Contribution
My Lords, I thank my noble friend Lord Holmes of Richmond, the noble Lord, Lord Clement-Jones, and the noble Baroness, Lady Bennett of Manor Castle, for their amendments in this group and for their thoughtful introductions and contributions to what is clearly a vital and timely discussion around the future of artificial intelligence in the workplace.
Amendments 148, 149 and 150, tabled by the noble Lord, Lord Clement-Jones, seek to define AI systems in statute, mandate a workplace AI risk and impact assessment, known as a WAIRIA—I was not sure how to say it—and impose statutory consultation duties on employers prior to the deployment of such systems. In my respectful opinion, these measures go a bit too far at this stage.
To begin with the proposed definition of an AI system, I fully accept the need for clarity in legislation, but the definition offered here is overly expansive and risks capturing a vast range of tools, from predictive text and email sorting to payroll systems and basic data analytics. Technology evolves rapidly, and we think that any attempt to lock such a broad and fluid concept into rigid statutory language at this point risks hindering innovation and forcing employers into compliance regimes for systems that may pose no meaningful risk at all.
On the proposal for workplace AI risk and impact assessments, the intentions behind this are understandable. However, the execution here reads a little more like a blueprint for a full-scale regulatory regime rather than a light-touch safeguard. Employers would be expected to carry out detailed documentation, consult staff, assess and monitor impacts on mental health, contractual terms, pay and more, and then repeat that process at least annually or upon any system change—and that is no small task. For large employers it might be possible, but for SMEs it would surely be burdensome and, in many cases, entirely unworkable. Our concern is not with the principle of transparency or fairness but with the disproportionate bureaucratic weight that these provisions would place on businesses, particularly those outside the technology sector, which simply may not have the capacity or technical knowledge to meet such a standard.
The third proposal, which is a statutory duty to consult employees or trade unions at least one month before deploying AI systems, again assumes a degree of foresight and technical certainty that may not always exist in practice. The development and use of AI systems is often iterative, and definitions, use cases and impacts evolve over time. Requiring formal consultation at every turn risks paralysing technological progress and may well deter even the cautious adoption of beneficial systems.
I recognise that the amendments are rooted in a desire to protect workers and uphold ethical standards, but we have to resist the temptation to reach immediately for sweeping, front-loaded legislation in a domain that is still very much in its infancy. Regulation in this space, as I am sure all noble Lords would agree, must be agile, proportionate and grounded in practical reality.
We are entering a new phase where AI is no longer confined to research labs or boardrooms. It is appearing across ordinary workplaces, public and private alike. There are tremendous opportunities to improve productivity, streamline operations and foster collaboration between human workers and AI tools, yet we think many of the proposals in this group—including those seen elsewhere, such as the mandatory appointment of AI officers, rights to personalised algorithmic explanations, opt-in clauses, and statutory principles of fairness and explainability—share a common shortcoming, which is that they attempt to legislate about highly technical and fast-evolving systems with a degree of rigidity that may prove counterproductive.
We on these Benches share the Government’s ambitions to become a world leader in this space, and therefore we must remain mindful of all those factors, as well as of existing protections. Several of these proposals risk duplicating duties already present under UK GDPR, data protection law and various existing employment safeguards. The creation of overlapping, inconsistent or duplicative regimes could confuse employers and regulators alike, all while doing little to prevent truly harmful practices.
While we acknowledge all the opportunities that are potentially offered by AI, we must remain vigilant to the risks that it poses, including algorithmic bias, opacity and decision-making—which we have heard a lot about—and the misuse of personal data. But that vigilance must be coupled with regulatory restraint. We ought to be cautious to not impose premature, overly burdensome rules that stifle innovation and overwhelm well-intentioned employers, particularly in low-risk use cases, such as rota planning, document handling or payroll automation.