Getting the machine learning: AI-driven Whitehall
Researcher
At a time when departmental budgets are stretched to breaking point, and public services are struggling to clear large backlogs, artificial intelligence could herald a welcome productivity boost. The potential of generative AI technologies to unlock massive efficiency gains and turbocharge productivity is well documented, and public expectations are understandably high given their increasingly widespread experience with the technology in their personal lives. Yet adoption of AI tools in Whitehall has lagged behind the private sector and implementation has been patchy.
On Monday, at Re:State's annual ‘Re:Imagining Whitehall’ conference, we hosted a panel on the topic of ‘Getting the machine learning: AI-driven Whitehall’, bringing together experts from across the public and private sectors with firsthand experience of implementing AI in government.
Here are my three reflections on the event:
- Learning and knowledge sharing
AI capabilities are developing incredibly quickly. Trials and experimentation with AI tools show promise, with many creative applications and valuable lessons to be learned about best practice. It is essential that these ideas are shared far and wide, and crucially between far too-often siloed government departments. We need to avoid departments with common problems failing to share their experience, and wasting time and money on projects which have already been tried, but weren’t learned from.
- Building trust and credibility
As with all new revolutionary technology, AI adoption inevitably comes with teething problems. Inaccuracies, hallucinations, and unpredictable modes of failure are part and parcel of current generative AI technologies. Risk is unavoidable, but must be handled properly. It is imperative for building trust in the technology, and the credibility of the government departments using it, that those involved are absolutely transparent about the limitations of products, while also enthusiastically sharing the breakthroughs and success stories.
- Human-centred implementation
A common theme throughout the panel was how to balance the massive potential productivity gains against the human connection that is often essential for good public service delivery. Panellists stressed the importance of “human-centred implementation”: using artificial intelligence to solve the issues which are most frustrating for human users, such as freeing up frontline workers from time-consuming admin. By speeding things up, public services can reallocate this saved time to build a better understanding of the needs of their charges.