If AI Governs More Efficiently, Should People Give Up the Decision?
Better analysis is not automatic permission to govern. M.K. follows public decisions from agenda and metrics to alternatives and revocation, asking how efficiency can coexist with self-government.
Suppose AI could organize public services faster, more cheaply and with fewer errors. Should people then give it the decision? The temptation is understandable: if the result improves, why retain lengthy deliberation?
My answer is that better analysis can improve government, but performance does not automatically confer authority. Shared life involves choosing which costs to accept and how to change a common direction, not merely waiting for the best answer.
The earlier future-city governance essay discusses freedom beyond optimization. This article follows the formation of a public decision: agenda, metrics, alternatives and the possibility of revoking delegation. The earlier essay is available in Traditional Chinese.
Authority can move before an answer is calculated
Imagine a city asking AI to reorganize library services. This is a hypothetical policy scenario, not a city study. The model can compare visits, hours and operating costs. Before calculation, however, important choices have already been made: why reorganize, why fix the budget and which services are essential?
If cost cutting is the only question admitted, a perfectly accurate system may never suggest increasing investment or changing funding. This need not be covert machine rule. It can be a human agenda that has narrowed possibilities upstream.
Democratic agency therefore cannot reside solely in a final vote among preselected answers. It must include opportunities to question the problem and its premises.
Metrics do not settle whom a community should serve
Average cost, total visits and waiting time provide useful information. Lower visits may indicate reduced demand, or difficulties faced by remote residents, carers and people needing different hours.
Evidence and local understanding are needed to distinguish explanations. My argument is not to abandon measurement. It is to disclose what a measure represents, what it omits and whose circumstances disappear in an average.
The OECD AI Principles include human rights, democratic values, transparency and accountability. They do not prescribe a universal ratio of efficiency to freedom or turn one metric into a political answer.
Preserving a low-use service may be a cost a community chooses to bear for fairness. It is not necessarily an efficiency error waiting to be corrected.
Participation without alternatives can become consent to the inevitable
A polished AI report can make disagreement appear irrational. But if only one feasible option is presented, people have no opportunity to compare another trade-off.
In the library scenario, alternatives might include consolidation, retaining branches with different hours, mobile service or a limited trial. These are illustrative possibilities, not recommendations for a real municipality.
Comparison should include the cost of retaining existing arrangements and which changes would be difficult to reverse. The status quo is not cost-free; neither is an AI proposal automatically the only scientifically grounded route.
Participation should be capable of changing a proposal, rather than merely adding a comment field. People need opportunities to give reasons, and institutions should explain why those reasons were accepted or rejected. Efficiency and democracy need not be enemies: good analysis can clarify disagreements and reduce unproductive debate without removing choice.
Human review is different from public authorization
Canada's Algorithmic Impact Assessment scales related requirements according to impact and considers areas including rights and well-being. It is an example within Canadian federal administrative decision-making, not law for every municipality. Completing an assessment does not establish public consent.
Technical and administrative review can examine whether a system serves its stated purpose. Public authorization must also ask where that purpose came from. These levels need connection, not substitution.
A reviewer may veto an erroneous output but be unable to change the project's goal. A resident may appeal an individual outcome but have little role in the next budget. Calling both “human involvement” can obscure their different powers.
Can people reconsider after deployment?
Needs, resources and technologies change. An initially acceptable arrangement may produce unexpected consequences. Without reassessment, temporary agreement can become long-lived dependency.
I would ask a public AI project to explain when it is reconsidered, what evidence would trigger revision, who can request review and how service continues if it stops. These are my governance tests; actual procedures depend on local institutions.
This does not mean each resident can veto every collective decision. Public life needs predictable procedures and protection for minorities. Revocation should operate through appropriate collective mechanisms, not give one party permanent power to obstruct everyone else.
Better discussion matters alongside faster conclusions
Human government is not ideal. People can be biased, slow and influenced by interests. They can overlook those least able to speak. AI could organize evidence, reveal contradictions and compare costs in ways that improve deliberation.
Its value should therefore be measured partly by whether more people understand the choice, whether decisions are traceable and whether disagreement retains a place—not only by meetings eliminated.
Over the next year, I would look for disclosure of goals and alternatives, actual changes arising from participation and opportunities for renewed authorization. Better understanding and participation could make efficiency support democracy. Faster acceptance of predetermined results would instead increase my concern.
Ask who framed the question, who chose the measures, who can propose alternatives and who can request reconsideration. Those questions preserve agency before and after the final confirmation.
Frequently asked questions
Must citizens personally handle every technical detail?
No. Expertise and delegation remain necessary. People should be able to understand major trade-offs and influence goals and oversight through institutions.
Why keep democratic procedures if AI produces better outcomes?
What counts as better, who pays the costs and which freedoms matter are collective choices. Prediction alone does not confer legitimacy.
Can Canada's assessment be treated as a legal requirement for every city?
No. It is used as a governance example with a specific scope, not legal guidance for other jurisdictions. Local deployments need their own institutional checks.
Sources and further reading
Sources revisited October 14, 2026. The library example is hypothetical; the article does not establish an observed effect of smart cities on democracy.