Part 2 - The Pension Regulator's AI plan - Accountability

In last month’s blog, I explored The Pensions Regulator’s (TPR’s) AI plan and what it could mean for pension schemes. The plan sets out a thoughtful approach to encouraging responsible innovation in this fast-moving area, while keeping the focus firmly on good outcomes and clear accountability for how AI is used. 

Over the coming articles (including this one), I’ll discuss some of the themes it raises in more detail. I’d suggest they’re probably best read in the order that I’ve written them to reflect the internal structure of TPR’s plan. 

So, we’ll start where the plan starts. With what - in my view - is its most important and simplest point. The one thing that must not change in the age of AI. 

Accountability 

Nothing about who is accountable changes. TPR’s plan states clearly “Trustees and scheme managers must ensure schemes are well run and deliver good outcomes for members, regardless of which technologies are being used. They remain accountable for decisions and outcomes even when they delegate activities to providers or advisers, such as administrators.” 

AI does not create a new category of decisions that somebody or something else answers for. Quite to the contrary, it enables new ways of reaching decisions that funds are already accountable for. 

This is an important nuance because AI brings with it a subtle temptation. Anyone who’s used ChatGPT knows that AI can make it feel easier to hand decisions over to technology. If we are not careful, “the AI model recommended we do this” could start to become an increasingly used explanation for poor outcomes. 

In fact, there have already been legal cases in other industries where organisations have been held accountable for errors made by their chatbots. As an example, Air Canada’s customer assistant famously hallucinated a non-existent discount fare. 

So far, these claims have been relatively minor, but for pensions, where the financial stakes can be very large, the consequences could be significant. 

The Regulator’s plan attempts to close that door before it can open. In doing so, turns what could be seen as a philosophical question, into a practical one with which to get started; “Do you know where AI is already working on your behalf?” 

For many funds, that might be a more uncomfortable question than it first appears. 

The likelihood is that many could already be using AI without a formal decision ever having been taken. AI has proliferated quickly over the last few years. As a result, it now may well sit inside an administration platform’s fraud screening, a member portal’s chat assistant or a supplier’s document processing. Much of that will have arrived not by way of an explicit board decision, but through IT procurement processes. These tend to focus more on concerns such as cybersecurity and infrastructure compatibility, than accuracy and fairness. 

So where should we start with all this?  

It’s maybe too obvious to state that good governance begins not with a policy but with visibility. You simply cannot govern what you cannot see, so understanding how and where AI is used at your fund, is a necessary precondition of good governance. 

The question then becomes not just whether AI was approved at the point of adoption, but whether it remains fit for purpose over time. 

If you would like to discuss anything further, please get in touch. 

Important information

This blog is based upon our understanding of events as at the date of publication. It is a general summary of topical matters and should not be regarded as financial advice. It should not be considered a substitute for professional advice on specific circumstances and objectives. Where this blog refers to legal matters please note that Hymans Robertson LLP is not qualified to provide legal opinion and therefore you may wish to obtain independent legal advice to consider any relevant law and/or regulation. Please read our Terms of Use - Hymans Robertson.