Computing Machinery and Intelligence

Inside the paper

The problem

The question of whether machines can think is difficult to settle while its key terms remain open to dispute. Turing looks for a more concrete question that can be discussed without first agreeing on a definition of thought.

The idea

He proposes a conversational imitation game and asks whether a machine could do well in it. He then examines objections and discusses how a machine might acquire the abilities needed to participate.

How it works

• An interrogator exchanges written questions and answers with unseen participants and tries to tell which is human.
• The machine’s task is to answer in a way that makes this distinction difficult, while the human participant also responds.
• Turing reviews objections ranging from mathematical limitations to claims about consciousness and particular human abilities.
• His discussion of learning machines considers starting with a simpler system and changing it through training and experience.

What it showed

The paper offers a proposed test and a sustained argument about its implications, rather than an experimental demonstration that a machine passes it. It also makes a forecast about future performance and explains why Turing considers several common objections unconvincing.

Read it for

Begin with Sections 1 and 2 for the game and the proposed replacement question, then choose objections in Section 6 that interest you. Section 7 on learning machines is a useful second stop; the technical discussion of digital computers can be skimmed if you already know their basic operation.

Written in our own words from the paper itself. A draft the editors have not yet reviewed.