Executive summary
AI alignment concerns whose purposes AI serves, whose experience informs its design, and who can challenge its effects. Open Eye One proposes that people should have meaningful influence over AI goals, deployment, acceptable risks, and the distribution of benefits. Participation should begin with the presumption that everyone belongs, supported by free and accessible ways to contribute.
AI could help people examine evidence, understand disagreements, and develop alternatives. It should support human deliberation under independent oversight. Rapid collection and synthesis of views must leave room for reflection, dissent, and correction. Human dignity, essential care, privacy, and autonomy must constrain every outcome, including one supported by a large majority.
This paper proposes a direction for research and public discussion, not a finished voting system or a claim of community consensus. Its starting proposal gives each person one vote and each company one vote regardless of size. Preventing manipulation, defining eligibility, and deciding how those votes translate into authority remain essential design questions.
Alignment needs a public voice
Imagine a community learning that an AI system will influence access to healthcare, employment, or public services. An invitation to comment offers little influence if the goals, supplier, and deployment decision are already settled. Meaningful participation starts while choices can still change and continues after deployment, when consequences become visible.
Our position is that people should help shape what AI is for, where it may operate, which risks are acceptable, and how its benefits and burdens are shared. Technical expertise is necessary to assess feasibility and danger. Lived experience is necessary to understand what a decision means to those affected. Neither should become a reason to exclude the other.
Participation must come with a clear account of what people can influence, who makes the final decision, and how contributions affected it. The OECD guidelines identify inclusion, accessibility, and feedback about the use of public input as important parts of participation processes. [1] We propose applying those expectations throughout AI governance.
Begin with everyone included
The starting question should be how everyone can participate. Begin from inclusion and work through the obstacles, rather than choosing a narrow group and asking whom else to invite. That includes people who lack technical knowledge, money, reliable internet, spare time, or confidence in formal meetings.
Contribution should be free. Public terminals, a toll free telephone service, accessible websites, translated materials, and supported community conversations are possible routes into the same process. They are proposals to develop and test, not services Open Eye One currently operates. Participants should be able to speak in ordinary language, review how their contribution was recorded, and correct mistakes.
An open website alone cannot demonstrate that a cross section of the community has been heard. Organizers should examine participation gaps, support those facing barriers, and publish the limitations of the resulting picture. Outreach should invite participation without pressuring people to disclose private circumstances or surrender their right to abstain.
Universal inclusion also requires a definition of the relevant community. A local deployment may affect residents, workers, visitors, and people elsewhere whose data is used. Future generations cannot vote today. Eligibility, representation, and ways to consider absent interests need explicit discussion before any result is treated as authoritative.
Equal weight and the company vote proposal
Our starting proposal is one vote per person and one vote per company, regardless of company size. Revenue, donations, technical capacity, and advertising expenditure should not purchase additional formal voting weight. A company may contribute evidence and operational knowledge without gaining votes in proportion to its wealth.
Under this proposal, a person who owns several companies could participate personally and through those companies. That consequence is intentional in the initial formulation. It nevertheless creates an unresolved vulnerability: an actor could establish or coordinate many entities to multiply influence. A rule that seems harmless at small scale must be tested against organized abuse.
Before implementation, the process would need verifiable company eligibility, disclosure of common ownership, controls against fabricated entities and automated submissions, and an agreed response to coordinated manipulation. Limiting votes by common control is one alternative for community consideration; it has not replaced the starting proposal. The treatment of nonprofits and other organizations also remains open.
Company participation in this proposed consultation framework should not be read as an established entitlement to vote in public elections. Whether organizational and individual votes should be combined, reported separately, or assigned different roles must be resolved transparently before the framework informs binding public decisions. Evidence should be judged for its quality and relevance, separately from counting preferences.
AI as a mediator under human oversight
AI could help organize submissions, identify shared concerns, compare evidence, and explain the strongest arguments on different sides. It could also explore alternatives that participants have not yet considered. Contributors should be able to inspect the synthesis, follow its sources, and challenge an omission or misleading interpretation.
Our approach draws inspiration from a Wesleyan search for a third way, understood here as a willingness to look beyond entrenched alternatives for a response that better serves people. This is our contemporary interpretation of an influence, not an attribution of an AI governance scheme to John Wesley. Christian and secular participants should be equally able to examine and challenge the resulting proposals.
Mediation should preserve genuine disagreement. A summary must distinguish agreement, conditional support, unresolved objections, and uncertainty. It should never describe silence as consent or manufacture consensus by removing inconvenient minority views. Sometimes no acceptable agreement will be available; the record should say so.
We propose comparing alternatives against human outcomes developed in our dignity paper: preservation of life, access to essentials, autonomy, community connection, and opportunities for creative flourishing. These measures still require definition and testing. World Happiness Report life evaluation measures offer one possible input, but its rankings are not a ready made governance score or a substitute for rights. [2]
Fast access with time to reconsider
The ambition of near instant participation is worth pursuing. People could contribute when an issue arises and receive an understandable account of emerging views. Speed should reduce the burden of participation while allowing enough time for evidence, reflection, and challenge.
We propose the following decision cycle for development and testing:
- Define the question, affected community, decision authority, and human rights constraints before seeking a result.
- Open accessible contributions and evidence gathering, with active attention to participation gaps.
- Publish an inspectable synthesis, competing options, uncertainties, and minority objections.
- Allow a proportionate cooling off period, challenges, and revisions before a final decision.
- Record the authorized human decision and reasons, including the response to public contributions.
- Monitor outcomes against agreed measures, with appeal, correction, pause, and exit mechanisms.
Thresholds for agreement, minimum participation, and treatment of abstentions must be published in advance. A narrow or unrepresentative response cannot honestly be presented as universal consent. Higher consequence decisions need stronger evidence and safeguards than routine administrative choices.
Public input can help shape laws and policies through the institutions authorized to make them. This paper does not propose that an AI summary automatically becomes law. Any future proposal for directly binding participation would require its own legal design and public authorization. Emergency arrangements should be narrowly defined, time limited, recorded, and reviewed afterward.
Correction should be possible without pretending every consequence can be undone. Trials, staged deployment, review dates, and advance plans for stopping a harmful system can limit exposure. People should not remain trapped in a demonstrably damaging decision merely because reversing it is procedurally inconvenient.
Human dignity constrains the majority
Participation serves a purpose: improving the human condition while preserving the worth and agency of every person. A popular decision that sacrifices a vulnerable group would defeat that purpose. Dignity, essential care, privacy, and autonomy must therefore constrain the options presented and the decisions ultimately taken.
Those constraints also need fair interpretation. An independent human review and appeal process should explain when a proposal violates them and allow that judgment to be challenged. An AI system, its supplier, or a temporary majority should not have unchecked power to define the limits alone. UNESCO’s AI ethics recommendation provides relevant context on human rights, oversight, and inclusive governance. [3]
Participation itself must respect privacy. Identity checks to prevent duplicate voting should collect the minimum necessary information, protect private ballots where appropriate, and separate verification from public disclosure. Public auditability should expose institutional conduct without exposing people to retaliation.
Auditable systems and honest technical limits
We propose complete accountability for the decision process: recorded system versions, governing instructions, evidence sources, relevant tool actions, proposed alternatives, changes, human approvals, and reasons for decisions. Independent auditors should have sufficient access to test compliance, investigate failures, and challenge claims made by operators. Public reporting should explain findings in accessible language.
The longer term research ambition is full observability of the reasoning and algorithmic processing that materially shape an outcome. We should not claim this has already been achieved. An AI generated explanation or visible chain of thought is evidence to examine, not proof of every internal cause. Anthropic research found that tested models sometimes omitted influential hints from their stated reasoning. [4] That finding supports using additional technical tests and independent evidence rather than relying on explanations alone.
On October 26 2026, OpenAI announced an AI produced refutation of the Exponential Time Hypothesis, accompanied by a Lean formalization and arrangements for independent examination. [5] This is a reason to pursue ambitious research, not evidence that transparent governance or complete AI interpretability has been solved. Our inference is that publishing inspectable artifacts and enabling independent challenge is a useful accountability example.
Audits must address manipulation, selective evidence, framing bias, identity fraud, security failures, and dependence on a single supplier. Human facilitators and appeal routes should remain available. If a system cannot meet the accountability requirements of a consequential use, that should constrain its role rather than become an excuse to lower the requirements.
From proposal to practical learning
The next useful step would be a voluntary, limited consequence pilot with a clear scope and no claim to legislative authority. It could test whether accessible participation and AI supported synthesis help a diverse group understand a question, preserve disagreement, and reach a decision they regard as fair.
Evaluation should consider who participated and who could not, how accurately contributions were represented, whether people could challenge the process, and whether the decision improved human outcomes. Speed and agreement rates alone would be inadequate measures. Funding, governance, supplier relationships, and conflicts of interest should be disclosed.
This is a proposed research direction, not an announced Open Eye One program. The voting model, participation safeguards, review bodies, technical standards, measures, and resources remain to be developed with the community.
Conclusion and invitation
Human alignment should give people a continuing voice in the systems that shape their lives. AI may help us listen across differences and examine choices more carefully. Its contribution should remain subject to human rights constraints, public scrutiny, and the ability to correct mistakes.
We invite you to challenge this proposal. How should company participation work? What would make it practical for you to contribute? Which safeguards would help you trust a mediator, and what would make you reject one? Experience, objections, and alternative designs are all welcome.
Contribute to this paper or inquire about the advisory council ↗
Select Position 3 — Human alignment requires human participation in the website form. You can also request general updates or express interest in the advisory council.
Sources and scope
The sources below support the specific factual context identified in the text. The voting model and governance proposals are Open Eye One discussion proposals; none of these sources endorses them. Source pages and links were checked on October 29 2026. Source pages and links were rechecked for publication on October 29 2026.
[1] OECD, OECD Guidelines for Citizen Participation Processes, 2022. The publication overview supports the discussion of accessibility, inclusion, and feedback on public input.
[2] World Happiness Report, Frequently Asked Questions. Explains life evaluation and the Cantril Ladder; used as measurement context, not a complete set of governance indicators.
World Happiness Report measurement explanations ↗
[3] UNESCO, Recommendation on the Ethics of Artificial Intelligence, 2021. Context for human dignity, human oversight, and inclusive governance.
[4] Anthropic, Reasoning models do not always say what they think, April 3 2025. The research describes limitations in the faithfulness of reported reasoning in tested settings; it does not establish that every explanation is unreliable.
Anthropic research on reasoning faithfulness ↗
[5] OpenAI, Working with academic communities to build scientific understanding, October 26 2026. Company announcement of the ETH result and independent examination arrangements. This paper has not independently verified the mathematical proof.
OpenAI research announcement ↗
Editorial record
Version 0.1 was developed through an AI assisted interview and drafting process using human responses about participation, voting, mediation, rights, and accountability. AI assisted with structure, research, and wording. Human editorial review and approval to publish were completed on October 29 2026. This published working draft is open for community discussion and refinement. It has not undergone independent peer review or advisory council approval and does not claim community consensus.