disproportionate to culpability. Nevertheless, the article ultimately concludes that the direct imposition of criminal punishment on artificial intelligence is not justifiable, because such an approach may entail substantial costs and would undoubtedly require fundamental and far-reaching changes to the existing legal framework. Accordingly, the authors contend that limited and proportionate reforms to existing criminal laws governing human persons, combined with the possible expansion of civil liability, constitute a more appropriate and practicable means of addressing crimes arising from artificial intelligence.