Artificial intelligence (AI) is transforming far more than information ecosystems—it is reshaping the constitutional foundations of democracy itself. In this interview with the ECPS, Professor Danielle Allen argues that the central challenge of AI is not merely technological but constitutional, requiring democratic societies to rethink representation, political power, and civic agency. Rejecting technological determinism, she contends that democratic institutions can still shape AI’s trajectory, provided they strengthen rather than diminish human judgment. Discussing populism, democratic backsliding, surveillance, intermediary institutions, digital citizenship, and constitutional reform, Professor Allen presents a compelling vision of democratic renewal grounded in one core principle: artificial intelligence should complement human capacities—not replace the judgment upon which liberal democracy ultimately depends.
Interview by Selcuk Gultasli
Artificial intelligence is rapidly transforming the institutional architecture of democratic politics. While much public debate has focused on misinformation, deepfakes, automation, and employment, a growing body of scholarship suggests that the more profound implications of generative AI lie elsewhere: in its capacity to reshape political authority, democratic representation, and the constitutional foundations of liberal self-government. As AI systems increasingly mediate public discourse, organize information, influence political preferences, and concentrate unprecedented forms of economic and computational power, the central question is no longer simply how societies should regulate a new technology, but how constitutional democracies themselves must adapt to preserve political equality, civic agency, and democratic accountability.
Few scholars are better positioned to address these challenges than Professor Danielle Allen, James Bryant Conant University Professor at Harvard University, Director of the Allen Lab for Democracy Renovation and the Democratic Knowledge Project, and one of the world’s leading political philosophers working at the intersection of constitutional democracy, citizenship, technology, and democratic renewal. In her widely discussed Journal of Democracy essay, “The Real Dangers of Generative AI,” Professor Allen argues that contemporary debates have fundamentally misidentified the nature of the challenge. Rather than treating AI as merely a technological disruption, she contends that it represents a constitutional turning point requiring a fundamental rethinking of democratic institutions themselves.
In this wide-ranging interview with the European Center for Populism Studies (ECPS), Professor Allen offers a remarkably expansive framework for understanding how AI intersects with democratic governance. Rejecting the conventional separation between technology and constitutional politics, she argues that “political institutions are themselves technologies" and therefore cannot be understood independently of technological transformation. From this perspective, AI is not simply another digital innovation but a force capable of altering the institutional mechanisms through which democratic societies organize representation, distribute political power, and preserve constitutional order.
Throughout the conversation, Professor Allen explores the twin dangers confronting liberal democracies: democratic collapse, produced by the erosion of trustworthy informational infrastructures, and democratic singularity, arising from unprecedented concentrations of computational, informational, and economic power. At the same time, she rejects technological determinism. Instead, she insists that democratic societies retain the capacity to shape AI’s trajectory through institutional reform, constitutional safeguards, and renewed democratic participation.
A recurring theme throughout the interview is the centrality of human agency. Professor Allen argues that democratic renewal ultimately depends not upon resisting technological innovation but upon designing institutions that preserve civic judgment and constitutional self-government. As she puts it, "technology always develops out of a moral vision," making AI fundamentally a political rather than merely technical question. Consequently, she maintains that the essential democratic task is not simply regulating machines but ensuring that "AI should strengthen human judgment, not replace it." That principle ultimately serves as the interview’s normative anchor, connecting questions of constitutional design, democratic education, AI governance, intermediary institutions, populism, surveillance, and the future of liberal democracy into a single coherent vision of democratic resilience in the age of artificial intelligence.
Here is the revised version of our interview with Professor Danielle Allen, edited lightly to enhance clarity, readability, and overall flow for publication.
Political Institutions Are Themselves Technologies
Professor Allen, welcome! Let me begin with your recent Journal of Democracy article, "The Real Dangers of Generative AI." You argue that public debate has focused too heavily on misinformation, deepfakes, and job displacement while overlooking AI’s deeper capacity to restructure political authority and democratic institutions. Why do you believe the greatest threat posed by generative AI is constitutional rather than merely technological, and how does this fundamentally change the way scholars should understand democratic backsliding?
Professor Danielle Allen: Thank you. It’s an important question. I would start by saying that I don’t draw a distinction between the merely technological and the constitutional. Political institutions are themselves technologies. Representation, when it was invented in the eighteenth century, along with electoral districts and voting methods and so forth, was a technology. It was a technology for channeling citizens’ voices into decision-making and producing a counterweight to the other forms of power that existed in monarchical and aristocratic structures. Consequently, any major technological change that affects information, communication, and related domains has the potential to disrupt that foundational technology through which we organize politics.
For the last four decades, policymaking circles have prioritized material well-being while substantially downgrading questions of political organization. As a result, we have underestimated and under-attended to the ways new technologies can disrupt our core political technologies because of this broader blind spot in thinking about politics. In that regard, what I’m trying to do is rectify a broader blind spot in policymaking—to help people remember that our political institutions, whether people have political empowerment, access to voice, and so forth, are essential parts of the story of human flourishing. We have to pay attention to those things as policymakers. And secondly, we have to pay close attention whenever a major disruptive technology directly interacts with those original technologies of representation.
The Concentration of Power Poses Dangers to Core Democratic Rights
You distinguish between what you call "democratic collapse" and "democratic singularity." Could you explain how these two concepts differ? To what extent does the concentration of computational, informational, and economic power in a handful of corporations create conditions that resemble the concentration of political power traditionally associated with authoritarian regimes?
Professor Danielle Allen: Great question. What my co-author, E. Glen Weyl, and I were trying to do in that article was to describe two dangers that liberal democracies have to navigate. We call these the danger of collapse and the danger of singularity.
The danger of collapse is one in which the informational infrastructures on which you rely for popular sovereignty and institutionally driven decision-making no longer function because of misinformation, deepfakes, the collapse of identity systems, and things like that. That’s one danger the new technologies are provoking as they interact with the traditional technologies of representation.
The second danger is the concentration of power that you’ve been describing. That concentration can emerge in several ways. There’s the issue of capital, the concentration of capital, and the sort of political power that flows from it. There’s also the issue of the concentration of power through control of data and surveillance capacity. Both of those things harbor dangers to core democratic rights, whether those are rights of privacy or rights of equal voice in a political system, and so forth.
So, these are the two dangers: on the one hand, the possibility of chaos and collapse; on the other, the concentration of power. The hard question, then, is how the institutions of liberal democracy can steer through this period of rapid change and transformation, secure control over the new technologies, and truly integrate their capacities into the project of liberal self-government.
What We Really Have Is a Clash of Intermediaries
Across Europe and North America, populist leaders increasingly claim to represent "the people" while simultaneously attacking intermediary institutions such as independent media, universities, courts, and civil services. Does generative AI risk accelerating precisely this populist logic by allowing leaders to communicate directly with citizens while bypassing democratic intermediaries altogether?
Professor Danielle Allen: That’s a great question. I would put the situation slightly differently. I would say it’s not so much that they’re bypassing intermediaries as deploying AI as a new kind of intermediary. In that regard, yes, the traditional intermediaries are under pressure from a new set of intermediaries. Those new intermediaries are the tech infrastructures that populists are using to communicate. In other words, when President Trump establishes Truth Social as an alternative to the existing social media platforms, he is not dispensing with intermediaries but creating his own intermediary, namely that platform.
So, what we really have is a clash of intermediaries. The challenge is that the old intermediaries—universities, newspapers, the traditional institutions, and so forth—grew up in a legal environment, and actively defended a legal environment, that placed a set of basic rights at its center: freedom of expression, freedom of conscience, freedom of civil society to maintain independence from government, and the like.
The new intermediaries, by contrast, are not necessarily dependent on those basic rights for their growth or committed to protecting them. That’s where the problem arises.
This is captured, for example, in the challenge surrounding Section 230 of the US Communications Decency Act. That provision has essentially exempted social media platforms from regulation on the grounds that regulating them would amount to regulating speech. It collapses speech and algorithms. We ought to be able to regulate algorithms, which essentially perform an editorial or publishing function, in the same way that the print media found itself regulated.
Print media was liable for libel, defamation, and similar harms because of its editorial decisions, whereas social media platforms have been exempted from all of that. Mistakenly exempted, that’s the point I’m making. Because they have been able to flourish without the kinds of careful mechanisms for balancing the protection of speech rights with protection against harms such as defamation and libel, we have experienced rampant problems of misinformation. So, the point I’m trying to make is simply that the old intermediaries have baked into them a more stable and useful rights-based structure than the new intermediaries do.
The Personalized Algorithmic Universe Is Becoming Completely Unreliable
Your work suggests that democracy ultimately depends upon citizens sharing a sufficiently common informational environment. Yet generative AI enables highly individualized persuasion at unprecedented scale. Are we moving from an era of mass propaganda toward what might be called algorithmic individualized populism, where every voter effectively receives a personalized political reality?
Professor Danielle Allen: That’s a great question. To some extent, I think the answer has to be ‘yes’. Although I think there’s a situation beyond that. What is really happening is that everybody is going to be thrown back on communication with their own trusted networks, and that’s what they will rely on as their source of information.
The personalized algorithmic universe is also one that people are already coming to recognize as completely unreliable. You can’t distinguish between AI-generated things and real things, and so forth. As we develop digital competence—and by that I mean the ability to use these tools ourselves so that we can understand how easily they can be manipulated and manipulate others—we will inoculate ourselves against some of the dangers of that algorithmic environment.
What that, then, means is that we will basically trust the people we actually know and our communications with those people. So, in a funny way, the very big world that we had made small through communications technology is about to become much bigger again. It is going to be harder for people to share trusted messages at scale, precisely because the level of distrust is going to be so high.
We Need Algorithms That Encourage People to Bridge Viewpoint Differences
One striking feature of contemporary populism is its capacity to transform emotional identity into political loyalty. To what extent do recommended algorithms and generative AI amplify precisely those forms of resentment, grievance, and identity polarization upon which contemporary populist movements increasingly depend?
Professor Danielle Allen: They certainly seem to be very active contributors to that. There’s important work being done now by technologists and political scientists to propose alternative structures for algorithms that are focused on pro-social incentives—algorithmic incentives that encourage people to build more bridges across viewpoint differences—and better contextualization tools so that people can see the full spread of opinion and understand where their own opinion sits within that landscape. That helps avoid some of the effects and emotional impacts that come from the siloization of opinion.
The problems you’re describing are all very real. They are the dominant problems in our social media and algorithmic landscape at the moment. At the same time, there really is innovative new work. I’m thinking of people like Aviv Ovadya and his colleague Luke Thorburn, and others who are generating an alternative set of structures for our algorithmic foundation.
Existing Scholarly Foundations Can Be Extended Into the Age of AI
Much scholarship on democratic erosion focuses on executive aggrandizement, judicial capture, electoral manipulation, and attacks on independent institutions. Does AI require us to rethink democratic backsliding itself by recognizing that democracy may now be weakened through the concentration of technological infrastructure rather than solely through constitutional manipulation?
Professor Danielle Allen: I’m not so sure if it requires fundamentally different analytical categories. When, for example, people were studying the genocides in Rwanda, Serbia, and Croatia, one of the key technologies that facilitated them was radio and its use to spread slurs, dehumanizing rhetoric, and memes that helped frame the ultimate victims of genocide as enemies and non-humans, and things like that. My point is that the study of autocracy has long included the study of technology and mass communication. Those pre-existing scholarly foundations can simply be extended into the contemporary context. I don’t think it’s a change of kind so much as a change of scale.
We Don’t Face an Emerging Problem—We Face an Already Existing One
You argue that democracy requires not merely political equality but also a broad distribution of power. Looking at the rapid consolidation of frontier AI among a handful of technology companies, do you believe liberal democracies face an emerging form of private constitutional power that existing democratic institutions are largely unequipped to regulate?
Professor Danielle Allen: I think we don’t face an emerging problem; we face an actually existent, interior problem. It has already emerged, and it’s a totally real problem. Europe has done a better job regulating this power than the US has. I’m thinking about the GDPR(The General Data Protection Regulation) and other forms of regulation, including the recent EU AI Act. That tells me it’s not that liberal democracies lack the capacity to regulate these technologies; it’s that the US, specifically, doesn’t have the capacity right now to regulate them. So, this is not a permanent problem but a political problem here in the US.
I also think we probably have a lot to learn from the late nineteenth and early twentieth centuries, the first Gilded Age, as we call it. We’re living through a second Gilded Age, and just as in the first, industrialization produced corporations with enormous concentrations of power, robber barons, and corporations that sought monopolistic control not only of the economy but also of political structures. We’re watching a repeat of all of that.
Those concentrations of power were broken up in the early twentieth century through a very serious array of anti-trust policies. People often point to the importance of the work of Supreme Court Justice Louis Brandeis, for example. Today, there’s a whole slew of what people call neo-Brandeisian anti-monopoly policies that many are advocating. I think we do need another era of trust-busting, breaking up those concentrations of power.
I know the dynamic right now between Europe, the EU, and the US is extremely complicated and painful. I will say that I take it as good news that Europe is starting to build a path toward empowering its own tech companies. I do think that taking on the monopoly of the US firms is an important thing to do.
There’s Going to Be a Battle Between Globe-Spanning Tech Powers and Nation-States
Throughout democratic history, monopolies over information have often strengthened authoritarian rule. How different is today’s AI revolution from earlier media revolutions such as radio, television, or social media? Does generative AI represent a quantitative extension of those transformations, or does it constitute a genuinely new constitutional moment?
Professor Danielle Allen: It’s a great question. To be honest, I don’t think I really have an answer. In some ways, it’s an obvious continuation, and I’ve already mentioned that. On the other hand, the speed and scale of change are wholly new for humanity. That’s honestly the part that concerns me the most: the rapidity and the totalizing nature of it.
Take the fact that somebody like Elon Musk can, in five years, manage to have half of the world’s satellites in space. It’s probably more than half of them by now. He has the capacity to intervene directly in geopolitical decision-making among the world’s nation-states. That’s what’s fundamentally different. The earlier technologies came into existence and existed essentially within nation-states. We now have corporations with a degree of political control that puts them in a more powerful position than nation-states.
It’s not that this has never happened before. In the eighteenth century, for example, Britain faced, roughly speaking, a similar kind of problem with the East India Company. Other European powers faced similar challenges with the colonial trading empires that they first allowed to grow and then discovered had gone beyond their control. That, of course, led to the struggles in the nineteenth century between nation-state governments and those colonial companies. We are likely to see another period of that kind of struggle.
I don’t think the nation-state is doomed. I don’t think it’s doomed primarily because people still live in places. Human beings are embodied, and we are instantiated in geographic locales. Our communities matter to us, and that feature of human life, which starts from the ground and extends outward, will continue to be an important part of governance and jurisdictionality. So, in that regard, there’s going to be a battle, for sure, between globe-spanning tech powers and nation-states. I do think nation-states can win in the end. And I would argue they should win. But it’s going to be a very hard period of struggle.
Technology Should Complement Human Intelligence, Not Replace It
In “How AI Fails Us,” you criticize the dominant vision of AI as autonomous intelligence and instead advocate technologies that augment human cooperation rather than replace it. How might this alternative vision reshape democratic governance? Could cooperative AI become a tool for democratic renewal rather than democratic erosion?
Professor Danielle Allen: I appreciate that question. What really matters here is that we sometimes think the way technology is currently unfolding is somehow a matter of the laws of nature or the laws of physics—that it’s simply a natural outgrowth of nature itself. That’s a mistake, and it already gives away too much power. Technology always develops out of some kind of moral vision. By moral vision, I mean a way of understanding human beings, human capacity, and the human relationship to nature and the world.
The argument of that paper is that the moral vision behind the current technological supernova rests on a flawed premise. That premise is that there is one thing called human intelligence, that it is singular, and that machines are trying to replicate and ultimately overcome it. Whereas, in fact, human intelligence is multiple. It comes in a whole range of varieties. Machine intelligence is also multiple. That plurality of our intelligences is valuable to specific societies and to humankind as a whole.
Plurality can be activated for tribalism, conflict, and division, or it can be activated for creativity, innovation, and cultural richness. One of the most important human projects, therefore, is to give ourselves the tools to activate the plurality of our intelligences for those positive outcomes.
If you begin with that basic vision of human beings and human well-being, then you can see a completely different approach to technological development. It is an approach based on activating that plurality in a positive way, one that requires complementing human intelligences rather than trying to overcome or replace them.
Very ordinary technologies, like the Zoom technology we’re using right now, provide a good example. Zoom complements what human beings can do. We can’t see across thousands of miles, so being able to do that through Zoom supplements what we are already able to do. It doesn’t actually replace anything we were previously doing. So, one can begin to envision a host of things that would be wonderful to be able to do that are all supplemental rather than replacing our core capacities.
We Need a Formally Articulated Bill of Rights Against the Powers of AI
Authoritarian governments increasingly combine sophisticated digital surveillance with artificial intelligence, while democratic governments are also rapidly adopting AI for policing, border control, and public administration. How can constitutional democracies harness AI’s administrative potential without gradually normalizing technologies of surveillance and executive concentration that undermine liberal constitutionalism?
Professor Danielle Allen: That’s a great question. I think we’re at a point where we need a formally articulated bill of rights against the powers of AI. In the eighteenth and nineteenth centuries, as nation-states were consolidating their power, people articulated bills of rights. In the US case, we have the first ten amendments to the Constitution: rights to association, rights to expression, freedom from search and seizure without warrants, rights to trial by jury, and so forth.
All of those rights apply to the AI universe, but it is worth taking the time to elaborate on what is required to ensure that they are protected under the surveillance conditions made possible by AI. That’s a project for legal scholars—to spell out, at meaningful length, what those requirements are.
I do think the GDPR was a good model for extending those rights protections into the environment of digital data and the like. We need to keep building that edifice of crystal-clear legal protections and limitations. Again, some of the lawsuits right now against Anthropic and others, for example, for stealing authors’ works, are really important. We already have this rights-based framework, and with more investment of time and effort, we could use it much more comprehensively to cover the world of generative AI.
Generative AI Is Going to Accelerate the Capacity of Everything
Recent years have witnessed growing interaction between transnational populist movements, digital platforms, and algorithmically amplified political narratives. Do you believe generative AI is likely to accelerate the emergence of increasingly coordinated transnational populist ecosystems that operate beyond the regulatory capacity of individual democratic states?
Professor Danielle Allen: Generative AI is going to accelerate the capacity of everything. It will also accelerate the capacity of pro-democracy transnational networks to operate across borders. In that regard, it only changes the balance of power if only the populists use it.
The First Thing We Need Is Functional Democratic Governance
Much current AI regulation emphasizes safety, content moderation, transparency, and risk mitigation. Yet your work argues that these approaches address symptoms rather than underlying distributions of power. What institutional reforms would be necessary if democratic governments were to regulate AI according to constitutional principles instead of primarily technological ones?
Professor Danielle Allen: The two major things we need in the US are to get rid of corruption and restore suffrage. We have experienced an incredible erosion of voting rights in the US. Those two things—corruption, by which I mean the impact of money in our politics, and the erosion of voting rights—mean that our institutions have been captured by small oligarchic minorities pursuing their own financial interests through cronyist arrangements with tech companies.
So, when people ask that question, what they’re often wondering is, "Okay, what’s the specific regulation that you need to rein in a tech company?" What I want to say is that that’s actually operating at the wrong level. The first thing you need is functional democratic governance, where power actually lies with the people. Then the people will be able to apply regulations to tech companies.
We can devise all kinds of regulations—that’s not really the problem. The problem is whether power actually lies with the people or with the people who are in bed with the tech companies. Right now, it lies with people who are in bed with tech companies. That’s why we need fundamental reforms in the basic operations of democracy in the US to change that balance of power.
Digital Competence Matters More Than Digital Literacy
Your broader scholarship consistently emphasizes citizenship, civic identity, democratic participation, and shared constitutional purpose. In an era increasingly mediated by AI-generated information, how should democratic education evolve to prepare citizens not merely to detect misinformation but to sustain democratic judgment itself?
Professor Danielle Allen: That’s a great question. There are a lot of important pieces there. I think highly of the work of Audrey Tong in Taiwan, who argues for an education in digital competence, not just digital literacy but competence. Again, if you know how to use the tools, you’re going to inoculate yourself against being abused by people who are using them.
So that’s one thing. But, to your point about judgment, cultivating the capacity for judgment is key. That means being able to—and here I’m drawing on some work by somebody named Scott Warren—to provide accurate accounts of what people who disagree with you think and why they think it. In other words, you have to train yourself out of your own silo in order to gain access to the full picture of what people are experiencing. One way to do that is by testing whether or not you can render the best version of your opponent’s argument. That’s a very basic skill, but it is no longer very common in our systems of instruction and education.
The work of judgment also requires understanding how to relate values and principles to empirical realities. To develop those skills, you actually have to spend time thinking about values. What are your values? Where do they come from? What are their sources? How do you justify them? What are other people’s values? How do they interact with each other? The work of connecting a values-based diagnosis to a values-based solution is also a learnable skill. So, those would be three things that I would point to as important to have inside our educational offerings.
We Should Deploy Technology to Elevate Human Judgment
Finally, stepping back from current technological developments, do you believe liberal democracy is entering a constitutional transformation comparable to those produced by the printing press, industrialization, or mass broadcasting? Looking ahead, what institutional innovations will be most essential if democratic societies are to ensure that artificial intelligence strengthens rather than weakens constitutional government, political equality, and democratic self-rule?
Professor Danielle Allen: Yes, I do think we are entering another period of major constitutional transformation of that kind. I wish I could give you a list of specific changes that will help us get where we want to go, but fundamentally, we need changes that strengthen human judgment. That runs a bit against the current at a moment when so much is eroding and replacing human judgment. That’s why I advocate various election system reforms in the US, because they would strengthen human judgment.
I also think there are many ways of deploying technology to elevate human judgment rather than denigrate it. AI sensemaking tools such as Polis and vTaiwan, which are deliberative democracy tools, use natural language to allow people to express themselves in their own words, while helping to bring the themes that people have generated to the surface and making us more visible to one another. Things like this can actually help strengthen human judgment.
The hard question, which I appreciate you’re asking, is this: How do we evaluate whether a technology is strengthening human judgment or weakening it? That seems to me to be one of the criteria we need to bake into our assessments of new models, new tools, and the other technologies that emerge into the world.
