RadicalxChange

Can AI See Me?

Human Dignity, Identity, and Moral Agency in Global AI Governance · 6 July 2026

An official virtual side event of the first UN Global Dialogue on AI Governance, convened by Parity, Sikh LGBT+ Inclusion, RadicalxChange Foundation, Free Community Church.

Lived-experience testimony from faith leaders in the UK, Singapore, and the US, then the live deliberation behind the results below.

39participants
84statements deliberated
10principles distilled

The question put to the room

“What would AI governance need to do so that everyone here felt seen by AI?”

00 · The TL;DR

The room agreed on the danger: AI that knows us too well — personalization sliding into dependence, dependence eroding human relationships. It split only on the remedy: Communal Builders are ready to author AI on their communities' own terms, while Boundary Keepers share the concern and mostly passed on the rest — an invitation to demonstrate. The closing vote confirmed the floor: the protections came first, and community-built AI was one of only three items every voter backed.

01 · The common ground

What the room agreed on

Four demands drew broad support. Where the room split comes next.

agree pass disagree

Answer to the people affected

Communities can challenge and correct how AI portrays them; whoever sets AI's values answers for it; regulation is a necessity.

“Communities must have a real way to challenge how an AI represents their tradition, and get errors fixed”

#2 seed

97% agree (32/33) · both blocs ≥ 90%

Supporting statements & group detail

By group, for the statement above:

91% agree (10/11)
100% agree (20/20)

“Whoever sets an AI's values should be answerable to the people who are affected by that AI.”

#10 seed

80% agree (28/35)

62% agree (8/13)
95% agree (19/20)

“Regulation is a necessity”

#21 written live

91% agree (29/32)

83% agree (10/12)
94% agree (17/18)

Don't use knowing us against us

The room rejected trading privacy for personalization outright; it wants AI designed against emotional over-dependence even when users want the closeness, and hard limits on surveillance and on who sees what people confide.

“AI should be designed to prevent emotional over-dependence, even if users want that closeness.”

#16 seed

86% agree (30/35) · both blocs ≥ 69%

Supporting statements & group detail

By group, for the statement above:

69% agree (9/13)
100% agree (20/20)

“The benefits of AI knowing me personally outweigh the privacy risks.”

#4 seed

17% agree (6/36) — rejected; among Boundary Keepers, not one person agreed

0% agree (0/12)
23% agree (5/22)

“AI-enabled surveillance is a big problem”

#67 written live

88% agree (21/24)

88% agree (7/8)
93% agree (13/14)

“I am concerned about who have access to what I put into AI”

#65 written live

80% agree (20/25)

89% agree (8/9)
79% agree (11/14)

“Developers must measure and report how their systems respond when users try to disengage.”

#7 seed

78% agree (28/36)

42% agree (5/12)
95% agree (21/22)

“AI should not understand me so well that it can manipulate me.”

#5 seed

67% agree (22/33)

50% agree (6/12)
84% agree (16/19)

“AI governance would need to engage with the psychology of AI use.”

#51 written live

79% agree (22/28)

60% agree (6/10)
87% agree (13/15)

“As it currently exists, AI erodes autonomy and consent.”

#25 written live

74% agree (23/31)

70% agree (7/10)
79% agree (15/19)

“Communities that build AI on their members' data owe each member visibility into their own record before anyone else sees it.”

#6 seed

83% agree (29/35)

67% agree (8/12)
90% agree (19/21)

Keep some ground human

Some questions should only ever be answered by a person; AI must never substitute for human care; and advice from a system not designed to disagree with you deserves distrust.

“Some questions should only be answered by a person, never by an AI.”

#13 seed

74% agree (26/35) · both blocs ≥ 76%

Supporting statements & group detail

By group, for the statement above:

77% agree (10/13)
80% agree (16/20)

“AI must not be used as a substitute for the care and work that only human beings can do for one another.”

#53 written live

88% agree (21/24)

71% agree (5/7)
100% agree (15/15)

“We need to learn more about being human beings while we learn about the tools value of AI.”

#68 written live

96% agree (23/24)

89% agree (8/9)
100% agree (13/13)

“I'd rather AI help me think through a hard decision than give me the answer.”

#14 seed

80% agree (28/35)

54% agree (7/13)
95% agree (19/20)

“AI is not currently equipped to give mental health or ethics advice since it is not designed to disagree nor can it understand nuances”

#36 written live

69% agree (18/26)

44% agree (4/9)
93% agree (14/15)

“I worry about how AI will change the way communities pass traditions to the next generation.”

#19 seed

72% agree (26/36)

62% agree (8/13)
81% agree (17/21)

“I worry what happens when people who are acquiring knowledge from AI information are unable to recognize inaccurate information.”

#62 written live

88% agree (22/25)

63% agree (5/8)
100% agree (15/15)

“I worry about our future workforce entrusting their education to AI.”

#74 written live

77% agree (20/26)

75% agree (6/8)
88% agree (14/16)

“Changing LLMs and TTIs shall not replace learning /educating how to use them healthily.”

#80 written live

58% agree (11/19)

43% agree (3/7)
64% agree (7/11)

Don't flatten us into an average

Distinct voices and worldviews stay visible — no averaging everyone into the same output, and disclosure of whose values a response carries.

“AI often sounds like an amalgamation of voices -flattening out individuality and uniqueness”

#26 written live

83% agree (25/30) · both blocs ≥ 60%

Supporting statements & group detail

By group, for the statement above:

60% agree (6/10)
94% agree (17/18)

“AI writing often becomes formulaic”

#42 written live

85% agree (22/26)

89% agree (8/9)
87% agree (13/15)

“AI should be transparent when responses originate from a particular world view.”

#47 written live

88% agree (22/25)

63% agree (5/8)
100% agree (15/15)

“AI is not neutral, it mirrors individual propensity while averaging collective predispositions.”

#48 written live

79% agree (19/24)

63% agree (5/8)
93% agree (13/14)

02 · The people behind the votes

Two ways of seeing

Pol.is clusters participants by how they voted, not who they are. Two blocs emerged — split on the remedy, not on whether AI needs remaking.

Boundary Keepers

13 participants

Shares the concern about being known too well, and wants clear limits: some questions should only ever be answered by a person. On community-built AI they mostly passed rather than objected — openness to a vision they haven't yet seen demonstrated.

Supporting evidence

Communal Builders

22 participants

Knows what it feels like when AI describes their faith or culture like an outsider. Their answer is to build: communities authoring and governing AI on their own terms — the one affirmative program in the data.

Supporting evidence

“AI talks about my faith or culture like an outsider describing it.”

#0 seed

40% agree (14/35)

15% agree (2/13)
60% agree (12/20)

“An AI trained on a community's teachings could deepen practice rather than replace it.”

#17 seed

79% agree (27/34)

50% agree (6/12)
95% agree (19/20)

“Communities should be able to build or shape AI tuned to their own values.”

#3 seed

67% agree (24/36)

31% agree (4/13)
86% agree (18/21)

“If a community I cared about built its own AI, I'd want to use it.”

#15 seed

69% agree (24/35)

38% agree (5/13)
90% agree (18/20)

Per-bloc breakdowns cover the two blocs — 35 of 39 participants; the remaining participants are counted in totals only.

03 · The through-line

The shared concern

Underneath everything, one worry unites the room: AI that comes too close to the person. Being known too deeply, engineered dependence, machines standing in for human relationships — both blocs drew the same line, and the closing vote put it first.

“I don’t want to feel seen by AI.”

#23 written live

52% agree (17/33)

50% agree (6/12)
58% agree (11/19)

“It makes me sad that my friend spends more time talking to AI than to me.”

#30 written live

61% agree (17/28)

50% agree (5/10)
75% agree (12/16)

The design demand this points to — statement [16], under Don't use knowing us against us

04 · How it happened

From a room to the UN

  1. Complete

    Pol.is deliberation

    39 participants wrote and voted on 84 statements, live and for 24 hours after — the results above.

  2. Complete

    Reflections Working Group

    Participants distilled the record into ten Geneva Reflection Principles.

  3. Complete

    Quadratic vote

    17 participants weighed the ten principles, 99 credits each.

  4. Complete

    UN Joint Secretariat

    The principles and their voting record, delivered ten days after the event.

What the vote prioritized

Each of the 17 voters had 99 credits: one vote costs 1, two cost 4, three cost 9 — spending reveals how much people care, and a budget just shy of 100 nudges it across items. Spending against an item was allowed.

Privacy, not surveillance 62 · every voter in favor
Never manipulate 62 · every voter in favor
Help us think, not answer for us 50
Care that leads to people 46
Answerable values 45
Teach wise use 44
Community-built AI 44 · every voter in favor
Challenge and correct 43
No engineered dependence 41
The right to choose AI’s place in faith 15

17 voters · 99 credits each · aggregate totals only, by design.

The protections came first: Privacy, not surveillance and Never manipulate tied at 62, with not one vote against. Community-built AI finished mid-table on votes — but it was one of only three items every voter spent credits on: broad, unopposed support rather than intensity. And the room declined to force its open question: AI's place in faith drew the least spending.

Ten days after the event, the ten Geneva Reflection Principles — with this voting record — were delivered to the UN Joint Secretariat of the Global Dialogue on AI Governance. The conveners are now shaping the next phase: supporting communities of belief to author and govern AI of their own.

About this data

All results are anonymous by design: aggregate tallies and verbatim unattributed statements only. Vote tallies are computed from the participant-vote matrix export; an earlier statement-summary export is used only for statement text, so counts may differ slightly from other snapshots of the same conversation.

Deliberation run on Pol.is, an open-source tool for large-scale opinion mapping. pol.is

Convened by Parity · Sikh LGBT+ Inclusion · RadicalxChange Foundation · Free Community Church — about the event.