Someone once sat in front of a keyboard in a room somewhere, most likely in the basement of a university building, with fluorescent lights and a water-stained ceiling tile in the corner. They were genuinely unable to tell if they were speaking to a machine or a human. On the surface, that moment seems unremarkable, but it is at the heart of one of the longest and most contentious debates in scientific history. In 1950, it began. It’s not over yet.
Instead of publishing “Computing Machinery and Intelligence” as a triumphant proclamation, Alan Turing did so almost as a means of avoiding an annoying question. He wrote, “Can machines think?” before quickly declaring that the question was pointless.
| Category | Detail |
|---|---|
| Originator of the Test | Alan Turing (1912–1954), British mathematician and logician |
| Seminal Paper | Computing Machinery and Intelligence, published 1950 |
| Original Institution | University of Manchester, England |
| Core Concept | The “Imitation Game” — can a machine convince a human interrogator it is human? |
| First IQ Test Developer | Alfred Binet (1857–1911), France — designed to identify children needing educational support |
| IQ Formula Pioneer | William Stern (1871–1938) — introduced Mental Age ÷ Chronological Age × 100 |
| Notable Practical Test | Loebner Prize Competition, first held in Boston in 1991; repeated at the Royal Society, London in 2014 |
| Top 1991 Competitor | PC Therapist — inspired by ELIZA, a 1960s language simulation program |
| Top 2014 Competitor | Eugene Goostman — simulated a 13-year-old Ukrainian boy |
| Key Critic of 2014 Result | Moshe Vardi, who argued the test “says little about machine intelligence” |
| Stanford IQ Adaptation | Lewis Terman (1877–1956), Stanford University — adapted Binet’s test for American adults |
| Further Reading | Stanford Encyclopedia of Philosophy – Turing Test |
Instead, he suggested a game that was sophisticated, a little theatrical, and incredibly ambiguous in ways that still give people headaches. A human interrogator. Two people in private. One machine, one human. Type your inquiries. Try to distinguish between them. It sounds easy. It’s never been easy.
Turing made numerous revisions to the game in his original paper, which is something that is easy to overlook. Scholars have discovered at least two different versions within those pages—what some now refer to as the “Original Imitation Game” and the “Standard Turing Test”—and there has been sincere, persistent disagreement over which one Turing truly intended to be taken seriously. The initial version included a gender dimension, according to some academics.
Others completely disregarded it. The most widely read version eliminated all nuance and focused on a simpler query: is it possible for a machine to respond to questions in a way that is identical to that of a human? That version continues to be the topic of debate at conferences more than 70 years later.
Depending on your personality, the actual attempts to administer the test have been either fascinating or depressing. A program called PC Therapist, which functioned similarly to ELIZA in the 1960s by repeating a patient’s own words back to them with enough variation to pass for understanding, won first place in the 1991 Loebner Prize Competition in Boston.
It wasn’t comprehension. It was a mirror wearing a lab coat. Eugene Goostman, a program that pretended to be a Ukrainian adolescent with poor English and little cultural awareness, won the 2014 edition, which took place at the Royal Society in London. It passed the Turing test, according to the organizers. Naturally, the academic community disagreed.
If you take a far enough step back, the Eugene Goostman episode has an almost comical quality. The machine’s intelligence did not help it win. It prevailed because it had an inherent justification for its lack of persuasiveness. It is reasonable to assume that a thirteen-year-old Odessa resident who does not speak English as their first language can say nearly anything and be blamed for being young and awkward in a second language.
That may indicate more about human credulity than machine intelligence, or it may indicate that the test was not well-designed to detect the difference. One of the more outspoken critics, Moshe Vardi, was unimpressed. He contended that the experiment’s specifics only supported his view that the Turing test provides us with frustratingly little insight into the true capabilities of machines.
And yet it continues. Even after more than 70 years of criticism, the test continues to appear in press releases from tech companies, philosophy curricula, and newspaper headlines whenever a new chatbot is introduced. That’s for a reason. Regardless of its shortcomings, the imitation game resonated with a genuine concern about the boundary between thought and performance, between mimicking intelligence and truly possessing it.
Naturally, Turing was not the first to raise the issue of intelligence measurement. It dates back at least to Sir Francis Galton, who established a laboratory in London in the late 1800s and attempted to gauge noblemen’s mental capacity by observing their reaction times and sensory capacities. He asked a serious question and developed testable hypotheses that later researchers could dissect and refine, even though he was wrong about nearly everything.
Alfred Binet came next, creating questions for French students in an effort to determine who required additional assistance. The formula for IQ was provided by William Stern. Lewis Terman expanded the system to include adults after bringing it all to Stanford. Intelligence testing had reached an industrial scale by the time the US military began sorting recruits using Alpha and Beta tests during World War I.
The fundamental disagreement was never resolved. Is intelligence primarily inherited? Does the environment shape it? Are you able to distinguish between the two? Hundreds of millions of dollars have been spent by researchers in an effort to pinpoint particular genes associated with cognitive ability. Decades later, the honest conclusion is that if genes matter, which they most likely do to some extent, thousands of them contribute minuscule pieces of variance, and none of them function independently of context. Genes react to their surroundings.
Cognitive function is impacted by nutrition. In some circumstances, an additional year of education significantly raises IQ scores. The notion that intelligence is a fixed biological quantity that is set in stone at birth is untenable.
It seems like scientists are continuously creating measuring devices for something they haven’t completely defined as they watch all of this unfold over the years. One of the first psychology historians, Edwin Boring, put it bluntly: the tests measure intelligence. Depending on your point of view, that could be either a profound observation or a devastating admission of circularity. Most likely, both.
For his part, Turing was more cautious than is typically acknowledged. He wasn’t saying that his imitation game proved anything. He was putting forth the idea as a thought exercise, a means of refining the query rather than providing an answer.
He kept changing the game’s parameters, testing it against criticisms from other thinkers, and honing it as a conceptual tool, according to academics who have closely examined the original paper. It was never intended to be a science fair competition with a winner and scoreboard. Perhaps the most bizarre result of the experiment is that it has been.
From Galton’s reaction-time labs to Eugene Goostman’s adopted adolescent persona, the entire plot of this tale implies that intelligence defies the classifications we create for it. Every time someone creates a precise definition and a trustworthy tool, something falls between the cracks. Whether that is a feature of the device itself or a problem with our tools is still unknown.
Perhaps intelligence, whether it comes from a machine or a human, is more of a relationship between a mind and the world it is attempting to understand. Though he never expressed it directly, Turing had a suspicion.
