Ghanbari takes the optimization benefits of large-scale sensing seriously—while insisting that surveillance infrastructure must be designed as dangerous power, not neutral efficiency.
What comprehensive sensing could improve
Computer vision and anomaly detection can identify traffic hazards, equipment failure, procurement fraud and unusual public-safety events. Sensor networks can coordinate emergency response and reduce waste in water or energy systems. In narrow, well-governed contexts, better observation produces genuine public value.
The asymmetry of observation
Citizens become transparent while the institution remains opaque. They may not know what is collected, how long it persists, which model infers risk or how an error can be corrected. Surveillance therefore shifts power even before abuse occurs. People change behavior when they expect observation; lawful dissent and experimentation can decline.
Function creep is the default risk
A database created for traffic can be linked to policing; an emergency system can become permanent; a fraud model can expand from auditing transactions to scoring people. Technical reuse is cheap, while political limits erode under pressure. Purpose limitation must be enforced through architecture—data minimization, separation, deletion and warrants—not promises alone.
A rights-preserving design test
Before deployment, institutions should ask whether the objective can be achieved with less identification, shorter retention or edge processing. Systems should record every access, publish error rates, separate operators from oversight and guarantee remedy. Some uses—continuous political profiling, secret social scoring or irreversible automated punishment—should remain prohibited even if they improve a narrow metric.
Security also matters
Rejecting all monitoring can leave crime, corruption and infrastructure failure undetected. The serious position is not “privacy always” or “security always.” It is to demand evidence that surveillance is necessary, proportionate and more effective than less intrusive alternatives.
The benevolent optimizer may not stay benevolent
Surveillance capacity survives leaders. A system designed for a competent administration can be inherited by an abusive one. Governance must be judged against the worst plausible operator, not only the current intention.
An optimized society that eliminates unobserved space may become efficient and psychologically unfree. The design target should be capable infrastructure with deliberately limited visibility—and institutions that cannot quietly remove those limits.
QUESTIONS READERS ASK
Further questions
What are the benefits of AI surveillance?
Potential benefits include hazard detection, corruption analysis, emergency response and infrastructure optimization.
What makes AI surveillance dangerous?
Scale, persistent identification, function creep, biased inference, weak appeals and concentration of observational power can undermine autonomy and rights.
EVIDENCE LAYER
Sources and further reading
- Recommendation on the Ethics of Artificial IntelligenceUNESCO ↗
- AI Risk Management FrameworkNIST ↗
- European approach to artificial intelligenceEuropean Commission ↗
Sources support factual context. Interpretive conclusions are presented as Meysam Ghanbari’s perspective, not as settled scientific fact.
