Today’s generative AI companies promise systems that will surpass human capabilities—and often portray that future as inevitable. Public policy historian Joy Rohde argues that these promises echo a long history of attempts—and failures—to use computers to predict conflict, model adversaries, and make policymaking more “rational.”
In her forthcoming book, People of the Machine, Rohde examines how Cold War-era efforts have shaped contemporary faith in artificial intelligence and computational governance. She teaches seminars on the history of public policy and on science and technology policy.
State & Hill spoke with Rohde about what we miss when we treat AI as an entirely new phenomenon, as well as what citizens and policymakers should ask before incorporating it into public institutions.
Tell us about the history of computational governance. Who are the "people of the machine"? How does this history help us understand the questions we face today about AI?
Much of today's discourse about AI presumes that AI technologies are new and their powers are unprecedented. This discourse implies that AI technologies have an inexorable internal momentum: The forms they take, the functions they perform, and the pace at which they develop are beyond human control. Whether we like it or not, they will replace us. But by tracing the history of AI since the 1950s, my book shows that the quest for artificial intelligence is a 75-year project that has been fundamentally shaped by U.S. political and national security interests.
Beginning in the 1950s, technologists and experts in international relations teamed up. Worried that humans were too poorly equipped to guarantee the survival of American democracy in a nuclear-armed world, they sought to create computer systems that could help government officials avoid political and military catastrophes. I call these experts in politics, policy, and computers the people of the machine.
Importantly, modern AI originated during the Cold War, when Americans believed they faced enemies that posed an existential threat to human freedom. In this context, the people of the machine viewed building AI systems that replaced humans with machines as more totalitarian than democratic.
So they intentionally designed systems that preserved human thought and agency. They hoped computers would sift information and provide interactive geopolitical models that analysts could use to make more informed decisions. The Department of Defense became more involved in their projects after the U.S. defeat in the Vietnam War. The military pressured researchers to create systems that tried to automate intelligence and decision-making. Because machines do not know, reason, learn, or choose like humans, their efforts failed repeatedly.
People of the Machine shows that the quest to replace humans with intelligent machines is a 75-year, multibillion-dollar failure."
When it comes to the urgent questions we face today, this history shows two things. First, AI design choices are not neutral; they have assumptions about human and machine intelligence encoded within them. Second, we have agency over those designs and the values our technological systems embody. We can design and deploy systems that replace humans. We can also design and deploy systems that preserve human agency and serve public benefit. Our AI future is not fixed or beyond our control.
AI has always been—and continues to be—shaped by human agency, values, and choices."
What do we misunderstand when we treat AI as a completely unprecedented problem?
We misunderstand our own power over our technological futures. We are not at the mercy of technologies that develop beyond our control, insulated from our values and goals. Our technological futures are shaped by our choices.
We also fail to learn from the past. Debates about whether AI benefits or is necessary for national security, or whether it is dangerous, are as old as the earliest AI projects. These debates are typically framed in terms we inherited from the Cold War: machine or human; control or freedom.
We have lost sight of some successful projects from the past, projects that started by asking: "What are humans particularly good at? What are computers particularly good at?"
These systems combined human experts with computers to make the most of their respective capabilities and were driven by human-centered values. We can choose to center humans now, and we will likely build more effective technologies if we do.
You describe the book's central tension as the desire for both "freedom and control." Why is that tension so important to understanding computational history?
The Cold War politics that shaped the projects in my book played out in the language of freedom and control. The people of the machine sought to preserve human freedom—intellectual freedom, political freedom, and national security, or freedom from fear.
At the same time, national security came to mean total security, total control. Their projects were therefore riddled with a fundamental contradiction: The aspiration for control conflicted with key political values of the era, such as the protection of individual freedom and agency.
These tropes continue to shape how we talk about computers and humans.
How did computer systems evolve from supporting human judgment to replacing it?
During the Cold War, the concept of national security came to mean total security—or as close to it as possible. This created an unattainable quest for absolute safety. As early computer projects failed to meet that standard, a group of developers decided that humans were impediments to control. They gave up their earlier commitment to human agency and designed systems that they claimed could predict and control the geopolitical future. Those systems failed, of course, but the promise of total global control was so tantalizing that deep-pocketed Department of Defense patrons funded them anyway.
Why does faith in computational governance remain so powerful?
One reason is that we have persistent and well-founded anxieties about human cognitive frailty in a complex world. There is so much information circulating—far more than unaided humans can take in and understand. Computers are great at information storage and retrieval, so it makes sense that we hope machines can do what we cannot: figure out what is happening and make decisions accordingly.
However, that hope has nourished a credulous acceptance of developers' claims—as old as the earliest AI systems—that computers will be able to think, reason, and learn like humans. We see that credulity, for example, when we say that large language models such as ChatGPT "understand" us or that systems "evolve" and "grow."
This faith also generates billions of dollars in government and venture capital investment, seemingly validating the promise.
What would it mean to design policy technologies that strengthen human judgment rather than displace it? What would it take to do this?
It would start by centering human values rather than the elusive goal of computational control or the attraction of venture capital. It would also mean working from a more forthright assessment of what machine "intelligence" is in the here and now, rather than in some hypothetical future.
It would mean weighing the financial, human, environmental, and social costs of these systems against their proven capabilities, creating the basis for an informed, evidence-based debate about our shared technological future."
What kinds of questions should citizens, policymakers, and scholars ask before adopting AI systems in public decision-making?
First, they should ask what's really behind the breathless claims about what systems can do? What do they actually do, how do they do it, and how often do they fail?
Then, they should ask what do public policy decisions require? To what extent do AI systems meet those demands, and to what extent do they fall short?
Today, much-touted generative AI systems run on statistical pattern prediction, which falls far short of the knowledge, moral reasoning, and wise judgment required to address the many complex problems we face.
Written by Rebecca Cohen (MPP '09)
Related Fall 2026 State & Hill stories
← Read all stories in the fall 2026 edition State & Hill
