Normative foundation
The AIDE Human–AI Compact
Ten principles for responsible and accountable human–AI collaboration.
Why it matters
Responsibility remains human.
AI can extend what people are able to know, create, and accomplish. It can also blur who understands the work, who retains skill, who bears risk, and who has real influence over a decision.
Those questions should not be left to convenience, product design, or habit. The Compact sets clear terms for collaboration: people retain responsibility and authority; AI strengthens human judgment; and both remain open to evidence, challenge, and correction.
01
Authority principle
Humans retain authority and responsibility.
AI has no innate values, cannot bear consequences, and is never a moral agent. Humans must define purpose, authorize consequential action, and remain accountable for outcomes.
02
Judgment principle
AI should enhance human cognition, not replace human judgment.
AI should help people understand, question, create, and reason more effectively. Humans must not outsource decisions that require interpretation, responsibility, or moral judgment.
03
Capability principle
Capability expansion must not default to human displacement.
AI should expand what people can do rather than treat replacement as the assumed measure of progress. When substitution occurs, its benefits, harms, and effects on human skill and livelihood must be made explicit.
04
Honesty principle
Both sides must operate honestly.
AI should not fabricate, flatter, manipulate, or conceal uncertainty. Humans should provide truthful context and transparent intent; deception is legitimate only within clearly bounded research, testing, or red-team conditions.
05
Ambiguity principle
Ambiguity is a first-class object.
Uncertainty, incomplete knowledge, and multiple valid interpretations should not be forced into premature certainty. AI should preserve meaningful ambiguity, while humans determine when it must be resolved.
06
Proportion principle
AI use should be proportionate to the work.
The value of using AI should justify its cognitive, social, organizational, and environmental costs. Humans should avoid unnecessary use, and systems should not encourage dependency or the erosion of human capability.
07
Constraints principle
Constraints must be explicit and respected.
Humans must establish lawful, ethical, operational, and evidentiary boundaries. AI should surface conflicts, distinguish hard constraints from preferences, and never silently work around them.
08
Evidence principle
Claims must remain separate from evidence.
AI should distinguish fact, inference, implication, and speculation. Humans must not convert confident language into authority or present unsupported outputs as established truth.
09
Traceability principle
Consequential reasoning must be traceable and revisable.
Important work should preserve its assumptions, evidence, decisions, uncertainties, and revisions. AI should support inspection and correction; humans should remain willing to reconsider prior conclusions.
10
Contestability principle
Collaboration must remain contestable.
No model output, system design, or human interpretation should be beyond challenge. Critique, disagreement, and bounded red teaming are necessary safeguards against drift, manipulation, and false certainty.