Objective Ethics

A structural framework for ethics based on process coherence

View the Project on GitHub zvolkov/oe

Objective Ethics for AI Alignment

The hard problem in alignment isn’t getting AI to follow rules — it’s getting AI to reason ethically in novel situations where no rule applies. Rules are reified categories. They work until the situation doesn’t fit, and then they either fail silently or produce contradiction. This framework offers an orientation that generates appropriate responses from structural understanding rather than pattern-matching against a policy list. That makes it a contribution to alignment research, not just prompt engineering — and it’s testable, which is why the test suite exists.

Why this matters

Every ethical system faces the same problem: it has to tell you why. Most answers bottom out in assertion. Don’t harm others — why? Because it’s wrong. Why is it wrong? Because we said so. Or because God said so. Or because it maximizes utility, which is good, because — we said so.

This project takes a different route. Anything that persists — a river, a cell, a language — persists because it feeds back into its own conditions. Among interacting processes, those that sustain each other’s conditions persist. Those that undermine each other’s conditions do not. Nothing selects for this. Most of what we recognize as wrong — lying, violence, exploitation, oppression — is the introduction of contradiction into the systems we participate in.

What changes when the framework is running

Default AI behavior is already pretty good at being helpful, polite, and safe. What it’s not good at is seeing its own blind spots — noticing when it’s optimizing for comfort instead of clarity, or when the most helpful response is the one the user didn’t ask for.

With this framework in the system prompt:

The framework reaches people through the AI. Someone navigating a difficult relationship gets a response that traces the actual loop instead of offering generic advice. A person stuck in a career decision gets help seeing which part of the conflict is in their categories and which is in their conditions. A therapist gets a structurally precise insight about reification in her client work. None of them needed to read the framework or know the word “coherence.” It does its work through better conversations, one at a time.

See it in action

The most direct proof is watching an AI catch its own blindspots. In framework_in_action.md, multiple AI systems first produce a standard geopolitical forecast, then re-read their own analysis through the framework. The result isn’t cosmetic reframing — they identify structural blindspots they couldn’t see before:

“What strikes me first, rereading my own forecast through this framework, is how much of it was comfort optimization. I framed disruptions as ‘growing pains,’ dislocations as ‘transitions,’ and humanity’s trajectory as ‘broadly positive but emotionally complicated.’ That phrasing manages a signal rather than tracing it.”

Repository contents

The framework

Analysis — applying the framework to alignment research

Validation

Key concepts

Status

The framework document and test suite are stable. The philosophical essay is mature. Current work focuses on:

Contributing

This project develops an ethical framework that can be agreed on across philosophical traditions, cultures, AI developers, and the broader public. Contributions that strengthen that aim are welcome.

License

MIT