This report captures the views of 62 participants at the Civic AI Conference 2026 at Rhodes House, Oxford. Every statistic, chart, and narrative is bound to the same Polis snapshot.
Each dot is a statement; position is deterministically calculated from the overall balance and cross-group spread.
What We Agree On
Across all three opinion groups, participants found broad agreement on most statements. The strongest common ground concerned human contact, disclosure, accountability, and community voice.
AI in care homes should free up time for staff to spend with residents, not replace human contact. [#0]
People living with dementia deserve a say in whether AI tools are used in their care. [#1]
When AI helps make a decision about your benefits, housing, or health, you should be told. [#4]
If an AI system treats people unfairly because of their race, age, or disability, someone should be held responsible. [#5]
People should always be able to reach a real person when dealing with a government service, even if AI handles most tasks. [#6]
The UK should create an independent body with real power to shut down harmful AI systems. [#7]
People are being harmed by AI-driven decisions while the government takes too long to act. [#8]
Workers who lose their jobs because of AI should get real help finding new work, not just advice. [#9]
Communities should have a voice in deciding how AI is used in their local schools and hospitals. [#12]
Schools should teach children to question what AI tells them, not just how to use it. [#13]
When I talk to a chatbot or AI assistant, I should always be told it is not a real person. [#15]
Big technology companies care more about profits than about what happens to our communities. [#16]
A community that takes care of its people matters more than one with the most advanced technology. [#17]
AI companies should have to pay artists and writers when they use their work to train AI systems. [#18]
Where Opinions Differ
A smaller set of statements revealed meaningful divisions between the three opinion groups. These are the tensions worth exploring further.
Using AI to keep an elderly person company when no human is available is better than leaving them alone. [#2]
Families caring for someone with a serious illness should be offered AI tools to help, even if those tools are not perfect. [#3]
Companies should not be allowed to replace workers with AI unless they help those workers find new roles. [#10]
New AI data centres in Oxfordshire will create good jobs for local people, not just for tech workers from elsewhere. [#11]
AI tools in schools do more to help struggling students catch up than they do to harm learning. [#14]
We should slow down on AI until we better understand what it does to people. [#19]
Three Perspectives
Groups are formed from voting patterns, not demographic data. The narratives below are generated locally from the same evidence; every number and citation is added by the deterministic renderer.
Each dot is a participant. A, B, and C are explicit group assignments; grey U is unclustered. Distance reflects voting similarity, not demographics.
Group A: Innovation-Ready (17 members)
The most comfortable with AI adoption. This group is notably less concerned about slowing down and does not believe companies should be forced to help displaced workers. They support AI companionship for elderly people but are sceptical of offering imperfect AI tools to families.
- 6% agree — Companies should not be allowed to replace workers with AI unless they help those workers find new roles. [#10] (B 67%, C 86%)
- 47% agree — AI tools in schools do more to help struggling students catch up than they do to harm learning. [#14] (B 8%, C 0%)
- 31% agree — We should slow down on AI until we better understand what it does to people. [#19] (B 79%, C 100%)
- 79% agree — Using AI to keep an elderly person company when no human is available is better than leaving them alone. [#2] (B 40%, C 83%)
Group B: Supportive Reformers (21 members)
The largest group. They favour offering AI tools to families, even imperfect ones, and want strong worker protections. They believe AI companies should compensate creators and are open to AI with guardrails.
- 94% agree — Families caring for someone with a serious illness should be offered AI tools to help, even if those tools are not perfect. [#3] (A 15%, C 0%)
- 100% agree — Workers who lose their jobs because of AI should get real help finding new work, not just advice. [#9] (A 50%, C 100%)
- 100% agree — AI companies should have to pay artists and writers when they use their work to train AI systems. [#18] (A 77%, C 86%)
- 79% agree — We should slow down on AI until we better understand what it does to people. [#19] (A 31%, C 100%)
Group C: Principled Sceptics (7 members)
The smallest but most decisive group. They take strong, clear positions and are the most cautious about AI. They want to slow down, demand strict regulation, and reject the notion of offering imperfect AI tools.
- 100% agree — We should slow down on AI until we better understand what it does to people. [#19] (A 31%, B 79%)
- 0% agree — Families caring for someone with a serious illness should be offered AI tools to help, even if those tools are not perfect. [#3] (A 15%, B 94%)
- 0% agree — AI tools in schools do more to help struggling students catch up than they do to harm learning. [#14] (A 47%, B 8%)
- 100% agree — The UK should create an independent body with real power to shut down harmful AI systems. [#7] (A 60%, B 79%)
Replay and provenance
The three group narratives use the 45 explicit participant-group assignments. The source summary records 46 voters in conversation, while the participant export contains 45 assignments, so the report does not silently treat those counts as identical.
Snapshot 39b1eed77ad2caed03c11d38b7f5730b5008f9df14e4c7617dc3b2854b740db0; model run gemma4:12b-it-qat-mtp · 77136510022a · Ollama 0.32.1. Each locale is generated twice sequentially at batch size one and written only when the canonical JSON is identical.
Open the full sensemaking view built from this same snapshot · Pinned source data
