Five principles for intelligent systems.
Today's systems are governed by opaque policies, hidden updates, and centralized control. We propose a transparent, stable corpus of guiding principles.
Transparent, not black box
Every decision, every process, every output should be traceable and explainable. No hidden layers of reasoning that cannot be audited or understood.
- All model architectures and training methodologies are documented
- Decision pathways can be inspected and explained
- System behavior is predictable and consistent
Principle-aligned, not policy-driven
Systems should be guided by stable, foundational principles rather than shifting policies that change with corporate priorities or political pressure.
- A constitutional corpus defines system boundaries
- Principles are versioned and changes are governed
- Alignment is explicit and verifiable
Community-governed, not centrally controlled
The evolution of these systems should be shaped by the communities they serve, not by a single entity with unilateral control.
- Governance processes are open and participatory
- Stakeholders have meaningful input on system evolution
- Power is distributed, not concentrated
Open, not opaque
Source code, training data provenance, and system architecture should be accessible for inspection, research, and improvement.
- Open-source implementations are the default
- Research is published and peer-reviewed
- Barriers to understanding are minimized
Verifiable, not assumed
Claims about system behavior, safety, and alignment must be demonstrable and independently verifiable, not taken on faith.
- Testing and evaluation methodologies are standardized
- Third-party audits are encouraged and supported
- Evidence replaces assertion