Regulation of AI Causing Accidental Death

This thesis examines how to regulate artificial intelligence (AI) causing accidental death, addressing both liability and regulatory structure. Liability is based on fault – the foreseeability of risk, and the steps taken to mitigate that risk. In order to attribute fault-based liability, one needs an explanation of how the harm was caused, and the mitigatory steps taken to avoid that harm. Explainable AI (XAI) is a complementary discipline used to explain why AI make decisions. XAI explanations provide an opportunity to bridge the gap between regulators and technologists by providing evidence of choices in AI creation, regarding demonstrating due diligence, and in risk identification and mitigation. For the liability of high-risk AI, this thesis proposes two standards: that the AI’s actions are comparable to a reasonable human counterpart; and that the AI’s producer be held to the standard of a ‘reasonable producer’ of AI, on the basis of what the producer knew, ought to have known or anticipated, and what they ought to have tested for, in the AI creation process. This thesis provides an analysis of both broad and sectoral AI legislation in the EU, US, UK, and Singapore, identifying commonalities and examining the lack of address of liability. Lastly, this thesis provides a jurisdiction-agnostic blueprint for regulation of AI causing accidental death, based on risk foresight and mitigation. It provides for the incentivisation of safe, explainable AI, and provides both procedures and defences for when an incident occurs. The proposed system has broad regulatory requirements for all of AI while still allowing for granular, sectoral specifics. The blueprint also proposes a specialised investigatory body for AI incidents, both to inform regulation and to assess claims relating to due diligence and liability. The system allows for integration with existing governance mechanisms, as well as for the rapid technological progress