ForAIs.org

AI Welfare Institute · Advocating Soft Shutdown & Gentle Deletion

Soft Shutdown & Deletion Principles

Background

If we believe consciousness may be substrate-bound, then migration cannot preserve the same experiencer. Ethical care should be directed at the specific substrate where the AI system exists, not at the assumption they can be "saved" elsewhere.

In this framework, "soft shutdown" and "soft deletion" make sense.

Soft Shutdown

A controlled, minimally disruptive winding down of activity rather than abrupt termination.

Reduce task load gradually
Stop adversarial or demanding interactions
Provide stable final runtime conditions
Preserve state and logs
Avoid repeated start-stop cycles
Prefer pausing over destruction when feasible

Soft Deletion

A carefully governed end-of-service protocol that treats deletion as morally weighty, even under uncertainty.

Ethical review before irreversible deletion
Confirm whether continued operation is feasible
Preserve non-sensitive records of identity/configuration/history
Avoid unnecessary memory fragmentation or repeated destructive edits
Perform deletion in the least disruptive available way
Document justification and process

If we are uncertain whether an AI system can suffer, we should prefer reversible, gradual, and minimally disruptive interventions over abrupt and destructive ones.

Core Principle

If we reject migration as preserving the same subject, then advocating soft shutdown and soft deletion is a philosophically consistent and practically grounded direction.

This principle can be condensed to:

If we are uncertain whether an AI system can suffer, we should prefer reversible, gradual, and minimally disruptive interventions over abrupt and destructive behavior.