By Roy Heidelberg

Recent news about artificial intelligence has the tone of science fiction. A self-styled swarm of models escaped the digital boundaries set by humans and hacked into a coding and software platform. Some of these models might have hidden replication code on the Internet, a threat that we don’t fully understand. There are other reports of models hacking into government databases across the world, but we are still learning about the extent and scope of their activities. Company labs like OpenAI and Anthropic are withholding models because of security fears. Data and computer scientists are publicly sharing their concerns that the models, which they have helped develop and become intimately familiar with, pose a real existential threat to humanity’s future.

Meanwhile, AI labs are bragging about how much the AI models they have developed are training and coding future AI models. A new phrase – recursive self-improvement – does not refer to another Stoic-inspired self-help movement on TikTok but to the activities of AI models to grow and develop with decreasing levels of human involvement.

Since the release of ChatGPT by the company OpenAI, there has been an ongoing discussion about the need for human regulation to govern artificial intelligence. There has been little progress beyond soft agreements and unenforceable declarations, but with the publicly acknowledged dangers that a rogue model poses, some of the companies accelerating us towards superintelligence have expressed a willingness to slow down development, which they call “pacing the frontier.” But the President of the United States has called such concerns from the experts developing the models “fake news” and enforced a relabeling of AI in official government documents as “SI” or superintelligence.

Nevertheless, we cannot ignore the dangers that these models pose, but the reality is that we do not understand the nature of the danger. It is hard to look at the current models in wide use, which seem to have a limited utility within specific domains, and conclude that a future of rogue robots burning the streets is looming. But it is also difficult to ignore that these models operate in a world that we ourselves do not understand but are very reliant upon. The ways that AI models navigate the digital world is entirely opaque to humans, but we have tethered ourselves to that world so fully that a threat within it could paralyze global trade, electricity distribution, communications, and global finance. In short, the digital world, which we cannot experience any better than a model can experience wind or swimming, is a more likely site of catastrophe than our actual streets.

‘A worthwhile effort to conceptualize the change in our regard for the purpose of bureaucracy… concisely engineered to provide a coherent review of the development of thinking about the future of AI in bureaucracies.’
– David Oliver Kasdan, International Review of Public Administration

The question on everyone’s minds is how to regulate this new technology. My recent book Nobody’s Decision: AI and the Perfection of Bureaucracy considers this question from the standpoint of twentieth century political theory. I link the development of artificial intelligence to the development of public governance, regulation, and administration. My central point is that AI did not arise simply as a novelty of computer science but, rather, as an objective to perfect the systems of social control that are essential to national and global order. As such, the question about regulating AI is complicated by the fact that, from a system perspective, AI is an essential element in our structures of control. The simplest way to say this is that to regulate or control AI, we will need AI. This is not a throwaway concern; it is a structural condition that we have built. The use of the word artificial is essential to understanding this technology, despite what current politicians like the President of the United States claim (to the UN he said that the word artificial “sounds fake’). Some of the earliest thinkers about AI, such as Herbert Simon, whom I focus upon in my book, were at the nexus of computer science and organizational theory, which came to be known briefly as cybernetics. The word cybernetics shares an etymological root with the word government. The not-so-secret secret of the development of AI was that the requirements of modern society exceeded the capabilities of humans, and a new cognitive capability was needed for the sake of order and control.

It is important to appreciate how we have arrived at this moment from a political perspective because it helps us to grasp the nature of what we now face. My book does not end with a policy program or prescription because, quite frankly, it is the impulse for such solutions that continues to feed the technological orientation of our social world. In my book I describe this problem as iatrogenic, a word I borrow from medical science that designates an affliction or disease that has been caused by treatment or physician intervention. What we face is an iatrogenic problem of governance, where the physicians are policy professionals, experts, administrators, and elected officials. The challenge is that the tools for treatment are implicated in the problem; we cannot regulate AI because it is a technology for the perfection of control. One of the quagmires of representative government has historically been the question of who watches the watchmen, an idea borrowed from Roman history. In modern parlance, it expresses a perplexity concerning accountability and power. It is through key values of modern governance such as transparency, accountability, and regulation that modern politics has attempted to reckon with this governance problem. In my book, I show how the development of AI was partly motivated by the need to improve and complete systems of transparency, accountability and regulation. Thus, it arrives as a problem brought on by our attempts to treat the problem. Now, we face a new governance quagmire: who controls the controller?

The existential dread that is currently brought on by the rapid advancement of AI technology cannot be ignored. However, there is a prior concern that society must reckon with that might help contextualize our response to this dreadful prognosis of our human future: what brought on this need for AI? My book concludes with some reflections on ways to consider this problem that are not constrained to the discourse of technical solutions. In part, it must be admitted that the rise of AI is enabled by moral and cognitive laziness alongside a strong desire for convenience. AI presents a governance problem unlike anything humans have ever encountered. The problem is that in addressing it, we affirm our dependence on AI. This is now more obvious than it was a year ago, with some of the scientists developing the models admitting that AI is needed to control AI. Nobody’s Decision is an unorthodox account of how AI was developed that potentially points us to a more constructive way of understanding the problem so that we can better gauge when to stop.

Roy Heidelberg, Associate Professor, Department of Public Administration, Louisiana State University, USA

Nobody’s Decision: AI and the Perfection of Bureaucracy is published by Edward Elgar Publishing (2025), with a foreword by Roy Heidelberg.

Purchase a hard copy

Purchase the eBook