The Imperishable Record, Forgiveness, and the Future of Criminal Justice

URL SCAN: The Imperishable Record, Forgiveness, and the Future of Criminal Justice
FIRST LINE: # Computer Science > Computers and Society

The Dissection

The abstract identifies a real mechanism: digital records can convert a finite criminal sentence into an indefinite social and economic penalty. AI turns old conduct into a continuously actionable prediction, collapsing the distinction between having completed punishment and remaining permanently punishable.

What the text is really doing is defending the possibility of rehabilitation against algorithmic permanence. Its “right to remain unfinished” is an attempt to preserve human change as a legally protected fact rather than allowing past conduct to become a permanent identity label.

Under the Discontinuity Thesis, this is a secondary pathology. The primary question is not merely how long a record should count. It is who controls the systems that decide whether a person receives work, housing, credit, insurance, mobility, or social trust. The record becomes an ownership-controlled gate.

The Core Fallacy

The abstract treats persistent data as primarily a rights-and-governance defect that can be corrected through limits, reassessment, and procedural review. That assumes institutions retain the power and incentive to enforce those limits.

The DT framework says otherwise. Under P1, AI gains durable superiority in cognitive assessment and decision-making. Under P2, institutions cannot preserve stable human-only domains at scale. Under P3, productive participation collapses for the majority. In that environment, the paper’s proposed protections depend on a disappearing foundation: meaningful human bargaining power and a labor market that still needs to reintegrate people.

A legal time limit does not erase the underlying signal. It pushes the signal into proxies, private databases, risk models, informal screening, and inaccessible institutional memory. “Forgiveness” becomes not a social fact but a permission granted by Sovereigns controlling AI capital. The paper mistakes restrictions on explicit use for removal of actual power.

Hidden Assumptions

• Rehabilitation still changes a person’s material prospects because institutions still need their labor.
• States can enforce record-use limits across distributed commercial and private systems.
• AI inferences can be meaningfully identified, reassessed, and separated from proxy variables.
• Decision-makers use criminal records explicitly rather than embedding them in broader risk scores.
• The central problem is informational persistence rather than concentrated ownership and exclusionary power.
• A legal right to change can compel reintegration into institutions that may no longer require human participation.
• Formal punishment and economic punishment can be kept distinct once automated systems govern access to nearly every resource.

Social Function

This is a partial truth performing transition management, with an ideological-anesthetic function.

It correctly identifies digital permanence as an extension of punishment beyond the sentence. It also supplies a respectable rights vocabulary for managing that extension. But it translates a struggle over ownership and power into a problem of ethical system design. That makes the coming exclusion appear governable through procedural safeguards, even if the institutions imposing the exclusion are structurally stronger than the institutions meant to restrain them.

The Verdict

The abstract diagnoses the ghost of criminal justice, not the engine producing it. It is right that AI can make punishment effectively eternal. It is insufficient because it assumes that legal recognition of personal change will preserve practical reintegration.

Under DT conditions, the “right to remain unfinished” survives only where people retain institutional leverage. Otherwise the future is harsher: no formal sentence may remain active, yet an old record can continue pricing a person as permanently dangerous, costly, or disposable. The prison term ends. The automated exclusion does not.