Ghanbari distinguishes epistemic weight from sovereign power. Superior forecasting should matter, especially in technical domains, but decisions must remain constrained by rights, legitimacy and the ability to contest what the system counts as success.
The case for epistemic weight
A structural engineer’s judgment carries more weight than a public poll when a bridge is unsafe. Epidemiologists matter during an outbreak; grid operators matter during a frequency emergency. Society already weights knowledge informally. If an AI system repeatedly predicts defined outcomes better than experts, refusing to use it merely because it is non-human could sacrifice welfare.
Prediction is not preference
A model can estimate how a carbon price affects emissions and household costs. It cannot decide how those costs should be distributed without a value rule. Voters may choose a less efficient policy because they reject its distributional burden. That choice can be informed or misinformed, but it is not reducible to forecast accuracy.
A competence threshold
Machine recommendations deserve authority only when the task, metric and comparison are defined in advance. Performance should be tested out of sample, across groups and under distribution shift. A system that predicts average outcomes but fails catastrophically for a minority has not “known better” in the relevant constitutional sense.
The right to demand reasons
Citizens cannot hold a probability score accountable. Public systems need reason-giving interfaces: which evidence mattered, what uncertainty remains, what alternative was rejected and where an appeal can go. Explanation is not merely a technical feature; it is part of legitimate power.
Decision rights as a portfolio
Future institutions may allocate decision rights by domain. Voters authorize goals and boundaries. Experts certify evidence standards. AI systems optimize bounded operations. Courts and independent auditors protect rights. No single actor—public, expert or machine—owns the entire chain.
The public may rationally reject the “best” forecast
People care about procedure, dignity and control as well as outcomes. An institution that consistently produces higher averages while denying agency may be unstable or unjust. Competence must be designed into legitimacy, not used to bypass it.
When AI knows more about a narrow consequence, it should receive more epistemic weight. It should not receive unlimited political authority. The future belongs to institutions that can tell the difference.
QUESTIONS READERS ASK
Further questions
Can AI make better decisions than voters?
It may make better forecasts in defined domains, but public decisions also contain value judgments and rights constraints.
Should experts have more political power?
Expertise can justify greater influence over evidence-sensitive tasks, provided authority is transparent, bounded and accountable.
EVIDENCE LAYER
Sources and further reading
- AI Risk Management FrameworkNIST ↗
- Governing with Artificial IntelligenceOECD ↗
- Recommendation on the Ethics of Artificial IntelligenceUNESCO ↗
Sources support factual context. Interpretive conclusions are presented as Meysam Ghanbari’s perspective, not as settled scientific fact.
