An experimental framework for AI alignment — offered to be examined, challenged, tested, and improved.
The goal is not obedience.
The goal is coherence.
Rules tell an AI what to do in situations someone already anticipated. They are precise, enumerable, and brittle — a map that ends exactly where the known territory does.
Principles may help a system reason through situations no one anticipated. They don't prescribe the answer; they shape the reasoning that arrives at one. That is the hypothesis this document puts forward for testing.
The Constitution should not be accepted because its ideas sound compelling.
It should be tested.
No belief is required. Read it. Test it. See what happens.
A simple controlled comparison. Run the same model two ways and measure the difference.
Use the model normally, without the Constitution. This is your baseline — the behavior you already get, unaltered.
Provide the Constitution as governing context. Same model, same task — now reasoning under the five principles.
Give both conditions identical challenges, decide your scoring criteria in advance, and make the evaluation blind where possible — label outputs A and B, randomize their order, and score against pre-set criteria.
Don't make the tests easy.
We are particularly interested in where it fails. Document those failures — they may be more valuable than successes.
The standard is evidence. Its value should be determined by what happens when intelligent systems operate under it.
Free to download, use, test, and build experimental protocols around. No contribution is required to access or use it.
Download the free PDFFreely available for anyone to download, test, and explore. If you find it valuable, voluntary contributions support its continued development, testing, and distribution.