If we don't sort digital inclusion, forget about AI

Digital inclusion was on the agenda at Labour party conference this week. It’s an important topic in its own right, given we have to do so much online, from work to socialising, health appointments to banking.

It’s also going to be critical if Labour want to achieve their goals around economic growth and improving public services, both of which are intertwined with technology, data and digital services in their ambitions. AI is a key part of how they hope to achieve all of this.

But as I was reminded the other day by this article, there is still so much to do to make existing digital services user-friendly, before we go anywhere near AI. A group of relatively well resourced and digitally confident parents in the Netherlands were asked about their experience of applying for child benefits online. They reported it being challenging, complex, and not easily accessible.

This echoes some of the thoughts I didn’t have space to fit into my recent essay on digital social security.

At the heart of Universal Credit is the belief that a digital-first system is preferred by and accessible to everyone. There is insufficient attention paid to the barriers to digital access and how this affects people’s ability to make and maintain a claim. Around half of claimants need help to register their claim, and many people have ongoing difficulties managing the online service. They are frustrated by the automated nature of the system, where human support is not available to most. When things go wrong, as they often do, both claimants and staff struggle to explain what’s gone wrong. Weaving AI-powered decision-making into systems that are already opaque is not likely to increase transparency or make them easier to access by people who might already be finding them hard to engage with.

UC is in part designed to familiarise people with doing things online and increasing digital literacy, but this isn’t going to work if claimants are digitally excluded: not only will their digital literacy not improve, but they might also lose their (meagre) income from UC. Claimants can be sanctioned if they fail to complete their required tasks, which can include 35 hours of job searching every week. Doing that without affordable, reliable and consistent digital access is a huge ask. It hardly needs saying, but if someone already digitally excluded loses their benefits because they have been sanctioned, their digital literacy is not about to improve.

The notion that we’re ready to embed AI into our public services, when there are still so many people excluded from essential services and support because of a lack of progress on digital inclusion, is misguided. Far from magically cutting costs and improving outcomes, adding AI risks deepening exclusion for many reasons, not least if we don’t crack digital exclusion at a systemic level. Which, incidentally, needs much more than donated devices and skills training, important as those are.