the second hundred · metaphor 212
Can you ever give a complete, true account of yourself — from the inside? Every honest self-description seems to leave out the very act of describing, and the fuller you try to make it, the further that horizon slides away.
There is a particular vertigo in trying to be fully honest about yourself. You narrate your motives, and then notice the narrating; you correct for your bias, and then wonder about the bias in the correction. Each pass adds a layer, and each new layer needs its own accounting, which you cannot give from where you are standing — because you are standing inside the thing you are trying to weigh. The account is never wrong, exactly. It is just never closed.
In 1933 the logician Alfred Tarski proved that this isn't a personal failing. Any language rich enough to talk about itself cannot contain its own definition of truth. Ask a system to certify which of its own sentences are true, and one sentence — a plain little liar — tears the whole assignment apart. The only repair is to step up and out, and speak of the language's truth from a language above it.
The idea
Write a sentence that says of itself: “this sentence is false.” Try to file it as true. If it is true, then what it says holds — so it is false. Try false instead. If it is false, then what it says fails — so it is true. There is no verdict the sentence will sit still under; every value it is handed, it overturns. This is the liar, and it is not a riddle to be solved but a structural fact.
Tarski turned the paradox into a theorem. Suppose a language L is rich enough to do a little arithmetic and to talk about its own sentences — enough to build a sentence that refers to itself. Then L cannot contain a predicate True(x) that correctly reports, for every sentence of L, whether it is true. If it could, the liar could be written inside L, and the assignment would contradict itself. So truth-in-L is not definable in L. You can only define it from a richer metalanguage, standing one level above.
The escape hatch is a tower of languages. The object language talks about the world but owns no truth predicate at all. “True” for it lives one storey up; “true” for that storey lives higher still. Truth is always spoken from above — never by the level it is about. The liar simply cannot be assembled, because the words it would need for its own truth are never in its own vocabulary.
What to try
The machine holds six sentences. Four are ordinary — they point at the world or at each other. Two point at themselves: a liar (“this sentence is false”) and a truth-teller (“this sentence is true”). Each mode is a different bargain with the paradox, and the instrument actually runs the bargain rather than describing it.
In classical mode every sentence must be true or false — and applying the truth rule again and again sends the liar into an endless flip, T, F, T, F, that never lands. That non-landing is Tarski's obstruction. Switch to allow a gap: now a sentence may be left with no value, and the same rule slides the liar into a stable undefined — consistent, but only by admitting truth can't reach it. Switch to stratify: each sentence gets a level, “true” always sits one level up, and the liar's self-reference is flagged as unwriteable — the paradox never forms, and truth becomes fully definable from above.
The mapping
Read the language as a person, and its sentences as the things that person believes about themselves. Most self-beliefs are ordinary sentences — “I was harsh in that meeting,” “I am tired” — pointing outward at the world or at each other, and they settle into stable values. The trouble is the one belief that is about its own truthfulness: am I being honest with myself right now, in this very act of self-examination? That is the liar's shape, and from inside, it will not hold still.
Tarski's structural verdict is oddly consoling. The reason you cannot produce a final, self-certified account of yourself is not that you are especially opaque or dishonest; it is that no rich system can certify its own truth from inside. The move that works is the same one the tower makes: you borrow a level above — a friend, a therapist, a diary read years later, a tradition, time itself — and let it speak the truth about the self that couldn't reach its own. Self-knowledge is real, but it is always spoken from one storey up.
Read as life lessons
The liar has no value not because we haven't looked hard enough but because the value it takes, it destroys. Some questions about yourself have no inside answer — pressing harder only spins them faster.
Every consistent repair works by stepping up a level. An honest account of your life is spoken from a vantage you don't currently occupy — later, elsewhere, or through someone else's eyes.
Only the strictly self-referential sentence breaks; the ordinary ones settle just fine. The lesson is narrow: distrust the loop that judges its own judging, not the plain observation.
In the wild
Tarski's theorem is Gödel's close cousin: both build a self-referential sentence to fence off a system's reach. It underlies why provability, and truth, must be handled from a metalevel.
Formal semantics stacks object language, metalanguage, meta-metalanguage — each defining truth for the one below. Kripke later showed how gaps let a language get closer to talking about its own truth.
Like Gödel and Heisenberg, it's dragged in to “prove” all truth is relative or the self is unknowable. It proves neither — only that one formal trick, self-defined truth, is unavailable.
The mapping, exactly
| Logic | Life |
|---|---|
| the object language | You, describing yourself in the ordinary way — able to talk about the world and about your own sentences. |
| the truth predicate | A final, complete verdict on which of your self-accounts are actually true. |
| the liar sentence | The belief about your own present honesty — the one self-judgment that judges the judging. |
| no fixed point | Why introspection loops: every verdict you hand that belief, it overturns, so the account never closes. |
| the truth-value gap | Letting the question simply have no inside answer — consistent, at the price of admitting a blind spot. |
| the metalanguage above | A friend, a diary, a later self, time — the outside vantage that can say what you couldn't from within. |
Where the metaphor tears
Tarski's theorem is exact about symbolic systems with a precise syntax and a bivalent notion of truth. A person is vague, revisable, and only sometimes bivalent. The metaphor borrows the shape of the limit — self-defined truth is unavailable — not a literal proof that you cannot know yourself. Read it as a caution, not a decree.
Only the strictly self-referential truth-sentence is unreachable. Enormous amounts of accurate self-knowledge remain: you can be right about your habits, your history, your effect on others. The theorem forbids one final self-certification, not honesty. Overreading it into “the self is unknowable” is exactly the abuse the theorem gets subjected to.
Stepping up a level resolves this liar — but the new level has its own truth predicate it can't define, and its own liar waiting. The outside vantage is always partial; there is no ultimate metalanguage that sees everything from nowhere. The consolation is real but finite: truth from above, forever, with no final storey.