Contribution to the handbook
Priority · Assessed 2026-10-27
The distinction between improved joint performance and citizens’ ability to understand, influence, or challenge decisions is material to authority and accountability. The evidence includes possible deliberative benefits and risks of deference, expert dependence, and loss of agency, without establishing either universally.
Does this challenge the goal?
In-scope gap
It assesses whether AI partnerships support or displace meaningful human participation, an open issue within the stated goal.
The evidence includes proposals, cautions, and study findings from particular settings, but does not establish that the same effects apply to governance decisions.
Why this category?
- meaningful alternatives
- consequential participation and authority risk
- evidence limits and testing need
The deliberation findings are limited to their described study settings; participant preference or reduced division does not establish consent, improved governance, or retained citizen authority.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
- Does working with AI help people understand and shape decisions, or can it replace their judgment?
- Personal AI helpers
- Some algorithms predict better in some fields
- Personal AI helpers could extend recording tools
- When systems are favoured for knowledge
- Human involvement may not be enough
- Better performance, less understanding
- Deference depends on the outcome
- Specialist explanations may not be enough
- Review may empower experts
- Good results can invite deference
- Considering process and results
- From agents to recipients
- Participants favored AI statements
- Mediation outperformed simple exposure
- AI-assisted groups became less divided
- Groups preferred revised statements
- Experts remain preferred
- Some support for limited AI influence
- No evidence of better AI facilitation
- AI suggestions and discussion quality
- Checking AI and protecting viewpoints
- Rough agreement and continuing differences
Making and interpreting laws →
Authority, accountability and resilience →
Tests, stop conditions and unresolved work →
Read the exploration goalAssess both the potential for improved human understanding and influence and the risk of deference to AI judgment; distinguish joint performance from meaningful human participation.
What we can say so far
The evidence supports a limited positive claim: in a deliberation study, participants’ critiques were incorporated into revised group statements, and AI-mediated discussion was associated with greater alignment than simply exchanging views. A separate chapter reports improved conversation quality from real-time AI suggestions, but gives too little methodological detail here to assess the result. These findings suggest AI can help people contribute to and shape some group outputs; they do not establish that participants understood the AI’s reasoning or exercised meaningful control over decisions. Danaher’s account raises the opposing risk: people may defer to opaque systems, and formal human involvement may not enable them to challenge a system they do not understand. Those are cautions and possible mechanisms, not evidence that deference is a general or measured effect. Overall, the supplied material offers partial evidence of improved joint deliberative outcomes and some influence, but does not determine whether working with AI improves human understanding or causes people to replace their judgment with AI judgment.
What remains open
- No supplied empirical finding directly measures whether working with AI improves users’ understanding of the decision process or the AI’s reasoning.
- No supplied evidence establishes whether participants retain meaningful ability to challenge, override, or control AI-influenced decisions across settings.
- No supplied empirical evidence establishes the frequency, magnitude, or causal effect of deference to AI judgment.
- The deliberation results do not establish that reported alignment or statement preferences represent fair, durable, or substantively better decisions.
Conditions and limits
- The Habermas-machine findings are reported from one study source; the supplied summaries are interpretations of shared source material, not independent corroboration.
- The reported preference for AI statements over human-mediator statements did not compare identical processes: the human mediators did not have a critique phase.
- The Plurality chapter reports an experiment but the supplied passage omits participant, method, and measurement details.
- Danaher’s deference and moral-patiency claims are presented as risks or possible developments, not established general outcomes.
- The Landemore evidence caveat is specific to the article’s December 2023 statement; its forecast of change is not evidence that change occurred.
Objections to consider
- Joint performance may improve without the human understanding or guiding the process. The chess example motivates this objection but does not establish how AI use affects participation in other settings.