Speculative Incarceration: Prisons for Crimes Not Yet Committed

Criminal justice has always been reactive. Someone commits a crime, is caught, tried, and punished. The sequence assumes the crime comes first.

AI prediction inverts this sequence.

If AI can predict criminal behavior with 90% accuracy—and it will be able to—the logic of prevention takes over. Why wait for the crime? Why allow the victim? Why not intervene before the act?

This is speculative incarceration. It sounds dystopian because it is. It is also the logical endpoint of predictive systems optimizing for harm reduction.

The Prediction Capability

What AI Can Already Predict

AI systems already predict:

• Recidivism: Who is likely to reoffend (used in bail and sentencing)
• Hot spots: Where crimes are likely to occur (used in patrol allocation)
• Risk profiles: Who is likely to be involved in violence (used in gang databases)
• Trajectory modeling: Who is on a path toward serious crime (used in intervention programs)

These predictions are not perfect. But they are better than chance, and they are improving.

What AI Will Be Able to Predict

As data collection expands and models improve:

• Individual probability of specific crime types
• Timing windows when risk is elevated
• Triggering conditions that precipitate action
• Intervention points where different outcomes are possible

The question is not whether these predictions will be possible. It is what we do with them.

The Accuracy Threshold

At what accuracy does preventive action become "justified"?

• 50% accuracy: No better than a coin flip. Clearly unjust.
• 70% accuracy: More likely than not. Still many false positives.
• 90% accuracy: Strong prediction. But 10% are still innocent.
• 99% accuracy: Very confident. But 1% of a large population is many people.

There is no accuracy threshold at which preventive incarceration becomes just. But there are thresholds at which it becomes tempting.

The Logic of Prevention

The Utilitarian Argument

A utilitarian calculus:

• One predicted murderer, if incarcerated, cannot kill.
• One actual murder prevents one death.
• If prediction is 90% accurate, preventing 10 predicted murders saves 9 lives at the cost of 1 false positive.
• The math favors prevention.

This logic is compelling to policy makers focused on outcomes.

The Slippery Slope

Once prevention is accepted for murder, why not for:

• Serious assault?
• Sexual offenses?
• Terrorism?
• Property crime?
• Traffic violations?

Each step down the gradient is "logical" once the previous step is accepted. The endpoint is total surveillance and preemptive control.

The Self-Fulfilling Prophecy

Predictions affect outcomes.

If someone is labeled high-risk:

• They may lose employment opportunities
• They may lose social support
• They may be surveilled heavily
• They may be arrested for minor infractions

These interventions may cause the predicted outcome. The prediction creates the conditions for its fulfillment.

The Justice Inversion

Punishment Without Crime

Criminal justice is based on the principle that punishment follows crime.

Speculative incarceration punishes before crime. It punishes for what someone would have done.

This is not justice in any traditional sense. It is risk management applied to humans.

Due Process Collapse

Due process assumes:

• A specific alleged act
• Evidence of that act
• Opportunity to contest the evidence
• Judgment by peers

Speculative incarceration has:

• A statistical prediction
• Algorithmic assessment
• No act to contest
• No evidence to challenge
• No meaningful appeal

How do you prove you would not have committed a crime?

The Minority Report Problem

In the classic formulation: the pre-crime system works until it produces a false positive that matters.

But in reality, the system produces false positives constantly. It just produces them among people who cannot effectively contest.

The false positives are not evenly distributed. They fall on the already marginalized.