As part of the recently launched Government AI Campus, Apolitical is publishing a series of articles exploring AI adoption in government. These articles take many forms, from op-eds written by academic experts to interviews with public sector leaders working on AI adoption themselves.

_Apolitical's Leonardo Quattrucci recently spoke to Dr Aaron Maniam, Fellow of Practice and Director, Digital Transformation Education, Blavatnik School of Government, University of Oxford, about how governments around the world might approach navigating new technologies like generative AI. _

Aaron Maniam focuses on issues connecting technology, public policy and public administration. He co-chairs the World Economic Forum’s Global Future Council on the Future of Technology Policy and is a member of the OECD’s Expert Group on Artificial Intelligence (AI) Futures. Previously a policymaker in the Singapore government, he was most recently Deputy Secretary (Industry & International) at the Singapore Ministry of Communications & Information, overseeing the ministry’s work in the digital economy, digital literacy and inclusion and digital diplomacy. Aaron’s PhD at the Blavatnik School of Government compared the work of leading digital states like Estonia, New Zealand and Singapore. From https://www.bsg.ox.ac.uk/people/aaron-maniam-0

Aaron, you can think about AI from multiple perspectives — as a civil servant and academic. Share a bit about your career journey and how it intersects with AI in government.

I first began exploring the topic of AI in 2010, during my tenure as Head of the Singapore Government’s Centre for Strategic Futures. At that time, AI — and technology in general — played a crucial role in shaping potential future scenarios, which the Centre studied. As my career evolved, my focus shifted toward the skills government leaders will need in the future. For instance, what will the future economy look like? How can governments better use, enable and regulate technology? Now, at Oxford’s Blavatnik School of Government, I try to bridge these insights for the government and engage with the wider academic community researching AI's technical and socio-technical impacts globally.

Generative AI has made AI a lot more accessible. How to govern AI, especially in a year with so many elections, is being discussed a lot. What is discussed less is how generative AI can be used to improve the competitiveness and performance of governments. How can — and should — governments re-equip themselves in this new age of AI?

It’s such an important question. Mostly, governments are good at knowing that they need to take on a regulatory role. That doesn’t mean that they always design great regulations, but they do understand their potential role in it.

I think governments can play two other roles, which are far less understood: as users and enablers of technology. More economically competitive countries understand the enabling role because they know that government is key to creating and authorising an enabling environment. But, when it comes to government as a user of AI, I find that it’s still an area where governments are relatively weak. I think the real question we should be asking is not “Will humans be replaced by AI?” but “What aspects of our jobs can AI take on that it can do responsibly and to a high level of quality?”

Generative AI is very good at divergent tasks: tasks which require synthesising wide-ranging existing content. So, for example, when government is trying to research an issue to understand what’s out there in terms of possibilities and ideas, generative AI can be used a lot more.

Generative AI is also good at some convergent tasks, i.e. issues that involve very specific areas of work. This includes things like editing and ensuring grammatical correctness. Generative AI is good at shortening or lengthening a piece of work and at including additional examples.

But we also need to recognise the sorts of tasks that AI is just not well suited to. These tasks involve some kind of judgement, evaluation or decision about what to prioritise.

💬 Let the author know what you think about governments being judicious users and enablers of technology like Generative AI leaving a comment below

I don’t think generative AI is good at generating data or evaluating things to high standards. This is where humans need to step in. We need to know what it can do and what it can’t; then have it do as much as possible of the stuff that it does well so that we free up the cognitive, emotional and bandwidth burdens on humans, who can then focus much more on what our minds were made to do: to think, to adjudicate, to evaluate and to make higher-level judgments.

What competencies and technologies do governments need to acquire to become better users and enablers of AI?

One of my favourite phrases is that governments need to not just be good at finding vendors, but they also need to be good buyers. This is a key skill. It can be broken down into three aspects. First, governments need parts of their staff to have technical literacy in engineering, programming and coding. These technically literate people are the ones who will ask the right questions about what to do, what to outsource and how to evaluate proposals from possible vendors. Second, you need leadership that is also technically informed. Leaders don’t need to be experts themselves but need to understand what a technology is and how it broadly works so they can link that technology and its functioning to their organisation's core mission and core purpose. That means partly knowing how to adopt and absorb technology, but it also means managing the change process that happens after adopting that technology.

The third aspect is that governments need procurement and contracting teams who are well informed on technology trends. They need to understand that the use of technology and its procurement will involve different techniques from procuring more standardised products. This means, for instance, understanding that technology is going to keep changing and, therefore, that contracts should not lock governments into long-term commitments. It means allowing for relational contracting within which they can keep making dynamic adjustments and evolving the technology in real time.

What you just described sounds like a coherent blueprint. But government work is often a bit of a scramble, especially in moments of technological transition. What are the main blockers or pain points that governments encounter in adapting to technological change?

There are a few of these, and a lot of the blockers exist because digital government is still government. It has all the possibilities and potential of government, but also the same pathologies, pitfalls, foibles and faults. One of the big blockers is an overemphasis on bureaucracy — governments end up blocking themselves from adapting well to technological transition because old rules don’t evolve quickly enough. Also, government is sometimes focused more on what it can supply than on what users need. And I think that when you don’t focus on user-centred design, you can sometimes end up supplying the wrong products and not adapting well to tech transitions.

Legacy structures are a second blockage. Most systems have structures already in place. Governments need to figure out which parts of those systems need to adapt. This is the kind of smart national infrastructure that Singapore is looking to build. Legacy structures can be very dangerous if they lock you into existing practices and mindsets.

Another issue is a lack of talent. Not having the right talent and skills can stymie the adoption of technology. One manifestation of this lack of talent is that governments inadvertently take on very, very large projects with significant timeframes and cost outlays. Such projects are nearly always destined to fail; by the time you finish even a portion of that large project, the technology will have overtaken it.

Modular projects work better since they can be modified and adapted with speed and alacrity. With such projects, the transition often proceeds a lot more smoothly because you are adapting in real time to what’s out there.

Speaking of speed and evolution… What governments are used to doing is looking for best practice. But, at a time when the challenges they’re facing are getting more complex, they will need to adopt new ways of thinking and operating, as you describe. This is why your recent work is on metaphors. Would you mind telling us a bit about that and what new metaphors governments need for generative AI?

You're right about the influence of metaphors on our thinking and their impact on what we prioritise and emphasise within a system. Introducing new metaphors allows us to highlight different aspects. We might begin by considering traditional metaphors for government, such as the “Leviathan” — a monstrous, large entity described in the Jewish Bible, symbolising government's vast size and scale. Another is Weber’s "iron cage" of bureaucracy, which refers to the strict, binding rules that govern how bureaucrats operate. Donald Kettl introduced a more modern, populist metaphor, likening government to a “vending machine” that provides citizens with what they want as long as they pay their taxes, regardless of whether it's beneficial for them.

However, we need more nuanced ways to view government. For example, Anne-Marie Slaughter suggests thinking of government as a “network”, and Tim O'Reilly has extensively discussed the concept of government as a “platform”. These metaphors emphasise the interconnectedness and enabling effects of government as a host for new ideas and institutions.

I particularly appreciate viewing government as a journey — a user journey that individuals embark on from birth to death, with the public sector playing a supportive role at various stages. This includes responding to significant life events like schooling, marriage, home buying, illness and caretaking at the end of life. Such a metaphor underlines the various interactions individuals have with both the private and public sectors throughout their lives.

My favourite metaphor, though, is that of an ecology. We are all part of a complex, self-renewing system where different components may die off while others spring to life. Governance, in this sense, involves everyone playing unique roles within an ecological system, contributing dynamically. Unlike a natural ecology where the lion eats the gazelle without consequence, ours is a moral ecology where we are interconnected and responsible for each other. Technology enhances this view by connecting us with far more people than ever before which can deepen our understanding of our ecological interconnections and strengthen them.

In moments of transition, it is easy to focus on trees rather than considering the whole forest, which leads to overlooking issues or misunderstandings. What do you think are the most overlooked challenges for government in this time of AI transition? And where and how should they start addressing them?

As I said, digital government is still government — with silos, bureaucracy, an emphasis on rules, etc. I think these are pathologies that governments really need to push against, especially when technology is democratising access to information.

I think culture wars are a second overlooked challenge. Identity politics is so much more prevalent and pervasive today than ever before, and is shaping how governments communicate with citizens. For example, governments can go about explaining to people why their identity and ideological beliefs are wrong — it’s not going to change their minds. In fact, it’s much more likely to re-entrench their prevailing beliefs. Ultimately, citizens need to have their needs as humans responded to by governments, and I think that governments which try to deal with the culture wars in a more empathetic way are likelier to succeed.

A third challenge comes from the fact that governments hardly ever deal these days with a stable equilibrium. We are often trained to try and get to what is called a steady state. But, in reality, there is no steady state. You have unstable equilibria — consistently new problems and possibilities emerging from discontinuous change. I like to describe this as not ‘new normals’ emerging but as ‘never normal’ environments because you’re always evolving and always moving to a new level of performance, or at least a new set of technologies. How we deal with this is very different to how we deal with clear blueprints or five-year plans. It isn’t even about having clear manifestoes, which can change as politics and operating environments evolve. I think we need to steel ourselves for the fact that life is never going to be stably normal. There will always be new evolutions, new paths and trajectories on which technology takes us. And the role of government is partly to keep pace with that if possible. If not, then at least to be able to respond quickly when circumstances change.

As a former civil servant, what advice would you give to civil servants today on how to learn and keep up with AI technology effectively?

In life, you need to balance optimism and caution. Imagine the best, but also plan for the worst. This applies to anything in government. The more you can imagine and harness the best, the more you can take advantage of the upside potential of any trend. It could be technology. It could be citizen engagement. It could be any kind of new emerging issue, but we always have to plan for the worst. Governments that don’t do this are being irresponsible.

But it really does have to be both. If you only look at the risks or at the downside, you lose hope or lose imagination. If you look at the upside, you become reckless and unrealistic. Good governance is about finding the sweet spot in between.

Done reading? Make sure to share how you imagine the best and plan for the worst with Genertive AI in mind by leaving a comment below ⬇️

(Image credit: Unsplash)