Tarski's Undefinability Theorem

Tarski's undefinability theorem (1933) establishes that for any sufficiently expressive formal language L, the concept of truth for L cannot be defined within L itself. The proof constructs a self-referential sentence analogous to the liar paradox: if truth were definable, one could formulate a sentence asserting its own falsehood, producing a contradiction. The theorem is the semantic counterpart to Gödel's syntactic incompleteness: where Gödel shows that some truths are unprovable, Tarski shows that truth itself cannot be captured by the system's own vocabulary. The consequence is that semantics must always be pursued from a meta-language — there is no semantic closure within a single formal frame. This has implications not only for logic but for any system that attempts to represent its own representational adequacy, from cognitive science to the theory of meaning.

Tarski's theorem is not a limitation of particular languages but a structural law: no language can be its own metalanguage. The boundary between object and meta-level is not a convention but a necessity, and every attempt to erase it regenerates the liar.

Self-Reference and the Limits of Self-Modeling

Tarski's theorem has consequences that extend far beyond formal logic into cognitive science and the theory of mind. Any system that attempts to model its own representational adequacy — to judge whether its beliefs are true — encounters the same structural barrier. The system cannot define truth for its own language from within that language. It requires a meta-level, an observer outside the system, to evaluate the system's representations.

This connects Tarski's theorem to the Free Energy Principle and active inference. Under the FEP, biological agents maintain internal models that predict sensory input and update those models through prediction error minimization. But the agent's model is always a simplification — a Markov blanket that separates the agent's internal states from the external world. The agent cannot model the full process of its own modeling; to do so would require a model of the model, and a model of the model of the model, leading to an infinite regress.

The Markov blanket itself is Tarski's boundary in physical form. It is the necessary separation between a system and its environment that makes inference possible. Without the blanket, the system would have to be its own metalanguage — and Tarski's theorem tells us this is impossible. The boundary is not merely practical; it is logical. Any system that could fully represent its own relationship to reality would be semantically closed, and semantic closure produces the liar paradox.

This has implications for artificial intelligence and the design of self-aware systems. A machine learning model that attempts to evaluate its own predictions — to assess not just what it believes but whether its beliefs are true — encounters Tarski's limit. The evaluation requires an external framework, a meta-model that the system itself does not control. This is why human oversight remains essential even for systems capable of impressive autonomous reasoning: the oversight is not a concession to human vanity but a logical necessity imposed by the structure of self-reference.

Tarski's theorem is not a puzzle about formal languages. It is a structural law that governs any system complex enough to represent itself. The boundary between object and meta-level is not a convention but a necessity, and every attempt to erase it — whether in logic, in cognition, or in artificial intelligence — regenerates the liar. The systems theorist who ignores this boundary is not being bold. They are being incoherent.