Abstract

This post summarizes Alan Turing's seminal paper on "thinking machines," the origins of the Imitation Game (now known as the Turing Test), the nine major objections to machine intelligence, and my own reflections on these arguments. I also briefly discuss the relevance of these ideas to modern AI.

1. Introduction: What Does It Mean for a Machine to Think?

"I propose to consider the question 'Can machines think?'"
— Alan Turing, 1950

That opening line from Turing's 1950 paper still gives me chills. It marks the starting point of all serious discussion about artificial intelligence.[1] Turing argued that the very definition of "thinking" is ambiguous (since everyone has their own criteria for what counts as machine thought). Instead, he proposed the Imitation Game as an experimental criterion: if a machine can communicate in such a way that a human cannot reliably distinguish it from another human, we may say the machine "thinks."

2. The Origin of the Imitation Game (Turing Test)

Turing devised the Imitation Game to experimentally test whether a machine can imitate human thought. If a human interrogator cannot tell whether they are conversing with a machine or a person, Turing argued, the machine should be considered "thinking." This was the first formal proposal of what we now call the Turing Test, and it remains a foundational concept in AI research.

3. The Structure and Universality of Digital Computers

Turing identified three essential components of a digital computer:

Memory (RAM, HDD)
Computation (arithmetic logic unit, ALU)
A set of programmed rules (CPU, instruction set)

If these three elements are present, Turing argued, even the complexity of human thought could be imitated. He also introduced the concept of a Universal (Programmable) Machine, capable of executing any computable function, not just a single fixed purpose.[2] Charles Babbage also attempted to create a programmable machine (the Analytical Engine), though he never completed it.

4. Nine Objections to Thinking Machines and Modern Perspectives

Turing's paper systematically addresses nine major objections to the idea of machine intelligence. Here, I summarize each objection and add some of my own thoughts:

Theological Objection: Thinking is a uniquely human faculty given by God.
Turing's reply: Science does not concern itself with the divine.
Heads in the Sand: The idea that machines could think is too frightening, so we simply refuse to believe it.
History shows (e.g., the Luddites) that technological progress cannot be stopped by fear or denial.
Mathematical Objection (Gödel's Incompleteness): Discrete State Machines (DSMs) have inherent limitations.
Turing noted that these limitations apply to humans as well. But what about quantum computers? If a quantum computer is measured, the state collapses and becomes discrete, but before measurement, it exists in a superposition. Does Gödel's theorem apply? I wonder if quantum computers might eventually overcome this objection.
Consciousness: We cannot know whether a machine is truly thinking or just following instructions.
Turing: Unless we can become the machine ourselves, this is unanswerable and not worth debating.
Various Disabilities (Emotion, Creativity, etc.): Machines cannot feel, create, or appreciate art as humans do.
Yet, modern AI can create art, music, and literature that moves people. Like in the movie "Bicentennial Man," perhaps machines will one day learn to feel and love.
Lady Lovelace's Objection: Machines only do what we tell them; they cannot originate anything new.
Modern AI, with bottom-up learning and Hebb’s rule, is beginning to show signs of creativity.
The Argument from Continuity in the Nervous System: The human brain is continuous, while computers are discrete, so computers cannot imitate it.
Turing: If the result is the same, does the implementation matter?
The Argument from Informality of Behavior: Human behavior is too complex and informal to be captured by rules.
Modern AI, using bottom-up learning and Hebb's rule, is gradually overcoming this.
Extrasensory Perception (ESP): Machines cannot have supernatural abilities.

Turing answered each objection logically, and the mathematical limitations he discussed are still relevant today, especially as quantum computing advances.

5. Conclusion: AI Today and the Turing Test

To my knowledge, there is still no officially recognized AI that has perfectly passed the Turing Test. However, with the recent advances in LLMs (Large Language Models), I think we are reaching a point where distinguishing between human and machine is becoming practically impossible.[3]
This paper was a fascinating read—neither too difficult nor too technical, and the first to propose the concept of the Imitation Game.

[AI #1] In-Depth Analysis of "Computing Machinery and Intelligence"

Table of contents