Thoughts on digital transformation and international politics

Stay tuned

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.

The Robo-Bureaucrat?

The Robo-Bureaucrat?
AI-generated picture asking for a public servant using artificial intelligence

Welcome to the seventh issue of my monthly newsletter

I'll be sharing analysis and short stories about digital transformation, practical recommendations, or recommended reading on this platform.

This time let's talk about AI and its use in the public sector. From prototypes to actual use-cases making ministers tumble, from turbo-charging election campaigns to facilitating policy-making, the potential and actual use-cases of artificial intelligence in different aspects of a government bureaucracy or democratic processes are myriad. Maybe it is high time to take a good look at the century-old theoretical underpinnings of liberal democracy and given technological possibilities of today and tomorrow think about an update.

Please enjoy!

Democratic processes can be cumbersome and slow with lots of apparent inefficiencies - though that inefficiency is often an explicit choice and more of a feature than a bug, think about deliberation for example - there seems to be huge potential for automation and the use of digital tools.

I've previously written about the specter of the "Robo-Politican" concluding that behind the noise and hype it remains unclear how exactly AI will shape our politics as we often do not think about the fundamental shift digital technologies mean for our institutions but get easily distracted by individual use-cases. This lack of systemic thinking is also prone to lead to overblown fears.

For example, much has been said about the dangers of using AI - especially generative AI - in election campaigns as we face a year of highly contested elections in key countries. But it appears - at least so far- that those fears have been overblown (here and here).

That isn't to say that AI in the public sector is without danger. The work of NGOs such as AlgorithmWatch, researchers and journalists has uncovered many instances in which bureaucrats used AI-systems with negative impact for the affected population, e.g. the case of the welfare algorithm that led to the dissolution of the Netherland government.

But even with pushback and heightened scrutiny: AI in the public sector is here to stay. The database of the OECD Observatory for Public Sector Innovation lists over 130 AI use-cases in various stages of the innovation cycle among it's member states today. And while those use cases mostly focus on machine learning applications, generative AI has sparked whole new ideas for how AI could be used in politics, e.g. supporting the drafting of new policy.

I've created the featured image above for a keynote on AI in the public sector. I do not believe that we will have "Robo-Bureucrats" anytime soon but I do believe that AI will become an integral tool and part of the inner workings of administrations. The key then will be for democracies to find ways of using the benefits and potential of digital technologies while managing the risks and understanding the limitations. The best of finding this middle-ground is to play around with the new technologies and show their potential but crucially also their limitations.

I'm personally fascinated by the potential use of AI in policy-making. When updating software packets, packet managers understand the potential dependencies to other software already installed. Imagine a similar system for the production of legal texts - automatically checking new proposed rules against the existing body of law, highlighting dependencies or even contradictions to policymakers and also ensuring compliance of a proposed policy with the constitution or other legal texts.

I initially wanted to work on this idea as part of a hackathon organized by the Federal Department of Justice in Switzerland and Open Data and picked up the thread for a little thought experiment. I was inspired by a training on AI alignment approaches and immediately taken by the so-called "constitutional AI" approach pioneered by AI company Anthropic and it's LLM Claude. By playing around with different models, I looked at what happens if you use the Swiss constitution with an LLM and ask it to help you with policy. Find out how this little experiment went in my separate blogpost. Spoiler: we're still gonna need lawyers for the foreseeable future.

Also, I'd like to leave you with some recommended reading. As written above, far too little thinking is being done about reimagining our democratic institutions and processes for the digital age. A book I've recently read and highly recommend does just that. In "Automated Democracy" - currently only available in German - the authors Christian R. Ulbrich and Bruno S. Frey offer a systematic way of reexamining and reimagining all aspects of a democracy for the digital age and conclude: it is high time to have the fundamental debate on which digital state we all want to live in!

Subscribe to this newsletter and follow me on social media for more publications and thoughts:

Latest issue