sciencebriefs
9:04in productionCh. 1 · From 'Can machines think?' to 'Can they fool us?'/ 9:04 · ceiling 15 min
Computing & AI

Turing test

Alan Turing

The Turing test doesn’t ask whether machines think—it asks whether we’ve stopped noticing the difference.

The Turing test is a precise, behaviour-first redefinition of intelligence—not proof of thought, but a standard for functional equivalence in dialogue. It endures because it is simple, testable, and ruthlessly limited.

Computing & AIArtificial intelligenceComputingConsciousnessArtificial intelligence across the network →
Chapters & takeaways4
  1. 1:10
    From 'Can machines think?' to 'Can they fool us?'

    Turing replaced an unanswerable philosophical question with a concrete, behaviour-based experiment.

  2. 2:33
    How the test actually works

    The imitation game is a structured party game—adapted into a text-only conversation judged solely on indistinguishability.

  3. 4:11
    What the test measures (and what it ignores)

    It measures resemblance—not truth, logic, knowledge, or consciousness—only performance-level equivalence.

  4. 5:35
    The overlooked educational hypothesis

    Turing proposed education of a simulated child mind—not pre-programmed adult competence—as the path to machine intelligence.

Worth your time?

Yes. Study the whole thing.

4.5/ 5
What works
  • replaces ambiguous metaphysics with testable behaviour
  • remains the most widely adopted benchmark in AI history
  • forces clarity on what 'intelligence' means in practice
What does not
  • measures correctness
  • requires truthfulness
  • implies understanding
  • tests nonverbal intelligence directly
Study it if
  • AI researchers defining evaluation criteria
  • philosophers examining operational definitions
  • developers building conversational systems
Skip it if
  • neuroscientists studying cognition
  • psychologists measuring intelligence
  • engineers building autonomous robots
The written brief1 min read

What the work claims

A machine can be said to ‘think’ if a human interrogator cannot distinguish it from a human through conversation. Intelligence is defined not by internal process but by external, observable equivalence in dialogue.

How it was done

Turing framed the test as a three-person ‘imitation game’—a human interrogator tries to distinguish a man and a woman by text-only questions. He then substituted a machine for one player. The modern version uses a text transcript of conversation between a human and a machine; the machine passes if the evaluator cannot reliably tell them apart.

What holds up

The core claim holds: replacing ‘Can machines think?’ with ‘Can a machine imitate human conversational behaviour indistinguishably?’ is a valid operational pivot. The test measures resemblance in performance capacity—and that metric remains coherent, replicable, and bounded.

What does not

It does not measure correctness, truth, reasoning, self-awareness, or internal states. It establishes no mechanism for intelligence. It says nothing about learning, embodiment, or generalisation beyond conversational indistinguishability.

Why it matters beyond the lab

It shifted AI from metaphysics to engineering: intelligence became something measurable, falsifiable, and contestable in real time. Its legacy is not in passing the test—but in forcing every claim about machine intelligence to confront the gap between mimicry and meaning.

Is it worth your time

Yes—if you care how we define intelligence, measure it, or avoid mistaking fluency for understanding. It remains the most widely cited operational definition in AI, despite its narrow behavioural basis.

Same field · Computing & AI4 of 46
9:52
Dartmouth workshop1956The Dartmouth workshop founded AI as a field by naming it and framing its ambition — not by delivering results.
9:19
Evolutionary computationEvolutionary computation is a family of global optimisation algorithms inspired by biological evolution. It originated conceptually with Alan Turing’s 1948 genetic search proposal. It emerged formally in the 1950s–1960s through independent developments: evolutionary programming (1962, USA), evolution strategies (1964, Germany), and genetic algorithms (1960s, USA). By the 1990s, the unifying term 'evolutionary computation' was coined, and genetic programming emerged as a fourth major branch. The approach is used both for engineering optimization and as an in silico tool in evolutionary biology.
13:00
Transformer (deep learning)2017The 2017 paper Attention Is All You Need, from a team of researchers at Google, proposed the transformer architecture, which replaced the step-by-step recurrent processing used by earlier language models with a self-attention mechanism letting every part of an input be processed against every other part at once, making the whole computation far more parallelisable on modern hardware. That parallelism is a major reason transformers could be scaled up to the very large models trained since, including BERT in 2018, the GPT series beginning that same year, and the wave of systems from GPT-3 in 2020 through ChatGPT's public release in late 2022 and beyond. These large language models are typically built in two stages, an initial pretraining phase in which the model learns to predict the next word across enormous quantities of text, followed by a fine-tuning phase, often using human feedback, that shapes the model into a more instructable assistant. Even as capabilities have grown substantially, well-documented limitations persist, including confidently stated factual errors known as hallucinations, weaknesses in multi-step planning and reasoning, and reliance on human-selected training data whose biases the models can reproduce, and researchers remain divided over how much of what these models do reflects genuine understanding versus sophisticated pattern recombination.
13:00
Perceptron1957Frank Rosenblatt, working at Cornell Aeronautical Laboratory, built the perceptron, a machine that adjusted its own connection weights from labelled examples rather than being explicitly programmed. First simulated on an IBM computer in 1957 and then built as physical hardware, the Mark I Perceptron, by 1960, it drew wildly overstated press coverage about machines that would soon think and reproduce themselves. In 1969, Marvin Minsky and Seymour Papert published a proof that a single layer of the kind Rosenblatt had built could not learn certain patterns, most famously exclusive-or, and although their book acknowledged that multiple layers could solve the same problems, the narrower message was widely read as a verdict against neural networks generally, contributing to roughly a decade of reduced funding and interest before the approach was revived with multilayer networks trained by backpropagation.
Up next in Science

Ultracentrifuge

· 9:44

It didn’t reveal life’s blueprint — it gave biologists the first tool to hold molecules still enough to weigh them.

Play nextBack to Computing & AI
9:44