"All-in on AI"?

The Government have published their AI Opportunities Action Plan, setting out fifty recommendations to grow the UK’s AI sector, drive AI adoption, and improve products and services in the public sector.

It is a welcome step from a lot of rhetoric to reality on the Government’s AI policy. ?But of particular interest to us here at Reform towers was the theme of “Adopt a “Scan → Pilot → Scale” approach in government”.

Scale is a rare commodity in government innovation. There are plenty of interesting AI pilots running in government — you can read about many of them online. The issue is that few ever get beyond that stage to working at scale, which is where the real benefits to the public come from — a kind of ‘pilotitus’ that we wrote about in September. Automation to reduce bureaucracy and cost, or novel services which improve diagnostics for life-threatening diseases, these are only valuable when the public accesses them.

But the plan tackles this head-on, with the aim of moving quickly from scanning ideas to deploying AI at scale across the public sector, with many recommendations which are familiar from our own paper, Getting the machine learning. Here are the key areas which jump out from our analysis:

Scan – “investing in building a deep and continually updated understanding of AI capabilities mapped to their highest impact challenges and opportunities”.

Key recommendations here, such as appointing an AI lead for each mission and developing an ‘AI evolution’ horizon-scanning capability in Whitehall are welcome. And yet this ‘birds-eye view’ approach risks driving AI adoption entirely from the centre, and not making it the responsibility of frontline services to identify and drive adoption in their area. Interviewees for previous Reform research papers and attendees at previous Reform roundtables have consistently argued that frontline public sector workers are, because of their experience delivering services, uniquely well placed to identify ways AI could improve their work, and to benefit from an improved service.

Pilot – “rapidly developing prototypes or fast light-touch procurement to spin up pilots in high-impact areas, robust evaluation and publishing results”.

As we argued in Getting the machine learning, improving access to talent and data, simplifying the pilot funding process and sourcing AI in a more varied manner would go a long way towards accelerating AI deployment across the public sector. Bringing in cutting-edge expertise is crucial, not just to understanding what users and frontline workers need, but what is feasible and possible with AI. And making data more interoperable makes it easier and lower-cost to test that formally. It is exciting to see the paper endorse a more flexible approach to procuring early-stage AI pilots, including using the Competitive Flexible Procedure in the new Procurement Act rules, which allows suppliers to bid based on demos rather than complicated bid documents which advantage large incumbents.

The test here will be how these recommendations are implemented in practice. A procurement process which “only layers bureaucratic controls as the investment-size gets larger” is similar to the much more risk-based approach we recommended — but similar promises have been made in the past. Watch this space!

Scale – “identifying successful pilots that can be applied in different settings… and rolling them out beyond organisational boundaries”.

Letting “1,000 AI pilots flourish” is a noble aim. But often the successful pilots are not scaled up and the unsuccessful pilots are not learned from, so time and money are wasted for no gain to the public sector.

It is welcome to see recommendations geared towards ending this issue, most notably the ‘scaling service’ for successful pilots with senior support and central funding. We called for targeted investment to fund new AI projects through an AI Transformation Fund in Getting the machine learning.

But scaling up successful pilots must go hand-in-hand with learning from failed pilots. Indeed, oftentimes this can be more important than scaling up successful pilots because it prevents further resources being thrown at an unviable idea. Alarmingly, a civil servant interviewed for our research was convinced that many pilots currently taking place are simple repeats of previously attempted failed pilots which were not adequately learned from.

Overall, the Opportunities Plan marks an exciting point of departure in the Government’s plans, from a lot of rhetoric to meaningful substance. But the job is not finished. We know Whitehall is often slow to act, struggles to prioritise, and is risk-averse — all issues which could make or break a plan on this scale. Of course the Government should be on the “side of the innovators”, a “great customer” for products and services and “build on UK strengths and catalytic emerging areas”. But it is easy to forget the hard choices which underpin them.