All that glitters is not gold

It’s not hard to understand why artificial intelligence technologies have captured the political imagination. Any technology which promises the potential to get much more done with far fewer resources will be compellingly attractive to governments around the world.

In the face of such temptation it is important to anchor assessment of artificial intelligence in the here and now rather than looking too far ahead to what may or may not be possible. To try and make sense of what’s realistic in a fast moving environment we need to avoid being too optimistic or too pessimistic.

The ramifications of wider adoption of AI technologies across society are so broad in scope that there are myriad perspectives by which to look at the issue.

Research by Rootcause into the framing of AI in media discourse shows that these new technologies are widely acknowledged to be risky. There’s lots of debate about the nature of the risks, ranging from existentially threatening sci-fi scenarios of computers gone rogue to more prosaic but profound damage that can be done when AI decision making goes wrong in the delivery of critical public services. We’ve already seen examples of this in the Netherlands where an algorithm led to false accusations of benefit fraud against thousands of families.

The speed with which generative AI has emerged onto the market presents a range of ethical, philosophical and practical conundrums. The data supply-chain on which these new tools are built appears to make wide use of copyrighted material. The way they work, drawing only on the past, inevitably replicates biases that already exist in our societies and the energy demands they create are challenging for meeting low-carbon aspirations.

In the digital information environment, generative AI has the potential to intensify all of the trends set out in the previous article in this series, driving further fragmentation, empowering less-regulated media actors, blurring the line between fact and fiction even more and placing a further premium on data in an economy where the most powerful people usually already have the most data.

With all of this in mind it’s important that government decision makers take a holistic approach to AI adoption which balances the need to manage risks with the imperative to innovate. When we hand over human agency to AI we need to be certain it is safe to do so. AI Safety will help.

AI’s ability to gather, analyse, classify and summarise huge amounts of data at breathtaking speeds offers us new ways of making sense of the world. Although there are reliability issues to resolve there’s already scope to identify meaningful insight into human behaviour from a wide variety of new data sources.

This sensemaking capacity offers real potential for governments. The ability of AI technologies to reveal patterns in large amounts of data through processes like classification can help to uncover important behavioural insights that can shape smart policy.

In the UK the Government’s Incubator for AI recently announced a series of AI-powered products for civil servants which make use of the summarising powers of large language models to offer insight into previous public consultations when working on new policy ideas, or to give an overview of parliamentary debates and the perspectives of politicians.

On the face of it these are useful tools but as we explored in the previous article, cultivating community and building trust is an important goal for many public institutions. This is why it will be important to consider the potential for AI to be used to aid democratic discourse and not just to summarise perspectives from the past. There’s a range of exciting AI applications being developed that can offer much more deliberative public input into the work of government that forward thinking democratic institutions would be well advised to explore.

One area where mass-market generative AI is clearly able to have an immediate impact is around personal productivity. Chatbots operating on large language models make for tireless brainstorming partners and can handle endless requests with ease. Asking these tools to critique work from multiple perspectives can reveal helpful insight and hidden blind spots. They can think through the pros and cons of proposed policies, map out arguments for and against particular interventions, offer critical analysis of drafts and generally provide an extension of the brain’s capacity to move through creative blocks. This isn’t to say AI tools can replace human creativity, they can’t, but they can certainly augment it. There’s a growing amount of research which demonstrates that staff in major institutions are already using AI on the quiet for routine tasks. This ‘Shy AI’ use may hide the full extent of AI adoption across government. In many cases this would be better guided by putting in place policies that enable colleagues to learn and share how to get the most out of AI tolls whilst acknowledging and managing privacy risks to government data.

With so much potential for future applications of AI it’s no surprise that there’s plenty of hype. The fear of missing out is driving lots of organisations to take a closer look at AI and when they do so there’s no shortage of new companies looking to sell them products and services. A large number of these are simply not worth the money. Many are just wrappers of large language models that do things it would be easy for staff to do for themselves with a small amount of training in creative uses of AI. Some offer more advanced applications of AI which appear to automate or remove the need for time consuming manual tasks but when put to the test many have a last mile problem, they can get 90% of a task done immediately but the last 10% still requires a human (especially if accuracy is important) and by the time they are finished the time and cost savings aren’t what they first appeared to be. This is the buyer beware zone for AI.

The future of AI use in government will be decided by people, not technology. The choices we make in how we adopt, regulate, and deploy it are critical. The institutions that thrive in the coming decades will not be those that resist change, nor those that blindly embrace every new tool, but those that balance innovation with accountability and efficiency with ethics. AI is not a substitute for judgment, leadership, or trust. All three will be as important as ever in the years ahead.

(This is the third article in a short series. You can read the first and second articles here)