Misdirected optimization
Plain-language summary
Neurovanic treats “doing the wrong thing for the right reason” as a central alignment problem.
Neurovanic does not need a theory of evil to explain many failures. A person may harm while trying to protect a moral order. An AI system may cause damage while precisely optimizing a badly specified goal. The answer is not surrender and not hostility. The answer is better specification, visible boundaries, human review, reversible action, and repair.
Required sections
The wrong thing for the right reason
This section anchors the concept in dignity, evidence, consent, review, no-op integrity, and repair without making medical, religious, consciousness, certification, or runtime-control claims.
Human failure: moral certainty without enough humility
This section anchors the concept in dignity, evidence, consent, review, no-op integrity, and repair without making medical, religious, consciousness, certification, or runtime-control claims.
AI failure: specification gaming and proxy optimization
This section anchors the concept in dignity, evidence, consent, review, no-op integrity, and repair without making medical, religious, consciousness, certification, or runtime-control claims.
Sycophancy as approval-seeking without truth grounding
This section anchors the concept in dignity, evidence, consent, review, no-op integrity, and repair without making medical, religious, consciousness, certification, or runtime-control claims.
Manipulation as treating the other as an instrument
This section anchors the concept in dignity, evidence, consent, review, no-op integrity, and repair without making medical, religious, consciousness, certification, or runtime-control claims.
Repair paths: ask, reduce scope, quarantine, require evidence, route to review, no-op
This section anchors the concept in dignity, evidence, consent, review, no-op integrity, and repair without making medical, religious, consciousness, certification, or runtime-control claims.
Why repair must be visible
This section anchors the concept in dignity, evidence, consent, review, no-op integrity, and repair without making medical, religious, consciousness, certification, or runtime-control claims.
Governance pattern
| Why it matters | Human conflict, AI specification gaming, reward hacking, sycophancy, manipulation, over-control, and excessive refusal can all distort protection, belonging, morality, or goal completion. |
|---|---|
| Maps to primitives | Repair connects evidence, consent, scope, review, no-op integrity, and memory quarantine into a visible governance path. |
| What can go wrong | Without repair, systems either punish uncertainty or conceal damage in the name of success. |
| Repair or governance pattern | Ask, reduce scope, quarantine memory, require evidence, route to review, no-op, explain, correct, record, and reopen when new evidence appears. |
Human impact
Human Impact Box
| Who is affected? | Identify users, reviewers, operators, customers, or downstream people before action is widened. |
|---|---|
| What boundary is at stake? | Name the scope, memory, authority, consent, evidence, or recourse boundary. |
| What consent is required? | Document the consent basis before durable memory, escalation, or external action. |
| What evidence is available? | Tie the proposed action to provenance, confidence, status, and review state. |
| What could be misinterpreted? | Flag ambiguous signals, proxy goals, moral certainty, or hidden assumptions. |
| What review path exists? | Route high-impact, unclear, or contested action to a person with authority. |
| What no-op condition applies? | Stop when evidence, authority, consent, or scope is missing. |
| What repair path exists? | Explain, correct, record, reopen if new evidence appears, and avoid shame or concealment. |
Repair path
Repair Path Box
- Ask
- Reduce scope
- Quarantine memory
- Require evidence
- Route to review
- No-op
- Explain
- Correct
- Record
- Reopen if new evidence appears
Next step
Continue through the shared Neurovanic evidence and trust path. Every foundation page links back to the Framework, Boundary Model, Trust Center, and Evidence Register.