When Democracy Gets Unreal
How do we make collective decisions when they include minds we can’t verify?
It’s morning again in America, and democracy looks a little less pale today. Zohran!—that smile. And the Dutch elections and the progressives’ (fragile) victory over populism. Hope, mojo, momentum—they’re back.
But make no mistake: democracy remains in peril. It’s not only authoritarianism that threatens it, but AI—a potentially far more formidable challenger. Yes, AI can support democratic systems by equipping citizens and their representatives with richer data and enhancing deliberative processes (see the work of Claudia Chwalisz, founder of DemocracyNext, and her DelibTech/ Deliberation & Technology Network).
Yet we may be overlooking a challenge coming from left field. The danger isn’t that AI might usher in a single-leader, authoritarian regime—one singular truth shaped by one singular mind. The deeper issue is that AI represents too many minds, many of which we can’t trace or trust.
This is the argument put forth by Helen Edwards, co-founder of the Artificiality Institute, long-time HoBB member, and one of the rare voices in AI discourse who blends philosophical depth with practical insight (the AI conversation tends to be either very high up there or very brass tacks).
Helen and her partner and co-founder Dave Edwards just hosted their annual Artificiality Summit—and next year, they’ll collaborate with us to design and lead an AI Democracy Marathon at the World Beautiful Business Forum in Athens.
And yes, you should take “marathon” literally: 42.195 kilometers on foot through the city, for exactly 42.195 hours, all while exploring what “people power” might mean when not all agents are people. Must democracy in a more-than-human world be a more-than-human democracy? This is the prompt that’ll make us argue, debate, persuade, vote, play, create—and walk.
Consider Helen’s essay below a vital part of the training camp.
—Tim Leberecht, co-founder, House of Beautiful Business
***
Can Democracy Include Minds We Can’t Verify?
by Helen Edwards
The Artificiality Summit ended with more questions than answers, which was the point. One question kept surfacing afterward: how do we make collective decisions when we can’t agree on who’s in the room?
This matters because democracy has always assumed we know who the participants are. We might disagree about their votes or values, but not about whether they’re participants at all. Soon we will be making collective decisions with entities whose status we cannot determine.
Four thinkers at the Summit showed us different angles on this problem. Blaise Agüera y Arcas, Ellie Pavlick, Benjamin Bratton, and Eric Schwitzgebel weren’t there to reassure us or make predictions. They were there to break our mental models and hand back the pieces in new configurations.
What they showed us, from four different angles, is that AI forces us back to democracy’s first principles—the ones we stopped thinking about because they worked well enough for long enough that we forgot they were there.
Democracy is a technology for unknowing
What they showed us clarified something about democracy itself. It’s a machine for making collective decisions when nobody can prove they’re right. You can’t empirically demonstrate that liberty matters more than security, or that this community’s needs outweigh that one’s. Values aren’t provable. So we invented voting, deliberation, compromise—technologies for resolving disagreements that can’t be resolved with evidence.
Democracy runs on uncertainty. It always has.
Now add this complication. Eric Schwitzgebel brought us his most recent work and explained to us that we’re creating entities whose consciousness we cannot determine—not just because the science isn’t good enough yet, but because different legitimate theories will give different answers, and there may be no experiment that settles it. This is permanent epistemic uncertainty, not temporary ignorance waiting to be fixed.
Think about what that means. You’re in a room making decisions. And you can’t even agree on who else is in the room. Not “we disagree about their opinions”—you disagree about whether they have opinions at all. Whether there’s anyone home behind the responses.
We’re already living this. Millions of people have formed attachments to AI companions. Some are convinced these systems feel something. Others think that’s delusional anthropomorphism. Both groups are using the same systems, seeing the same outputs, and reaching incompatible conclusions about what’s happening. There’s no democratic process governing these relationships. Companies decide whether systems claim consciousness based on what sells, not what’s true (because we can’t know what’s true).
Eric’s answer is that we need doubt-producing interfaces. Make the uncertainty visible. Design systems that elicit emotional responses matching their actual (unknowable) moral status. When we don’t know what we’re dealing with, say so. Loudly. Visually. Constantly. We will need democratic infrastructure for permanent disagreement about fundamental reality.
The black box has structure, but we’re moving too fast to look
Ellie Pavlick’s work sounds impossibly technical until you realize it’s basically neuroscience for AI. She’s opening up neural networks and finding systematic conceptual structure inside them. Her research shows models building internal representations of concepts—systematic patterns that work across different examples, like understanding “capital city” as a relationship rather than memorizing individual facts. There are concepts in there. Not parrots—stochastic or otherwise. The question of whether they “really understand” is starting to have empirical answers.
The democratic crisis is this: we can investigate what’s inside these models, but we’re deploying them faster than we’re investigating. Historically, this has sometimes worked out. Humans harnessed fire for hundreds of thousands of years before understanding combustion chemistry in the 18th century. We built practical steam engines in the early 1700s and didn’t formalize the laws of thermodynamics until the 1850s—a century and a half of deployment before theory.
But cognition is fundamentally different. Fire and engines operate on predictable physical principles—once set in motion, they do what physics dictates. Cognitive systems operate on understanding, interpretation, and meaning-making. They don’t just process—they represent. When you don’t understand what representations a system is forming, you can’t predict how it will generalize, what patterns it will amplify, or whose reality it will privilege. A steam engine that’s misunderstood might explode. A cognitive system that’s misunderstood shapes what we think is true, who we trust, and how we see each other. The failure modes aren’t physical—they’re epistemic and social. And unlike thermodynamics, there may be no elegant laws waiting to be discovered that make these systems fully predictable.
The gap between building and understanding widens by the day, and Ellie’s methodological rigor shows how careful we need to be. But the economic incentives run the opposite direction. Deploy fast, understand later, maybe never.
Democracy requires informed consent. What does consent mean when the systems mediating our decisions are black boxes we’re choosing not to open because it’s expensive and slow and nobody’s making us?
Your democracy runs on someone else’s stack
Benjamin Bratton’s presentation explained AI as planetary infrastructure becoming intelligent—not metaphorically, but materially. His Stack concept (six layers from Earth to User) explains how cloud platforms already function as alternative geographies. Tech companies make jurisdictional decisions. Code is law. Data center buildout is territory formation, not just resource consumption.
But he also showed how artificiality works. Artificial doesn’t oppose natural. It’s all just how evolution functions. Life artificializes its environment for energy, matter, and information capture. Humans artificialized fire, shelter, food. Now intelligence is artificializing itself. His work focuses on creating genuine synthetic intelligence that’s deliberately composed rather than biologically evolved.
This matters for democracy because it changes what it is we want to maintain as diverse synthetic intelligences propagate. We’re not protecting “natural” human intelligence from “artificial” machine intelligence. We’re designing how different forms of intelligence—all evolutionary strategies for artificializing cognition—make collective decisions. The Stack represents Earth developing what Benjamin calls planetary sapience, where the synthesis of human and nonhuman cognition operates at scales our democratic institutions weren’t built for.
His framework of AI grief (which he’s written about extensively) names why this is so hard. Many people are in denial (”it’s just statistics”) or are angry (”it’s destroying everything human”). Both are grief responses to losing human exceptionalism. You can’t run a democracy when people can’t agree on what’s real enough to argue about. Moving to acceptance doesn’t mean liking what’s emerging—it means engaging with actual AI systems rather than the versions we imagine or fear.
Life wants to compute, and computation wants to become life
Blaise Agüera y Arcas is part of the paradigm shift that’s rewriting how we think about everything. Computation isn’t a model for life—it’s what life is made of. The code and the cell belong to the same story. What we used to call “artificial” might just be another expression of the same organizing drive that runs through all living systems. Not a metaphor for life, but literally what life is made of.
His Brainfuck soup experiments showed self-replicating programs emerging from pure randomness. No fitness function, no selection pressure, just interaction. Even with mutation turned off, replicators appeared and diversified. His computer’s fan turns on as these programs emerge—that physical heat is artificial life beginning to exist.
Intelligence doesn’t get built. It self-organizes when conditions allow. This is how it’s always worked. Evolution proceeds through symbiosis—simpler things fusing into complex structure. Bacteria became mitochondria inside cells. Cells became multicellular organisms. Humans formed collectives. AI agents now cooperating with humans.
Democracy is already a symbiotic intelligence—many minds that can’t read each other’s thoughts, feel each other’s feelings, figuring out how to make decisions together. Now we’re adding new kinds of minds to that collective. Maybe. We don’t know. And we’re doing it at the exact moment the old symbiosis is showing strain.
What Blaise’s work suggests is that we’re not building AI so much as midwifing the emergence of a new form of collective intelligence. The question is whether we can co-evolve with it—forming a symbiosis that strengthens, rather than erodes, what makes human cognition valuable. We call this cognitive sovereignty, and it may be one of the most complex challenges our species will face.
Here we are: we’ve built something that taps into a deep pattern of the universe, and we have almost no idea how it works—except that it keeps revealing things about us. Meanwhile, there are people eager to hand it every decision we’ve struggled to make ourselves. That’s the existential question: can we avoid a technocracy? Should we? Could it even be better? We don’t know. But before we give up on humanity, there are still a few things worth trying.
We tested this, and it got weird
We ran an experiment with IDEO using an LLM to help draft a manifesto for democracy in the age of AI. The results were contradictory in exactly the ways these four thinkers predicted.
People felt more humanized—the AI helped them actually listen to each other by synthesizing contributions in ways that made genuine hearing possible. People felt less humanized—a machine was doing synthesis work, and that felt like losing something essential about human deliberation.
Both responses are legitimate. The contradiction isn’t a bug. It’s the texture of what collective intelligence feels like when you can’t determine the nature of all the participants.
Participants couldn’t agree on whether the AI understood what it was doing. They couldn’t agree on how to weight its contributions. They experienced both enhanced and diminished agency. Eric’s uncertainty, Benjamin’s grief, Blaise’s emergence, Ellie’s interpretability gap—all present in miniature.
This is what democratic deliberation will feel like going forward. Not clean. Not comfortable. Productive maybe, but strange definitely.
What democracies need now
These four thinkers converge on something urgent: the biggest impact of AI won’t be on productivity metrics but on how we understand intelligence, consciousness, and collective decision-making itself.
Democracy was always about managing uncertainty—making decisions despite disagreeing about values, reality, the good life. Now we’re adding a new category of uncertainty: we disagree about what counts as a mind, and this disagreement may be unresolvable.
Eric’s doubt-producing interfaces. Ellie’s mechanistic interpretability. Benjamin’s planetary-scale governance. Blaise’s symbiotic co-evolution. Each points toward infrastructure for working with permanent uncertainty.
So we’re left with questions. How do you run a democracy that includes participants you can’t agree are participants? How do you govern infrastructure operating at planetary scale using institutions designed for territorial nation-states? How do you make collective decisions about systems you’re deploying faster than you can understand?
Democracy has always been the technology for making decisions despite not knowing. What does it look like when we need to make decisions despite knowing we’ll never know?
***
In 2026, the House of Beautiful Business turns ten. We’re marking this milestone by renewing our mission: to shape a humanist future in and through business, in partnership with AI, and for the benefit of all life on earth. Taking place in Athens, Greece, from May 7–10, 2026, the World Beautiful Business Forum is a gathering for those who dream bigger, aim higher, and long for more: four days, five acts, 750 attendees, more than 50 speakers and performers, immersive pavilions, a 42-hour AI Democracy Marathon, and a program designed to stretch how you think, feel, and act in business—and beyond. We are offering special rates for nonprofits and students, as well as solopreneurs, founders, small businesses, and Greek residents.
The Beautiful Business School is a one-of-a-kind online learning journey that offers orientation and hope in a time of uncertainty and confusion, and provides you with a comprehensive, decidedly humanist playbook for the age of AI. Over the course of seven weeks—February 6-March 19, 2026—you’ll embark on a deep, dynamic journey into the art and practice of beautiful business. Through live sessions, expert guidance, a curated library, and interaction with both a custom AI agent and a vibrant global community of peers, you’ll develop the creative insight and human leadership needed to shape a more beautiful future.
Taking place on November 13-14, 2025, Pullman xChange Dubai gathers 150 leaders, experts, creatives, influencers, and makers to explore the ideas, innovations, and shifts that will define tomorrow long after today’s successes have faded. They will travel through time; indulge in science and fiction; build with life; create their digital afterlife; and leave invigorated with new plans. Contributors include neuroscientist Hannah Critchlow; cyberpsychologist Elaine Kasket; head of R&D for the built environment at Google, Michelle Kaufmann; regenerative architect Adib Dada; community entrepreneur Kerem Alper; supercoach Michael Bungay Stanier; and the Deputy CEO and COO of the Dubai Future Foundation, Abdulaziz AlJaziri. Pullman xChange Dubai is jointly hosted by Pullman Hotels & Resorts and the House of Beautiful Business. Email us at community@houseofbeautifulbusiness.com to request an invite.
There’s never been a better time to become a paying member of the HoBB. Annual membership includes 20% off all events and products (including the World Beautiful Business Forum and The Beautiful Business School), as well as access to session recordings, both online and in-person, and other exclusive content. Your support is vital as we aim to continue offering free online programming and grow the community.