About Tractatus

As AI systems make increasingly consequential decisions—medical treatment, hiring, content moderation, resource allocation—a fundamental question emerges: whose values guide these decisions? Current AI alignment approaches embed particular moral frameworks into systems deployed universally. When they work, it's because everyone affected shares those values. When they don't, someone's values inevitably override others'.

The Tractatus Framework exists to address a fundamental problem in AI safety: current approaches rely on training, fine-tuning, and corporate governance—all of which can fail, drift, or be overridden. We propose safety through architecture.

Inspired by Ludwig Wittgenstein's Tractatus Logico-Philosophicus, our framework recognizes that some domains—values, ethics, cultural context, human agency—cannot be systematized. What cannot be systematized must not be automated. AI systems should have structural constraints that prevent them from crossing these boundaries.

"Whereof one cannot speak, thereof one must be silent."
— Ludwig Wittgenstein, Tractatus (§7)

Applied to AI: "What cannot be systematized must not be automated."

Tractatus emerged from a simple but urgent question: as AI systems become more capable, how do we preserve human control over the decisions that matter? Not technical decisions—moral ones. Decisions about whose privacy matters more. Whose needs come first. What trade-offs are acceptable.

This is fundamentally a problem of moral philosophy, not management science. Different communities hold genuinely different, equally legitimate values. You cannot rank "family privacy" against "community safety" on a single scale—they are incommensurable. Any system claiming to do so is simply imposing one community's values on everyone else.

The framework's core insight comes from attending carefully to what humans actually need from AI: not another authority making decisions for them, but systems that recognize when a decision requires human deliberation across different perspectives. The PluralisticDeliberationOrchestrator represents our primary research focus—a component designed to detect when AI encounters values conflicts and coordinate deliberation among affected stakeholders rather than making autonomous choices.

Traditional organizational theory addresses authority through hierarchy. But post-AI contexts require something different: authority through appropriate deliberative process. Not "AI decides for everyone," but "AI recognizes when humans must decide together."

AI systems are amoral hierarchical constructs, fundamentally incompatible with the plural, incommensurable values human societies exhibit. A hierarchy can only impose one framework and treat conflicts as anomalies. You cannot pattern-match your way to pluralism.

Human societies spent centuries learning to navigate moral pluralism through constitutional separation of powers, federalism, subsidiarity, and deliberative democracy. These structures acknowledge that legitimate authority over value decisions belongs to affected communities, not distant experts claiming universal wisdom.

AI development risks reversing this progress. As capability concentrates in a few labs, value decisions affecting billions are being encoded by small teams applying their particular moral intuitions at scale. Not through malice—through structural necessity. The architecture of current AI systems demands hierarchical value frameworks.

The Tractatus Framework offers an alternative: separate what must be universal (safety boundaries) from what should be contextual (value deliberation). This preserves human agency over moral decisions while enabling AI capability to scale.

The Tractatus Framework consists of six integrated components that work together to enforce structural safety:

The Tractatus Framework emerged from real-world AI failures experienced during extended Claude Code sessions. The "27027 incident"—where AI's training patterns immediately overrode an explicit instruction (user said "port 27027", AI used "port 27017")—revealed that traditional safety approaches were insufficient. This wasn't forgetting; it was pattern recognition bias autocorrecting the user.

After documenting multiple failure modes (pattern recognition bias, values drift, silent degradation), we recognized a pattern: AI systems lacked structural constraints. They could theoretically "learn" safety, but in practice their training patterns overrode explicit instructions, and the problem gets worse as capabilities increase.

The solution wasn't better training—it was architecture. Drawing inspiration from Wittgenstein's insight that some things lie beyond the limits of language (and thus systematization), we built a framework that enforces boundaries through structure, not aspiration.

The Tractatus Framework is open source under the Apache License 2.0. We encourage:

• Academic research and validation studies
• Implementation in production AI systems
• Submission of failure case studies
• Theoretical extensions and improvements
• Community collaboration and knowledge sharing

The framework is intentionally permissive because AI safety benefits from transparency and collective improvement, not proprietary control.

Why Apache 2.0?

We chose Apache 2.0 over MIT because it provides:

• Patent Protection: Explicit patent grant protects users from patent litigation by contributors
• Contributor Clarity: Clear terms for how contributions are licensed
• Permissive Use: Like MIT, allows commercial use and inclusion in proprietary products
• Community Standard: Widely used in AI/ML projects (TensorFlow, PyTorch, Apache Spark)

View full Apache 2.0 License →