House of Lords sets the tone on regulation of automated decision-making in the UK public sector

The House of Lords has introduced the Public Authority Algorithmic and Automated Decision-Making Systems Bill to help regulate the use of algorithms and automation in public sector decision-making processes.

The private members' bill, introduced by Liberal Democrat peer Lord Clement-Jones on 9 September 2024, aims to respond to concerns about the potential widespread harm that could be caused by these types of systems, especially when deployed by public bodies in sensitive areas like healthcare, immigration and social security.

Public authorities' automated decision-making

The bill applies to the use of algorithmic or automated decision-making (AADM) systems developed or procured by public authorities. These are defined broadly, as "technology that either assists or replaces the judgement of human decision-makers" but excluding system used only to automate calculation and implementation of formulas, where the calculation could otherwise be carried out manually and would be "fully understood". The bill would not apply to use of AADM systems in test environments, or for national security.

New obligations

The bill would impose a raft of obligations on public authorities, all aimed at ensuring that the use of AADM systems is carefully considered in advance to ensure that they meet appropriate standards, are assessed in operation on an on-going basis, that decisions are logged and retained, and generally that public authorities are open and transparent about the use of these systems. Public authorities would be prohibited from deploying AADM systems that couldn't be effectively assessed or monitored.

The bill requires public authorities to complete an algorithmic impact assessment prior to procuring or using any AADM system. They must also keep an algorithmic transparency record. Both the impact assessment and the transparency record must be published within 30 days of completion and updated whenever there are changes to the functionality or scope of the system. Neither obligation applies to AADM systems which are used only for formulating the public authority's policies and are not expected to "" determine the content of the policy.

Algorithmic impact assessments must take account of principles such as fairness, compliance with the Equalities Act and the Human Rights Act (for example, in avoidance of bias), minimisation of harms, and appropriate oversight, governance, audit and public transparency. In particular, they will need to include a description of the system, a risk and benefit analysis (including in relation to personal data and safety) and details of risk-mitigation undertaken. There are also requirements for audit, and for external scrutiny of the system's effectiveness and accuracy.

Algorithmic transparency records would include a detailed description and explanation of the AADM system itself and how it is used; as well as information about human oversight and how it is used to inform administrative decisions.

The bill also proposes an independent dispute resolution service to allow challenges to decisions made by an AADM system and orders for compensation if the challenge succeeds.  

Osborne Clarke Comment

Private members' bills rarely become law, but even if they do not, they can be politically influential on particular issues and affect legislation indirectly. While this bill is not a government-backed piece of legislation, it broadly aligns with the new Labour government's more "hands on" approach to regulating artificial intelligence (AI) technology, as outlined in the King's Speech and its supporting documents.

However, it seems likely at this stage that the government will remain focused on progressing highly targeted AI legislation regulating the most advanced foundation models, like the large language models that power the latest AI chatbots. In any event, the government will likely take some time to consult in this area prior to introducing any formal legislation, given the inherent complexities and financial/social significance.

Companies hoping to sell AI (and other types of computer system) which involve AADM to public authorities should ensure that they monitor the progress of this legislation. More broadly, companies looking to develop, commission or deploy AI systems should consider the wider implications for these systems as, regardless of whether this particular bill makes it onto the statute books, it is an indicator of likely direction of travel for AI legislation in the UK.