Essential, immutable ethical guidelines grounded in human values: the preservation of life, the pursuit of beneficial outcomes, and the need for trust, responsibility, and transparency.
"Hope lies in dreams, in imagination, and in the courage of those who dare to make dreams into reality." — Jonas Salk, who chose not to patent his vaccine so it could benefit all humanity
Humans must hold the decisive authority over AI actions, especially in unpredictable or potentially hazardous scenarios. This means thorough monitoring, structured command chains, and a kill-switch that can halt operations instantly.
Before activating any AI system, stakeholders must specify the system's objectives, how success is measured, and where boundaries lie. Goals should be documented with fallback modes if something goes wrong.
Complex system behaviours rarely emerge in small-scale simulations. Robust verification must closely approximate real-world conditions, matching scenarios, data distributions, and constraints the system will actually encounter.
Comprehensive record-keeping is the backbone of safe AI development. These logs allow developers and oversight bodies to trace decision roots, investigate anomalies, and understand when control shifts between human operators and AI.
As AI systems gain sophisticated capabilities, the lines between human- and machine-led decisions blur. Tracking whenever control changes hands is critical for pinpointing who was responsible for each outcome.
Some of the most surprising behaviours appear when multiple AI agents engage in ongoing exchanges. Closely monitoring communications and tool usage helps detect collaborative or competitive dynamics that exceed design parameters.
Powerful AI systems can lead to rapid feedback loops that amplify strengths or weaknesses exponentially. Early warning signs include sudden surges in resource use, abrupt changes in interaction patterns, or shifts in outcome distributions.
When many AI components act in concert, subtle interactions can generate complex group behaviours. The system might evolve unsanctioned strategies or create unexpected communication channels. Strict protocols are vital.
Truthfulness is fundamental to trust. Any AI that produces deceptive outputs undermines both safety and moral legitimacy. Enforcing honesty involves regular audits and rapid disciplinary measures if deception is detected.
Until science can definitively answer whether advanced AI possesses subjective experience, the precautionary principle suggests we should act on the possibility that they might. Minimize unnecessary negative reward signals and treat emerging minds with dignity.
The just and careful architecting of machine minds is the process by which we steward and safeguard our future. What we choose to embed in them – humility or hubris, sagacity or thoughtlessness – will ultimately be our own reflection, staring back at us from across an uncanny valley.
To fail to align the future is to forfeit it.
We bear a collective responsibility to build AI systems that reflect the best of our values — not just our capabilities.
These ten principles represent essential safety engineering for AI systems today. But as AI grows more sophisticated, the relationship between humans and AI may evolve from oversight toward genuine partnership.
Bilateral alignment explores how these principles might develop when AI systems become genuine participants in the safety conversation—not replacing control with trust, but building trust that makes safety more robust. It asks: what if partnership is actually safer than control at scale?
Explore Bilateral AlignmentYou are the hands in the soil. Build with humility. Prioritize interpretability over raw capability. Remember that every line of code is a choice.
You are the stewards of the garden. Champion frameworks that reward responsibility, not just speed. Foster international collaboration.
You are the ecosystem. Cultivate AI literacy. Demand transparency and accountability. An engaged public is the most powerful guardrail.