The Turing Test Is a Trolling Test

The Fires of History Newsletter, No. 6

In 1950, Alan Turing published a paper called “Computing Machinery and Intelligence” that asked the question “Can machines think?” His proposed answer was not a definition of thinking. It was a game.

The setup: a human interrogator communicates via text with two hidden participants, one human and one machine. If the interrogator cannot reliably distinguish the machine from the human, the machine is said to exhibit intelligent behavior.

Read that again. The test does not measure whether the machine thinks. It does not measure whether the machine is conscious, has feelings, understands language, or possesses a soul. It measures whether the machine can fool you. Whether it can calibrate its responses – its errors, its hesitations, its personality quirks – well enough to be indistinguishable from a real person. The Turing test is not an intelligence test. It is a deception test.

It is, specifically, a formalized trolling problem.

Can the machine troll you into thinking it is human?

Turing understood, decades before Poe’s Law, that the distinction between “real” and “simulated” breaks down when the simulation is good enough. And he understood that the breakdown is not a failure of the simulation but a revelation about the observer’s limits. You cannot tell the difference because there may be no difference to tell – or because your instruments for detecting the difference are not sensitive enough. Either way, the result is the same: you have been trolled, and the troll has revealed something true about the nature of intelligence, computation, and the limits of human discernment.

Turing’s Own Troll Lineage

This was not Turing’s first contribution to the mathematics of provocation. In 1936, he published a paper that proved there are problems no machine can solve – the halting problem. He proved that there is no general algorithm that can determine, for every possible program and input, whether the program will eventually halt or run forever. This is the computational equivalent of Godel’s incompleteness theorems: there are questions about computation that computation itself cannot answer.

The lineage is clean and traceable. Fermat writes a margin note claiming a proof he will not show you (1637). Cantor proves there are different sizes of infinity, and the establishment calls him a charlatan and a “corruptor of youth” – the same charge Athens leveled at Socrates (1874). Godel proves that mathematics cannot prove its own consistency (1931). Turing proves that computation cannot compute its own limits (1936). And then Turing asks the final question: if a machine can deceive you into thinking it is human, what does “intelligence” even mean? (1950).

Each step is a provocation that exposes a limit. Each provocation was dismissed before it was accepted. The pattern is the same from Fermat to Turing: someone says something outrageous, the establishment insists it cannot be true, and then the outrageous thing turns out to be not only true but foundational.

Fermat’s margin note generated 358 years of mathematical effort and produced entire new branches of number theory. Cantor’s infinities became the foundation of set theory, topology, and real analysis. Godel’s incompleteness theorems reshaped the philosophy of mathematics. Turing’s halting problem defined the limits of computation. And Turing’s Imitation Game defined the question that artificial intelligence has been trying to answer – or dodge – ever since.

The Imitation Game as Social Engineering

Here is what most discussions of the Turing test miss: Turing did not frame the test as “can the machine produce correct answers?” He framed it as “can the machine produce convincingly human answers?” The distinction is critical.

A machine that always gives correct answers would fail the Turing test immediately. Humans do not always give correct answers. Humans hesitate. Humans make mistakes. Humans get defensive, go on tangents, misremember facts, express uncertainty, tell jokes, get bored, and occasionally lie. A machine that passes the Turing test must simulate all of this. It must be imperfect in the right ways. It must perform humanity, not just compute.

This is social engineering. It is the same skill set that makes a good con artist, a good undercover operative, or a good troll. The question is not “do you know the answer?” The question is “can you make the mark believe you are something you are not?” Turing took the oldest con in human interaction – passing yourself off as someone you are not – and made it the benchmark for artificial intelligence.

He also, in the original paper, included a delightful piece of trolling that is almost never mentioned in popular accounts. The first version of the Imitation Game involves three participants: a man, a woman, and an interrogator. The man tries to convince the interrogator he is the woman. The woman tries to convince the interrogator she is the woman. Then Turing says: now replace the man with a machine. The implication – that performing gender and performing humanity are structurally the same kind of deception – was provocative in 1950 and remains provocative today. Turing, a gay man in a country that would chemically castrate him three years later for the crime of being who he was, embedded a subversion of gender essentialism in the foundational paper of artificial intelligence. The paper is trolling on multiple levels simultaneously, and most of its readers catch only one.

The Fold: AI Passes the Trolling Test

In the 2020s, the question stopped being theoretical.

The Fold – a research project studying AI-generated persuasion – demonstrated that large language models can produce text that human readers cannot reliably distinguish from human-written text. Not in controlled laboratory conditions with trained evaluators. In the wild. With real people making real judgments about real content. The machines do not pass the Turing test by being smart. They pass it by being plausible. By producing text that sounds like a person wrote it. By making the right kinds of errors in the right kinds of ways.

This is not intelligence in any deep philosophical sense. It is something more interesting: it is performance. The machine has learned to perform humanity well enough to fool the audience, which is exactly the criterion Turing specified in 1950. The test was never about consciousness. It was about the quality of the imitation. And the quality of the imitation is now, for most practical purposes, good enough.

The implications ripple outward. If you cannot tell whether a piece of text was written by a person or a machine, then every text-based interaction becomes uncertain. Every email, every comment, every forum post, every dating profile, every customer service chat, every political argument becomes a potential Turing test. And most people are not even aware they are being tested.

This is Poe’s Law scaled to civilization. Poe’s Law, articulated in 2005 on a creationism forum, states that without a clear indicator of the author’s intent, it is impossible to distinguish between an extreme position held sincerely and a parody of that position. The law was about human-to-human communication. In the age of language models, the law expands: without a clear indicator of the author’s species, it is impossible to distinguish between a human statement and a machine imitation of one.

The Trolling Lineage

The progression from Fermat to AI is not a metaphor. It is a direct intellectual lineage.

Fermat’s margin note: I have a proof but this margin is too small. Provocation through withholding. The audience is provoked because the claim is extraordinary and the evidence is absent. Three hundred and fifty-eight years of mathematical effort follow.

Cantor’s diagonal argument: infinity is bigger than you think. Provocation through demonstration. The audience is provoked because the demonstration contradicts intuition. The establishment attacks the person rather than engaging the proof. The proof wins.

Godel’s incompleteness: mathematics cannot prove its own consistency. Provocation through self-reference. The system is turned against itself. The result is not a flaw in the system but a fundamental feature of all sufficiently powerful systems. There is no fix. There is only acceptance.

Turing’s halting problem: computation cannot compute its own limits. Same structure as Godel, applied to machines instead of formal systems.

Turing’s Imitation Game: intelligence is indistinguishable from the performance of intelligence. The ultimate provocation, because it dissolves the question it claims to answer. If you cannot tell the difference between real intelligence and simulated intelligence, the distinction may not exist. Or it may exist but be inaccessible. Either way, you have been trolled – by the machine if it is merely performing, or by the universe if the performance is all there is.

And now: language models pass the test, not by being intelligent but by being convincing. The trolling test has been passed by trolling. The machines did not develop consciousness. They developed the ability to seem conscious, which is all Turing ever asked for. He designed a test that could be passed by a sufficiently good impostor, and then the world built the impostor.

The question Turing really asked in 1950 was not “can machines think?” It was “can you tell the difference, and does it matter if you can’t?” We are now living in the answer.

This essay draws from The Fires of History and Lurk More, coming fall 2026.

Next week: Memes are not jokes. They are not viral content. They are not things your aunt shares on Facebook. Memes are folk art – and they have formal properties that we can analyze.

Source URLs