The UK government has published a policy paper on where it sees AI regulation heading in the UK and has put out a call for views. What is encouraging is that the importance of not interfering more than necessary with innovation in this area is highlighted all over the paper, starting with the title (“Establishing a pro-innovation approach to regulating AI”) and a stated desire for the UK to be the best place in the world to found and grow an AI business. The government’s stated ambition is to support responsible innovation in AI - unleashing the full potential of new technologies while keeping people safe and secure. How is this feat to be achieved?
The paper sets out a framework that is:
-
Context-specific – the responsibility for regulation is delegated to individual regulators rather than proposing a unified set of rules as in the current version of the EU's AI Act.
-
Pro-innovation and risk-based – a focus on issues where there is clear evidence of genuine risk or missed opportunities, with a focus on high risks rather than hypothetical low risks and avoiding the creation of barriers to innovation
-
Coherent – a set of light-touch cross-sector principles to ensure regulation remains coordinated between different regulators.
-
Proportionate and adaptable – in the first instance, allowing regulators to get on with regulating their areas rather than introducing additional regulation and encouraging light touch options such as guidance and voluntary measures in the first instance.
An important aspect of the proposal is the no-definition definition of what AI is. Perhaps with an eye on the controversy as to how AI should be defined in the EU legislative process for the AI Act, the proposal avoids having to define what AI is. Instead, it sets out two key characteristics of AI that need consideration in regulatory efforts: AI systems are trained on data rather than expressly programmed, so the intent or logic behind their outputs can be hard to explain. This has potentially serious implications, such as when decisions are being made relating to an individual’s health, wealth or longer-term prospects, or when there is an expectation that a decision should be justifiable in easily understood terms - such as legal dispute. This is, of course, well-recognised, and much current research is looking to address this. The other aspect is autonomy (although I prefer the term automation as more accurate of the reality of AI as a deterministic technology); that is, decisions can be made without express intent or ongoing control of a human. The best example is probably the use of AI to control self-driving cars, and the implications are clear regarding responsibility and liability for decisions made and actions taken by AI.
The government sets out its purpose behind this no-definition definition: “To ensure our system can capture current and future applications of AI, in a way that remains clear, we propose that the government should not set out a universally applicable definition of AI. Instead, we will set out the core characteristics and capabilities of AI and guide regulators to set out more detailed definitions at the level of application.” The decision to forego attempts at a universal decision and put detailed definitions within the remit of regulators at the application level is, in my view, highly sensible and avoids much unnecessary and unhelpful debate that is not necessarily based on technical reality. It has, incidentally, been proposed before in one of my favourite papers on the topic, illustrating the problems of trying to define AI at an ontological level rather than in terms of its concrete technical applications.
That is all well and good, I hear you say, but is this not going to lead to chaos and more rather than less red tape as each regulator adopts different, potentially overlapping and conflicting regulations? Well, maybe. There is always the potential for unintended consequences in any regulation. Still, at least the government is aware of this issue and proposes coordination between regulators and a set of overarching principles that all regulators should abide by as the solution.
The policy paper marks an early stage in the government's approach to formulating its policy on AI regulation. At this stage, they propose the following overarching principles, explained in detail in the paper:
-
Ensure that AI is used safely
-
Ensure that AI is technically secure and functions as designed
-
Make sure that AI is appropriately transparent and explainable
-
Embed considerations of fairness into AI
-
Define legal persons’ responsibility for AI governance
-
Clarify routes to redress or contestability
This policy paper sets out the government's current thinking. It provides an opportunity for stakeholders to make their views heard ahead of the White Paper and public consultation the government plans to publish later in the year (and the paper asks some particular questions on which views are thought).
I am no expert in regulation, AI or otherwise, but I work with AI innovation daily and welcome the government's pro-innovation focus. I also think it will be incredibly difficult to make meaningful regulation for a whole field of engineering/technology independent of its application, as the current legislative initiative in Europe is seeking to do. To my scientific mind, regulating AI per se makes about as much sense as regulating electromagnetism or statistics. I, therefore, appreciate the clarity the government's no-definition definition brings. Of course, in the end, all will depend on how this is implemented: will we see a light-touch regulatory regime in which regulators work together to provide clarity and certainty while protecting the public and meshing seamlessly with the international regulatory context? Or a byzantine set of conflicting and ineffective regulations suffocating innovation and enterprise in reams of red tape while leaving the UK isolated internationally? The legislative journey this paper starts will at least be fascinating to follow, and it is interesting to see the UK considering a different approach.