Automated Driving Systems (ADS) are moving from closed, geofenced environments intoopen public roads. In golf courses, industrial sites, and campus shuttles, current technologyalready delivers what can reasonably be called automated driving. The remaining questionsfor courts and regulators are not whether ADS are technically possible, but under whatconditions they can be deployed at scale while preserving safety and assignableresponsibility.This technical note introduces a conceptual framework based on Second Physics, in whichlegal responsibility ρ is treated as a conserved quantity within a socio-technical system wherehumans, institutions, and technical artefacts interact. In this view, AI systems cannot be thefinal bearers of responsibility: only human or institutional actors endowed with an underlyingsource f₀ can ultimately carry ρ. Applied to ADS, this implies that talk of ‘AI fault’ cannoteliminate responsibility; it can only obscure the flows of ρ between manufacturers, softwareproviders, operators, insurers, regulators, and users.The note identifies three structural constraints that any acceptable legal regime for ADSshould respect: (i) an AI Non-Terminality Principle — AI cannot be the ultimateresponsibility bearer; (ii) a Control–Benefit Alignment Principle — responsibility shouldfollow effective control over ADS behaviour and long-term benefits from its operation; and(iii) an Evidence Preservation and Disclosure Principle — logs, sensor data, and softwarestates must be recorded, retained, and made available to the competent authorities afterserious accidents. Without such a framework, victims cannot realistically prove ‘AI fault’,and prudential solvency constraints will deter manufacturers and insurers from large-scaledeployment, regardless of technical feasibility.
Toshisada Utsunomiya (Sat,) studied this question.