AI would act as cartographer, translator and question-maker—not judge. Language models can recognise that differently phrased contributions may share an underlying concern, suggest where two claims rely on incompatible definitions and explain specialist evidence in accessible language. Retrieval systems keep every synthesis attached to original contributions and primary sources. Multiple models or agents deliberately produce competing interpretations so that the interface never presents one machine summary as neutral truth. Participants can correct how their position is represented, inspect every transformation and refuse inclusion in generated synthesis. Human facilitators remain responsible for process legitimacy and consequential decisions.
Common Ground
“Can AI help people disagree without making them enemies?”
The beginning
Common Ground begins with a frustration: public conversation has become exceptionally good at sorting people into camps and exceptionally poor at helping them think together. I imagine a new kind of civic institution—not another social network and not an AI moderator that decides who is right. It would be a place where a difficult public question can be unfolded slowly enough for people to see the evidence, values, fears, incentives and trade-offs inside it. The purpose is not artificial consensus. The purpose is a better disagreement: one in which participants understand what is actually contested, what is merely misunderstood and where a shared next experiment might still be possible.
Most digital debate systems optimise attention, speed and visible reaction. They flatten an argument into posts, reward certainty and hide the structure underneath a disagreement. Two people may use different words for the same value, cite evidence measured at different scales or appear to disagree about facts when the real conflict concerns risk, identity or fairness. Institutions then receive a loud stream of opinion without a legible map of why people believe what they believe. Conventional surveys compress complexity into percentages; comment sections amplify performance; expert reports may be rigorous but socially distant. The missing object is a shared, inspectable model of the disagreement itself.
What it could become
The product would be a deliberation space built around a living argument map. A community, city, company or public institution begins with a consequential question. Participants contribute experiences, claims, evidence, concerns and desired outcomes in their own language. The system separates these elements without stripping away their human context, links compatible or conflicting claims, identifies missing evidence and shows how conclusions change under different assumptions. Small facilitated groups can explore the map, ask questions of absent perspectives and propose limited real-world trials. Instead of ending with a vote alone, every process produces a public reasoning record: what was learned, what remains unresolved, which values were protected and what experiment should happen next.
For whom
- Citizens participating in local decisions
- Public institutions designing policy with communities
- Employees navigating contested organisational change
- Journalists and researchers explaining complex public questions
- Facilitators, mediators and civil-society organisations
Core capabilities
- Multilingual claim, value and evidence mapping
- Source provenance and contradiction tracking
- Facilitated deliberation tools for small and large groups
- Perspective synthesis without erasing minority positions
- Experiment design, decision records and follow-up measurement
The project becomes meaningful only when a new technical possibility is translated into a clear human advantage, an experience people can understand, and a system capable of earning trust over time.
Intelligence and mathematics
The argument map can be represented as a typed, temporal graph connecting people, claims, evidence, values, assumptions and proposed interventions. Graph clustering reveals coalitions of reasoning rather than demographic tribes. Bayesian updating shows how beliefs might change when evidence changes, without pretending values are probabilities. Social-choice theory helps expose how different aggregation rules produce different outcomes. Information theory can identify which unanswered question would reduce the most uncertainty, while robust optimisation searches for interventions that remain acceptable across several plausible models of the situation. The mathematics serves legibility: it should reveal sensitivity and trade-offs, not manufacture objectivity.
How it might live
I imagine Common Ground as a public-benefit technology institution with several layers. Local communities and small civil-society groups should have access to an open essential version. Cities, public agencies, universities and large organisations could fund structured deliberations, secure deployments, facilitation and longitudinal analysis. A governance foundation would protect the method, publish audits and prevent the platform from becoming political advertising infrastructure. Revenue would come from the quality of the process and the durability of institutional learning—not from engagement, voter profiling or selling behavioural data.
For me, a venture is more than an interesting technology. It needs a narrow first user, a repeated problem, a distribution path, a credible advantage and a reason to improve as more people use it. I would test those conditions before deciding whether this idea should become a company, a product, an open technology or an ongoing research programme.
Rules for making it real
- 01
Consensus is not the goal; mutual intelligibility is.
- 02
Every synthesis must remain traceable to people and evidence.
- 03
A minority position must not disappear because it clusters poorly.
- 04
AI may organise a disagreement but cannot legitimately settle it.
- 05
Participants must be able to correct the representation of their own view.
- 06
The process should end with a testable next step whenever possible.
From question to company
- 01Frame
Choose one local question with a trusted civic partner and a bounded decision.
- 02Prototype
Co-design the contribution language and consent model with participants and facilitators.
- 03Prove
Build a transparent argument map with claim, evidence, value and assumption layers.
- 04Build
Run parallel human-only and AI-assisted deliberations to compare distortion, learning and trust.
- 05Advance
Publish the complete method, failure analysis and model-behaviour audit.
- 06Advance
Expand into a reusable civic reasoning infrastructure only after legitimacy is demonstrated.
What could go wrong
Serious imagination includes the possibility that an idea should change radically—or should not exist. These are the tensions the project would need to resolve:
- Powerful institutions using participation as theatre after a decision is already made.
- AI summaries subtly privileging dominant language, education or cultural styles.
- The map making contested moral values appear technically resolvable.
- Sensitive political beliefs creating surveillance or retaliation risks.
- Bad actors flooding the process with coordinated synthetic participation.
- Participants mistaking a legible model of disagreement for a complete model of society.