KIU Publications

Artificial intelligence (AI) is increasingly capable of making autonomous decisions that can cause harm. This prompts important questions of criminal liability. The existing legal frameworks built on agency and intent are inadequate to accommodate situations where there is no direct human actor that caused the harm. This paper assesses the ongoing debate about whether AI should just be treated as a tool for which humans are liable for or whether AI ought to be treated as its own legal entity that can be held accountable. The research applies the retributive theory of punishment and classical legal rationales relating to mens area, the causal relationship and legal personhood to identify whether there are conceptual issues with attaching liability to AI. The research used a qualitative descriptive legal-philosophical approach to review secondary literature, international political documents and comparative legal models across multiple jurisdictions. The research identified key areas of law which are not currently adequately reflected in liability regimes, naming some areas of connection with corporate liability and product liability, as well as considering emerging ideas such as electronic personhood, strict liability frameworks and preventative regulatory frameworks. The preliminary conclusion is that although there is no consensus on whether there will be direct criminal liability for AI, much more reform is needed to avoid accountability gaps and to ultimately preserve public confidence in the law. The paper concluded with policy recommendations for better clarity between human accountability and possibilities for regulated AI personhood, requirements for safety and transparency, training for legal agents, the scope of international cooperation, etc., which would work towards a legal framework which serves the interests of justice in relation to autonomous AI systems while incorporating key conventions.