University of Birmingham (Law School & School of Computer Science) and UK Parliament

Taking seriously the need for 'human rights by design measures' based on engineering conventions to ensure safe systems and products as a key element in effective and legitimate AI governance.

• Protection and promotion of human rights
• Transparency, accountability, and human oversight
• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI

These are the basic foundatinons for buliding and deploying AI systems that serve the needs and interests of humanity, rather than the quest for profit fuelled by unchecked capitalism

Democratic participation and implications are not readily visible in the current thematic agenda

AI governance to ensure that it is trustworthy, reliable and rights-respecting in the UK has been continually downplayed by the current and previous governments.

Taking a broad section of views across all sectors of society, whose voices are repeatedly overshadowed by the tech industry due to its vast resources and intensive lobbying

We believe that a 'product safety approach' to the governance of AI systems, adapted to protect human rights from AI-generated harm, offers the most potential to serve as a legitimate and effective foundation for securing rights-respecting AI governance. All persons have a universal basic interest of safety. Over time, safety engineers have developed well-established methods, systems, and conventions that reliably and effectively ensure product safety. These are embodied in modern safety product safety laws, placing legal obligations on the producers of goods to ensure that they are safe for use before sale, backed by criminal sanctions, economic disincentives and independent regulators. It entails structured approaches of risk assessment, conformity testing and, where warranted, third-party approval. Through the concept of 'state of the art', this incentivises an economic rising tide while driving up safety improvements. Four safety engineering conventions are central: • systematic identification of hazards and hazardous situations, including reasonably foreseeable 'risk scenarios' to identify potential safety risks during design and development; • the identification, implementation and testing of safety risk controls to prevent safety failures at source, including the so-called 'hierarchy of risk controls'. 'Inherently safe design' risk controls are the most effective controls. 'Human rights by design' controls can, by analogy, be embedded into AI system design; • the systematic verification and testing of technical systems to evaluate the effectiveness of system functionality and of risk control, as part of a systematic approach to safety risk management; and • automated logging and recording mechanisms to capture system behaviour, facilitating ex post investigation of serious incidents, fostering accountability and continuous learning and improvement. AI governance should embrace a product safety approach that mandates that human rights protections are engineered, as far as possible, into the design, development and deployment of AI-systems to secure end-to-end accountability and human rights protections across the AI lifecycle.