The Problem
What was Turing trying to solve?
Turing opens his paper with a sentence that sounds simple but is actually a trap: “I propose to consider the question, ‘Can machines think?’”
Then, immediately, he steps back. He says this question is too poorly defined to answer. The words “machine” and “think” are so vague, and so loaded with assumptions, that any attempt to answer the question directly would just become an argument about definitions. Everyone would talk past everyone else.
Here is the problem he identified: we use the word “thinking” to mean something that only we can observe in ourselves. When I say “I am thinking,” I know what I mean — I am experiencing something from the inside. But when I say “she is thinking,” I am making an inference. I cannot observe her experience directly. I observe her behaviour — what she says, how she responds, what she does — and from that behaviour, I conclude she is thinking.
So when we ask “can a machine think?”, what are we really asking? Are we asking about internal experience (does it feel like something to be the machine)? Or are we asking about behaviour (does the machine act in ways indistinguishable from a thinking being)?
These are completely different questions. And Turing argued that the first question — about internal experience — is unanswerable and possibly meaningless. We cannot even be certain that other humans have genuine inner experiences. We only infer it from their behaviour.
So Turing did something clever: he replaced the unanswerable question with an answerable one.
The problem in plain language
Instead of asking “Can machines think?”, Turing asked:
“Can a machine behave in a way that a human cannot distinguish from another human?”
This is still a hard question. But it is a testable one. You can run an experiment. You can get a yes or no answer. You can make progress.
In one sentence a Class 10 student can understand:
Can a machine talk to you in a way that makes you think you are talking to a person?
In technical language:
Can a machine produce natural language outputs that are behaviourally indistinguishable from those of a human interlocutor, as evaluated by a naive human judge in a blind comparison test?
Why did this problem matter so much?
Because the stakes were enormous.
If the answer is “yes, in principle a machine can pass as human”, then:
- There is nothing in human intelligence that is fundamentally non-mechanical
- The brain is a computing machine of some kind
- Building artificial intelligence is a matter of engineering, not philosophy
- The question is “how?” not “is it possible?”
If the answer is “no, a machine can never pass as human”, then:
- There is something in human intelligence that computation cannot replicate
- Perhaps consciousness, creativity, or emotion are genuinely non-physical
- Building true AI is either impossible or requires something we don’t yet understand
In 1950, almost everyone believed the answer was “no.” Turing was betting — and arguing carefully — that the answer was “yes.”
And he was right. In 2022, Google engineer Blake Lemoine publicly claimed that the language model LaMDA was sentient. In 2023, millions of people reported feeling that ChatGPT was genuinely understanding them. The question Turing raised in 1950 is not just alive — it is more urgent now than at any point in the last 75 years.
Why should a student in small-town India care about this?
Because you are living in the moment when this question is being answered in real time.
The AI systems being built today — the ones you can talk to, argue with, ask for help with your homework — are direct descendants of the ideas in this paper. When you use ChatGPT or Claude or any other AI assistant, you are, in a sense, participating in an ongoing Turing Test.
And the philosophical question underneath — what is intelligence? what is understanding? what makes a mind? — is one of the deepest questions human beings have ever asked. You do not need to be at IIT to think about it. You just need to be curious.
Next: The Core Idea →