Contribution to the handbook
Priority · Assessed 2026-10-28
Readiness is consequential to whether, where, and under what conditions AI systems could participate in civic decisions. The packet includes an explicit author assessment that current language models are not ready for full augmented-democracy systems, study-specific prediction limitations and subgroup differences, and contrasting evidence about AI-assisted deliberation. It supports a central, conditional assessment—not a claim that any system is ready or that one governance arrangement has authority.
Does this challenge the goal?
In-scope gap
The question asks for readiness criteria and evidence for consequential civic use, which the definition invites the exploration to assess; it does not reject the goal of investigating AI governance.
The question’s readiness judgment is normative as well as empirical. The packet supplies competing considerations, not an agreed threshold or citizen endorsement.
Why this category?
- consequential risk and limits
- readiness testing
- meaningful alternatives
- citizen participation and authority
The Hidalgo findings are from one study and share a source root; they do not establish performance for other populations or settings. The deliberation evidence concerns a different task and process, and includes limitations such as omitted fact-finding stages. The packet does not settle the normative criteria for readiness or the boundaries of delegated authority.
This is an AI assessment of contribution to the goal, separate from whether the question has an answer.
- Are these systems ready to use in real civic decisions?
- Not ready for real decisions
- Individual prediction results
- Differences by political view and education
- Choices and option order
- A proposal for personalized AI helpers
- The assembly left out fact-finding stages
- Use it within a broader process
- Agreement rose with AI mediation, not exposure alone
- A risk to participation and understanding
- People need to understand the process
- Citizens authorize rules; AI acts within their mandate
- Citizen participation with separated checks
- Invest in safe democratic tools
Authority, accountability and resilience →
Implementation and transition →
Tests, stop conditions and unresolved work →
Read the exploration goalAssess their demonstrated reliability and limitations for real civic use, distinguishing current evidence from future possibilities.
What we can say so far
The evidence supports a limited conclusion: these systems have shown measurable performance in specific studies, but the supplied evidence does not demonstrate reliable suitability for real civic decisions generally. In one study, fine-tuned models predicted participants’ recorded pairwise preferences with reported accuracy above a simple party-based rule in most comparisons; the study also found high retest reliability for the group’s aggregate preferences. Separate deliberation studies reported increased group agreement, including in a virtual citizens’ assembly. These are results on particular participants, tasks, and procedures—not evidence that civic decisions made with the systems are more accurate, legitimate, or beneficial. The evidence also reports subgroup accuracy differences and sensitivity to option order. The Hidalgo authors explicitly conclude that current language models are not ready for full augmented-democracy systems; that is an attributed author assessment, not itself an empirical readiness test. Overall, the studies provide evidence of promise and specific demonstrated capabilities, but not demonstrated reliability for broad real civic use.
What remains open
- No supplied evidence establishes performance or reliability in consequential real civic decisions across varied populations, issues, or institutional settings.
- The supplied studies do not establish that model-assisted decisions improve decision quality, representativeness, legitimacy, or civic outcomes.
- The supplied material does not resolve the implications of subgroup accuracy differences or option-order sensitivity for deployment in real civic decisions.
Conditions and limits
- The reported preference-prediction and aggregate-reliability results concern the study’s participants, models, proposals, and comparison methods; they do not establish performance for other populations or decision settings.
- The deliberation findings concern the tested procedures and groups. The assembly mainly replicated the discussion phase and omitted fact-finding and expert-testimony stages.
- Increased agreement is an observed outcome, not evidence that the resulting views were correct or that the decision was better.
- The authors’ statement that current systems are not ready is an attributed assessment; the supplied passage does not specify a formal readiness standard.
- The evidence consists of interpretations of supplied source material and is not independent new evidence.
Objections to consider
- Reportedly lower prediction accuracy for some political and education groups raises a material representativeness and reliability concern. The supplied evidence does not establish its cause, extent beyond the study, or remedy.
State whether those limitations justify using them in consequential civic decisions, making the judgment’s criteria explicit.
What we can say so far
On the supplied arguments, the limitations justify not using current language-model systems as autonomous or decisive authorities in consequential civic decisions. My criterion is that a system should not materially determine a public decision unless affected people can meaningfully participate, understand the reasons and process, contest the decision, and retain effective human and institutional control. Inclusion in a broader deliberative process and protections against capture are also relevant safeguards. The supplied authors’ assessment that current language models are not ready for full augmented-democracy systems supports this cautious judgment, but it does not establish that every civic use is unsuitable. Limited, nonbinding assistance may be acceptable where these conditions are demonstrably met; the supplied evidence does not establish that any particular system or setting meets them.
What remains open
- The evidence does not provide empirical evaluations showing whether particular systems satisfy the participation, understanding, contestability, control, inclusion, or anti-capture criteria.
- It does not define thresholds for adequate public understanding, effective oversight, or acceptable levels of risk, nor identify which safeguards would be sufficient in particular decision settings.
- The evidence does not compare the civic benefits of use with the risks or establish which limited uses, if any, meet the criteria.
Conditions and limits
- This is a normative assessment based on the supplied authors’ arguments and positions, not independent verification or a finding that all systems fail the stated criteria.
- The judgment concerns systems that materially shape consequential civic decisions; it does not rule out all limited or nonbinding assistance.
- Any proposed safeguards must be effective in practice, not merely formal, before they can justify consequential use.
Objections to consider
- The potential practical and procedural benefits of algorithmic systems could weigh against excluding them, but the supplied evidence does not identify or compare those benefits in particular civic settings.
- The control categories do not show when monitoring or override is effective in practice. The judgment therefore treats effective control as a condition, but the evidence supplies no test for satisfying it.
- Embedding a system in a broader deliberative process may be a safeguard, but the supplied recommendation does not show that it is sufficient to make consequential use acceptable.
- The proposed enhancement route may be delayed and its effects are uncertain, so it cannot currently establish that the relevant limitations will be remedied.