How, if at all, could shared or public data arrangements reduce citizens’ disadvantage in benefiting from or scrutinizing AI-assisted governance, and what limits or risks would remain?

Contribution to the handbook

Priority · Assessed 2026-10-30

It raises a consequential design alternative for citizen access and scrutiny, including the dependency on analytical capacity and residual risks of manipulation and capture. These issues materially affect accountability and resilience.

Does this challenge the goal?

In-scope gap

It asks about a possible mechanism and its limits within the existing goals of serving citizens and resisting capture, rather than challenging those goals.

The question does not assume that shared data arrangements are effective, safe, or sufficient.

Authority, accountability and resilience →

Tests, stop conditions and unresolved work →

Explain how shared or public data arrangements could affect citizens’ ability to generate, access, and analyze large datasets, and how this might influence their ability to benefit from or scrutinize AI-assisted governance. Compare relevant alternatives without assuming that access to data alone creates equal analytical capacity.

What we can say so far

Shared or public data could, in principle, reduce one part of citizens’ disadvantage. Citizens pooling contributions might collectively generate a larger dataset than an individual could assemble; public access to an existing dataset might let them examine information otherwise held by institutions. These are possibilities inferred from the resource gap, not arrangements evaluated in the supplied evidence. Compared with an individual using a personal data tool, either arrangement could give citizens more material to analyze and potentially use for their own benefit or to check patterns relevant to public decisions. Machine learning is one possible way to analyze large datasets. But access is not the same as the ability to analyze: organizations with data, tools and expertise may still have an advantage. A shared dataset would also need to be relevant to the governance system in question; patterns in it would not by themselves establish why a particular decision was made. Danaher argues that meaningful scrutiny requires people to be able to examine and question the factors a system uses, not merely know its broad purpose. Expert review might help interpret a system, but could also leave citizens dependent on specialists they cannot effectively challenge. Thus shared data could support scrutiny without necessarily making decisions understandable or contestable.

What remains open

• The supplied evidence does not assess a specific shared or public data arrangement, its feasibility, or its comparative effectiveness in generating, accessing or analyzing data.
• It does not establish what data and other access would allow citizens to identify the factors behind actual decisions or successfully challenge them.

Assess what evidence establishes about effects on citizen services, scrutiny, or concentration of power, and what remains unquantified or untested. Examine residual risks such as manipulation and capture without treating openness as a demonstrated safeguard.

What we can say so far

The supplied evidence does not establish whether shared or public data improves citizen services or scrutiny, or reduces concentrations of power. Danaher reports concerns that data-mining benefits favor organizations able to generate and control large datasets; the cited passage does not assess whether shared or public data would narrow that gap. Residual risks remain plausible, not quantified effects. Danaher argues that review of complex systems could shift power to a small group of experts rather than citizens. Hidalgo and coauthors warn that manipulation and capture are possible under proprietary, open-source, venture-funded and government-supported models. They suggest that switching between providers might help, but do not show that competition prevents capture. Landemore also identifies bias, privacy, surveillance and legal risks across AI applications. Openness is not established as a safeguard against any of these risks.

What remains open

• No supplied evidence measures the effect of shared or public data on citizen services, scrutiny or concentration of power.
• The size and distribution of the data-access disadvantage, and any reduction achievable through sharing, remain unquantified.
• The effectiveness of provider switching or openness against manipulation and capture is not established.
• The incidence and magnitude of the identified privacy, surveillance, bias and expert-control risks under shared-data arrangements are not established.