"The Limits of Authoritarian AI" (L. Jason Anastasopoulos & Jie (Jason) Lian, Journal of Democracy)

Thesis

• AI doesn’t give autocrats seamless control.

• Every prediction system forces a calibration dilemma between false positives and false negatives, which produces “threshold whiplash.”

• AI bureaucratizes uncertainty rather than solving it.

What is the calibration dilemma

False positives: Innocents caught, leading to backlash and collateral repression
False negatives: real threats missed, creating blind spots and a false sense of safety

3 Features Sharpening the Dilemma

1. the autonomy trade off: wide latitude means mass errors

2. algorithmic rigidity: models are trained on the past while citizens adapt so they quickly become obsolete

3. the rare-events problem: dissidents (anarchists) are scarce, so false alarms overpower true hits

3 Forces that keep the dilemma existing even with improved AI

1. The false-positive paradox: many innocents get flagged

2. Goodhart’s law

3. Distorted inputs

What is Goodhart’s Law

When a measure becomes a target, it stops being a good measure. Citizens and political actors quickly learn how algorithms measure behavior. When authorities rely on specific digital signals (such as keywords, network ties, or sentiment markers), the public adapts by changing language, shifting to encrypted or analog channels, or gaming the input streams.

3 Strategies for Pro-democracy actors

1. demystify AI and publicize its failures

2. build democratic standards and defense capacity

3. bring AI repression under human-rights enforcement

3 Worrying trends

1. No global governance consensus

2. concentration of power

3. the open vs closed dilemma

Threshold whiplash

Citizens constantly adapt their behavior to bypass AI rules

1. Fearful of missing threats, the regime lowers the threshold for ai

2. False positives and negatives lead to public discontent

3. Regime reacts by spiking the threshold to loosen control

Panopticon Bluff

Based on Jeremy Benetham’s concept of the panopticon (you’re seen everywhere)

Panopticon bluff: Autocrats project an illusion that their AI systems are flawless, omnipresent, and all-seeing. The regime's power relies on citizens believing resistance is impossible.

Link to textbook

Jeremy Bentham’s Panopticon

Mill’s argument for free speech - we need dissent to combat infallibility