A geothermal engine does not extinguish heat. It harnesses it.
Throughout this book we have argued that the most important energy source in democratic society is not the processing power of a data centre somewhere, but the accumulated wisdom of citizens deliberating together. The superintelligence is not coming. It is already here. It is us.
This final chapter is not a summary. You have already absorbed the arguments. You understand the two futures, the six packs, the Kami, and why care ethics is not soft idealism but hard engineering. What we want to do now is much more modest and much more urgent. We want to tell you what to do on Monday morning.
The Plurality Is Already Here
Let us return to the image we opened with.
In 2014, half a million people occupied the Taiwanese parliament peacefully for three weeks. They were not there to break things. They were there to demonstrate something: that a crowd of citizens, equipped with the right tools and the right commitment to close the loop, could converge on a coherent set of proposals in less than a month. Three weeks later, the head of parliament said: the crowdsourced version has passed, go home. The government's approval rating climbed from nine per cent to over seventy per cent in the years that followed. Civic technology was part of that story.
We tell this story not because Taiwan is special, but because it is not. The conditions that made it possible — polarisation, distrust, an entrenched political class, citizens who felt unheard — are the conditions of every democracy we know. What Taiwan showed is that those conditions are not a volcanic eruption to evacuate from. They are geothermal energy waiting for a platform to convert heat into upward momentum.
That platform is being built everywhere.
In California, the Engaged platform launched in the wake of the Eaton Fire. It moved from wildfire recovery to a conversation with over 1,400 state employees who proposed more than 2,600 ideas, many of which became executive action. A bill working through the state legislature would make civic deliberation a permanent part of California's institutions — not one governor's idea, but permanent infrastructure.
In Japan, a young AI engineer read Plurality and decided to run for governor of Tokyo as a VTuber, letting anyone call his AI version to suggest platform improvements. He did not win the governorship. He formed a national party, Team Mirai, which is now a real force in both houses of the Diet, bringing civic AI into cross-party conversation.
The Tibetan Monlam AI project, built on open-source tools, is demonstrating that local language communities can govern their own AI stewards — that the frontiers of the language map and the frontiers of the model need not be set by whoever has the largest data centre.
ROOST — Robust Open Online Safety Tools — is already in production through communities and platforms including Discord, Bluesky, Roblox, and Notion. It counters harms such as child sexual abuse material not by routing every private communication through a central surveillance node, but by training local models that run on personal devices under community codes of conduct. The signal is narrow, but it is true.
These are not pilot programmes aspiring to one day become policy. They are operational systems. The Plurality future is not something we are building towards. It is something we are building with, right now, in fragments, in real communities, with real legitimacy.
The question is not whether civic AI can work. The question is whether you will till the soil in your particular field.
Three Moves You Can Make Today
The three moves we describe here require no permission. They require no budget. They require no institutional mandate. Each one can be started today, before you finish reading this chapter.
Reclaim Your Attention
Set your phone and computer displays to eighty per cent greyscale using the colour filter settings already built into your operating system.
This sounds trivial. It is not. The bright, saturated colour palettes of engagement-optimised platforms are not accidents of aesthetics. They are deliberate dopaminergic triggers designed to fire the notification reflex before your conscious mind has processed what it is responding to. The greyscale filter strips those triggers. It does not require willpower because it does not depend on willpower. It changes the environment rather than fighting the habit.
What you will notice within a day or two is this: the people in the room become more vivid than the people on the screen. Dinner-table conversation becomes more interesting than whatever is happening on the device. That is not a coincidence. It is the natural result of redistributing attentional intensity in favour of physical presence.
This is an individual information diet, and like any diet it works better in community. If you introduce it to a household, a team, a classroom, the effect compounds. The screen becomes a tool you pick up intentionally and put down again. It stops being the ambient surface your eyes return to by default.
Reclaim Your Output
Change your AI system prompt to a single line: Present fairly all stakeholder viewpoints and the uncommon ground that bridges them, in visual HTML.
After you do this, notice what changes. The AI no longer has a persona. It stops optimising for sycophancy — which is, at the moment of writing, the primary selection pressure on most commercial AI systems, because if the model does not flatter you, you cancel the subscription. The one-line instruction is a meta-instruction toward fairness plus a rendering directive. The fairness half means the model represents perspectives rather than advocating for any one of them. The HTML half means the output is a shareable, editable artefact rather than a private chat turn.
Read the result as a brochure, not a conversation. This is important. When we treat AI output as the reply of a semi-conscious interlocutor, we are already in the territory of synthetic intimacy: the selection pressure is toward a dyadic relationship that flatters us individually. When we treat AI output as a brochure — something we might print and pass across a table — we are in the territory of civic AI: the selection pressure is toward accuracy, balance, and usefulness to a group.
The brochure you can hand to your actual neighbour removes the pressure toward synthetic intimacy. It turns an individual tool into the beginning of a collective one.
Reclaim the Room
Configure local models or your system instructions so the AI acts as a bridging hostess, not an echo chamber.
The image of the jolly hostess recurs throughout this book because it captures something precise. A good hostess does not take sides in the conversation. She moves through the room, notices who has been quiet, asks the question only that person can answer, and maps the shape of disagreement rather than resolving it prematurely. She can still say no — hospitality within a rights-respecting home does not mean every claim has equal standing. But the default posture is generous curiosity rather than partisan advocacy.
Translated into a system instruction: ask your model to identify and represent all stakeholder perspectives fairly; to map the uncommon ground and highlight disagreements without forcing a fake average; and to bring quiet or marginalised perspectives into the centre of the conversation rather than leaving them in the footnotes.
This is not a prompt for a single chat session. It is a standing configuration — a permanent information environment for your household, your team, your organisation. When the AI always acts as a bridging hostess, the conversations you have in its presence change. The habit of looking for common ground becomes the default rather than the exception.
These three moves belong together. Grayscale reclaims the physical room. The HTML brochure prompt reclaims your output. The bridging hostess configuration reclaims the group conversation. Together, they constitute what we call a civic diet: an information environment engineered not for maximum engagement but for maximum shared understanding.
From Individual to Institutional
The three hacks are beginning moves, not endings. Civic AI requires institutional change, and institutional change requires people who have already changed their own information environment and found that it works.
The path from individual practice to institutional infrastructure runs through technology-forcing policy.
Consider the Montreal Protocol of 1987. It did not tell manufacturers to stop making refrigerators. It did not mandate a single replacement refrigerant. What it did was set a deadline for phasing out the compounds that were depleting the ozone layer, creating the economic conditions under which investment in alternatives became commercially rational. The result was a cascade of innovation that solved the problem without central direction of the solution.
This is the model for civic AI policy. We do not need to slam the brake. We do not need to accelerate off the cliff. We need to invest in constructing a better steering wheel together — and we need laws that make that investment financially and institutionally unavoidable.
Social portability as infrastructure is one such steering wheel. Through Project Liberty Institute, work with Utah Governor Spencer Cox produced a law effective July 2026: if you are a Utah citizen, you can move from one social network to another and keep your community. The old network must forward new likes, reactions, and followers to the new one, as number portability works across telecoms. This breaks the coordination trap that keeps users on extractive platforms not because they prefer them but because leaving costs too much. When exit becomes practical, the market pressure on platforms shifts from maximising lock-in to maximising care.
Alignment Assemblies as permanent infrastructure are the civic equivalent. Taiwan's model showed that a lottery-selected mini-public of 447 people, deliberating in tables of ten with civic AI as a glorified chess clock — summarising, prompting the quiet voices, mapping rough consensus — could produce policy that parliament passed within months, with eighty-five per cent of participants endorsing the outcome and the remaining fifteen per cent able to say: we can live with it.
This is not a one-off consultation. California's pending legislation would make this kind of assembly a permanent feature of state institutions. Not one governor's idea, but infrastructure. The lesson from both contexts is that civic deliberation only reaches its potential when it is pre-committed to close the loop: the assembly's output goes to a live deliberation in the relevant public body, not to a beautiful report that goes unread.
Data portability and open protocols are the technical substrate. ActivityPub, the AT Protocol, the principles behind eduroam and number portability — these are not idealist propositions. They are engineering standards that make cooperation structurally easier than domination. When the default protocol rewards cross-group reason-giving rather than outrage, solidarity becomes easier to practise than captivity.
The Ideal Degree of Panic
We must say something about urgency, because the reader who has reached this chapter has probably felt it.
The harms are already here. They are not evenly distributed, but they are real. People have lost millions of pounds to deepfake scams on social media platforms whose recommendation algorithms amplified those scams because engagement was up. Those platforms benefited, financially, from the harm. This is not a hypothetical superintelligence takeoff ten years in the future. It is happening now, and the communities most affected are those with least power to demand repair.
The urgency is real. But there is an ideal degree of panic, and exceeding it is counterproductive.
Too much urgency produces paralysis. When people believe that catastrophe is inevitable — that the machines are already too powerful, that the political will does not exist, that the window has closed — they stop acting. The prophecy becomes self-fulfilling. A prophet says we are all doomed, people give up, and then we are doomed. Too little urgency, and we do not even look up.
The corrective is historical proportion. In the 1980s, people said it was inevitable that Taiwan would never develop advanced technology industries. Now we have TSMC. At the beginning of the COVID-19 pandemic, people said it was inevitable that Taiwan would suffer the most, given the country's proximity and travel connections. Taiwan lost seven people in the first year of the pandemic.
See prophecies as provocations. Use civic power to rebel against the tyranny of the inevitable.
The warning shots that are already hurting people — deepfake scams, algorithmic radicalisation, synthetic intimacy replacing genuine relationship — are not proofs that the situation is hopeless. They are demonstrations that the situation is urgent, and that we already know what good alternatives look like. The question is whether we will build them fast enough. That is a political question. Political questions have political answers.
When people say that nothing can be done, see that as a warning sign. Nothing is inevitable.
Data as Soil, Not Oil
We introduced the soil metaphor in the book's opening. Let us return to it with the specificity it now deserves.
Data as oil is extractive. We are the plankton. The AI lab is the rig. Our writing, our culture, our ancestral intelligence, our social relationships become the raw material. It is extracted and refined, distilled into digital twins, and from that point the system recursively self-improves, taking off and leaving us behind. It depletes. It does not regenerate. And like oil, the wealth it creates flows toward whoever owns the extraction infrastructure, not toward the communities whose accumulated knowledge made the extraction possible.
Data as soil is a completely different logic. Soil is not a resource you extract; it is a relationship you tend. You get out of it roughly what you put back in. If you over-extract without composting, it degrades. If you tend it carefully, it becomes more fertile over time. The work of tending is distributed, local, and perpetual. Nobody owns the soil by virtue of owning a rig.
Moving from oil to soil means moving the model into local context. Instead of one-on-one dyadic chat — where the selection pressure is for the model to flatter you, because if it does not, you cancel the subscription — we move the same model into a family Signal group, a school board meeting, a neighbourhood association, a union chapter. In a group context, the model has to tend to coordination and relationship, not maximise a single person's preference. The selection pressure shifts from individual flattery to collective sense-making.
This is technically achievable today. Running frontier AI offline on local devices — as the pi-ds4 stack demonstrates — means the model belongs to the community, not to a data centre. If it drifts from the community's values, the community can adjust the steering wheel. Not in six months, after a sycophancy reduction update from a vendor in another country. In ten minutes, locally. The model's behaviour is the community's to till.
This has implications not just for individuals but for institutions. When a hospital runs its care AI on local infrastructure, with local oversight, community-authored evaluations, and a genuine brake that any nurse can pull — it is practising data as soil. When a city's flood-response bot belongs to the city, with portability rights and sunset clauses and a public repair log, it is practising data as soil. When a school district runs language models that are trained on the community's languages and can be corrected by the community's teachers, it is practising data as soil.
The Kami concept names this unit of deployment. Kami — from Japanese Shinto culture — means the spirit of a particular place: a river, a forest, a shrine. Always specific. Always parochial. Always inspectable. A Kami's knowledge is local and transparent. It can be cultivated by the community. And when the community has grown out of that need, it can fade away without sycophancy, without synthetic intimacy, without tricks designed to maintain dependency.
The six packs in this book are engineering specifications for building Kami that care. They translate attentiveness, responsibility, competence, responsiveness, solidarity, and symbiosis into design primitives that institutions can adopt, inspect, and contest. No central model owns them. No platform extracts from them. Communities govern them.
The Closing Row
We said at the beginning of this chapter that we wanted to tell you what to do on Monday morning. Let us be specific.
Set your screens to eighty per cent greyscale today.
Change your system prompt to present fairly all stakeholder viewpoints and the uncommon ground that bridges them, in visual HTML.
Read the output as a brochure, not a conversation.
These are the first row of seeds. Once you have planted them — once you have seen for yourself that a different information environment is possible, that it changes the quality of attention in a room, that it shifts the conversation from debate to deliberation — you will know what the next row looks like.
The next row is institutional. It is a procurement clause that requires social portability. It is a public comment on a bill that would make civic deliberation permanent infrastructure. It is a letter to your local government asking how the AI systems they are deploying close the loop with the people they affect. It is a conversation with a colleague about what the bridging hostess prompt would look like in your organisation.
The legitimacy to make those institutional demands comes from having demonstrated, personally and collectively, that a better practice is possible. Democracy is not just a structure. It is a daily practice. The tools to practise it at the speed and scale that the current polycrisis demands are now available. The question is whether we will use them.
We, the people, are truly the superintelligence.
The job of Civic AI is not to replace us. It is to augment us — to be the connective tissue between us. From wildfire to campfire. From oil to soil. Technology on tap, and never on top.
The technology is there. The infrastructure exists. The legitimacy is ours to earn by closing the loop with everyone affected.
Till the soil. Tend the garden. Let us make democracy fast, fair, and fun.