Raymond Smullyan's Logical Puzzles

Module 4 · Section 4 of 6

Raymond Smullyan’s Logical Puzzles

Raymond Smullyan was a mathematician and logician who spent his career making formal logic feel like play. His books — What Is the Name of This Book?, The Lady or the Tiger? — are full of puzzles that look like riddles but are actually training in systematic reasoning.

His most famous setting: an island where knights always tell the truth and knaves always lie. You meet someone who says, “I am a knave.” What are they?

Work through it. If they are a knight, they must be telling the truth — but then they would be a knave, which is a contradiction. If they are a knave, they must be lying — but then “I am a knave” would be false, which means they are a knight. Another contradiction.

The statement is logically impossible for either type to make. That is the point. Smullyan is not trying to trick you — he is showing you what a contradiction looks like when you trace it carefully, and teaching you to trust that feeling of “this cannot be right” as the start of an investigation rather than the end of one.

What These Puzzles Train

Smullyan’s knights and knaves puzzles train a specific reasoning skill: tracking assumptions as you move through a chain of logic, and noticing when an assumption leads to a contradiction.

In any puzzle of this type, you start by assuming something — “suppose this person is a knight” — and follow the implications forward. If the implications eventually contradict each other, the assumption was wrong. You backtrack, flip the assumption, and try again. The contradiction is not a failure. It is diagnostic information pointing you back to the broken premise.

His backwards-reasoning puzzles sharpen this further. Smullyan often presents a conclusion and asks what must have led to it. You cannot move forward from given information — you have to reconstruct the chain that produced the outcome you are looking at. That reversal is uncomfortable. It is also one of the most useful thinking moves available when something has gone wrong.

Applying This to AI

When an AI gives you an output that seems wrong — a recommendation that does not fit, an explanation that contradicts something you know, a summary that cannot be reconciled with the source — the instinct is to treat the output as the problem and discard it.

Smullyan’s method suggests something more productive: treat the wrong output as a contradiction, and work backwards.

A contradiction means at least one assumption in the reasoning chain is false. The task is to find it. Start with the conclusion the AI reached, and ask: what would have to be true for this conclusion to follow? List those premises. Check each one against what you know. The broken one is your diagnosis.

This is exactly what happens in his island puzzles. You do not reject the puzzle when you hit a contradiction — you use the contradiction to eliminate possibilities and narrow down what must be true. The bewilderment is productive. It is pointing you somewhere.

The same applies to AI debugging. If an AI-generated summary contradicts a document you have read, do not just note “the AI got this wrong.” Ask: what would the model have needed to believe about the document for this summary to make sense? That question often reveals something about what the model was actually doing — misidentifying the main claim, conflating two separate points, relying on a template that did not fit the specific content.

Comfortable With Not Knowing

Smullyan’s greatest contribution to how people think was normalising productive confusion. His puzzles leave you temporarily bewildered. That bewilderment is not a sign you are doing it wrong — it is the point where the real reasoning starts.

AI outputs can create the same kind of confusion. Something looks right. Something feels off. You cannot immediately say which part is wrong. The instinct is to either accept the output or discard it. Neither is the reasoning move. The reasoning move is to sit with the confusion, identify where it lives, and trace it back to its source.

Smullyan showed that paradox and contradiction are not obstacles to clear thinking. They are tools for it.

---

The bridge: Smullyan’s method — assume, follow implications, find the contradiction, backtrack — gives you a procedure for diagnosing broken reasoning. The next section shows this method operating at scale, under time pressure, with real stakes: the Bletchley Park codebreakers used known facts to constrain possible explanations and eliminate impossible ones. That technique, called cribs, is directly applicable to AI output review.