The rise of digital computers motivated scientists to reason about the possibilities and limitations of such machines. In 1936, Church and Turing independently proved that there are problems which are not possible to solve with a digital computer. Other questions remained unsolved, one of which still concerns society today: Can machines think?
In 1950, Turing published “Computing machinery and intelligence”. In it, he proposed a practical test which is to this day synonym with testing machine intelligence: the Turing test. But the paper offers more. It gives an overview of common objections against Artificial Intelligence. Many of them are still brought up today. Others get weaker and weaker with every new challenge computers tackle. And it is the paper which makes Turing one of the founding fathers of Artificial Intelligence, as he proposes to teach machines just as we would teach a child.
“Computing Machinery and Intelligence” is a seminal paper, incredibly ahead of its time and most of all: still worth a read. As it is one of the earliest works in computer science and Artificial Intelligence, it is approachable without much background knowledge. But it also touches so many aspects that even the more AI-advanced readers will be blown away by. This blog post will only give an overview of the paper; I can only encourage everybody to pick it up and read it for yourself.
Before we dive into the paper itself, it is worth considering the time the paper was published.
In 1950, the term ‘computer’ most commonly described humans, women mostly. It was a job description for somebody who did computations, using computation tables. Human computers reached their peak during the world wars. But with the rise of digital computers, their downfall began. Digital computers soon outperformed human computers and in 1954 most humans were already replaced by digital computers.
Turing’s paper was published just in this transition time. Two years prior to the first electronic computer, the Manchester Mark I ran for the first time. But it was still humans who did the most of the computations. His paper, therefore, contains several sections explaining the notion of a digital computer to the reader and describing it as a machine following a human-computer rule book.
Can machines think?
To answer this question, the terms ‘thinking’ and ‘machines’ would have to be explained. But instead of doing so, Turing proposes to replace the question altogether. Instead of asking whether machines can think, we should be asking if the machine can succeed to win in the imitation game.
The imitation game
The imitation game requires three players, a woman (A), a man (B) and an interrogator (C). A and B are separated from C in a different room. C’s objective in the game is to figure out which of the two players (X and Y for him) is female. C may ask any question to A and B, which they have to answer. A tries to fool the interrogator by either pretending to be male or by suggesting that B is actually female. B on the other side will try to help C to correctly identify the woman. All communication is done in written form so that the voice will not give the sex of X and Y away.
Turing then proposes to replace A by a computer. Is there an imaginable digital machine which is able to fool a human interrogator as often as a woman is able to fool a human judge that she is a man? In Turing’s eyes, this is the question we ought to ask.
What does it take to win the imitation game?
From today’s perspective, one can see that a number of AI sub-disciplines would be needed for a computer to participate in the imitation game.
First of all, it would need Natural Language Processing to communicate with the interrogator. Some form of knowledge representation and automated reasoning would be needed to reason about the questions asked and provide satisfying answers. Lastly, machine learning will be necessary so that the machine can adapt to new circumstances.
Taking a step back here, Turing wrote his paper when even the notion of a digital computer was foreign to the majority of people. I assume today’s reader is familiar with those machines, but there is more to Turing’s excursion to computing.
You might have heard about Turing machines. In a prior paper, Turing proved that every definite-state machine can mimic every definite-state machine’s behaviour given enough resources. All computers are basically the same. This means the computation unit in your microwave could, with the right code and enough resources, do the same computations as your fancy MacBook pro.
Following this logic, if we can imagine one Turing complete machine (as digital computers are) to succeed in the imitation game, we will have answered the question for all digital computers. And Turing believes that such machines are possible.
Turing anticipates that not everyone will agree with either the replacement of the original question or with the possibility that machines can think. He addresses theological, mathematical and even objections from extrasensory objections (he actually takes ESP more serious than religion).
A few quite interesting ones are the argument from consciousness, the objection of various disabilities and the Lady Lovelace objection.
The argument of consciousness
For this argument, Turing quotes Professor Jefferson who wrote in 1949: “Not until a machine can write a sonnet or compose a concerto because of thoughts and emotions felt, and not by the chance fall of symbols, could we agree that machine equals brain—that is, not only write it but know that it had written it. No mechanism could feel (and not merely artificially signal, an easy contrivance) pleasure at its successes, grief when its valves fuse, be warmed by flattery, be made miserable by its mistakes, be charmed by sex, be angry or depressed when it cannot get what it wants.”
Turing takes this quote as a justification for his imitation game. In his eye, the question-answer approach of the imitation game does allow the interrogator to ask the machine questions about art, feelings and perceptions. As stated before, it does not matter if the machine is actually able to feel the way we humans feel. It only matters if it can make humans believe it feels and perceives as we do.
The argument from various disabilities
This argument is a particularly interesting one, as it gets weakened with every advancement in the field of AI. The argument claims that a machine will never be able to do X, where Turing states X to be:
Be kind, resourceful, beautiful, friendly (p. 448), have initiative, have a sense of humour, tell right from wrong, make mistakes (p. 448), fall in love, enjoy strawberries and cream (p. 448), make someone fall in love with it, learn from experience (pp. 456 f.), use words properly, be the subject of its own thought (p. 449), have as much diversity of behaviour as a man, do something really new (p. 450). (Some of these disabilities are given special consideration as indicated by the page numbers.)
Page numbers refer to the original publication in Mind, Volume LIX, Issue 236, October 1950e
From today’s perspective, we see, however, that machines proved they can do some of the X’s. It seems less likely that they won’t be able to achieve the others. They will just never be able to be human.
Of course, there is also the mathematical objection. There are certain things which are mathematically impossible for a machine to solve. Turing claims that humans are also not flawless and frequently make mistakes. So even if a computer cannot solve the Entscheidungsproblem, they could be programmed to give a random answer and pretend to have done a mistake in case they were mistaken.
Lady Lovelace’s objection
Lady Lovelace is commonly considered to have been the first programmer. She wrote algorithms for the “Analytical Machine”, a mechanical computer. In her memoir written in 1842, she states “The Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform”.
But what if we tell the machine to learn and originate? In the refutation of this argument, Turing introduces learning machines, which he dedicated a whole chapter in his paper.
Turing admits that he cannot make a sufficient refutation of Lady Lovelace’s objection. But he proposes a machine which would a) give a solution to the imitation game and b) address Lady Lovelace’s objection.
In his eyes, creating a machine that does well in the imitation game is merely a question of the program. He believes that the computational requirement will improve in the future, that they will not restrict the aim.
The real challenge will be to write a program that can fool a human interrogator. He proposes to go back to the roots of what shapes the intelligence and ability of the human mind:
- the initial state of the mind at birth
- the education the human obtains
- and the experiences made
Instead of trying to write a program, mimicking an adult brain, we should mimic a child’s brain: a blank notebook with only a couple of inference mechanism, which allows the machine to learn.
The child-machine would obtain an education by a teacher as a normal child would do. Turing believes that this education process would probably take as long as for a human child. It would get rewarded if it produced good results and answered correctly. It would get punished for disobeying and wrong answers. He basically proposes to implement a machine that can learn. Or slightly reformulated, he proposes Machine Learning. Or for the real nerds: Reinforcement Learning.
Turing expects that the teacher will not always know what exactly happens in the child’s mind, much as a teacher cannot always tell what the human pupil thinks. And if we cannot tell what happens inside the machine when it learns, how could we possibly have told it to do exactly this action? So if we can make a machine learn as Turing proposes, it would do something we did not specifically tell it to do, which would refute Lady Lovelace’s argument.
From the Imitation Game to the Turing Test
What Turing proposed as the imitation game has in some form become a practical approach for measuring intelligence. Turing believed that by the end of the 20th century, nobody would be contradicted when stating that machines think.
We are not quite there yet. No machine has officially passed the Turing test. Eugene Goldman, an AI pretending to be a Russian boy has fooled judges by claiming that his bad communication skills are due to his inferior English level. But expert’s opinion is that it was cheating the game.
As Turing said, winning the imitation game is a matter of the right program. And the AI world did not focus on producing the right program for passing the Turing test. The focus lays on solving real-world problems. We have computers that can detect cancer, personal digital assistants, robots that help improve the social skills of autistic children and even sex robots.
Turing asked, whether there is any imaginable machine which would do well in the imitation game. We are not there yet. But given the advances in the field of AI, most people would be able to imagine such a machine.
“Computing Machine and Intelligence” helped to lay the foundation of what is today known as Artificial Intelligence. It is yet another example of Turing’s genius.
You can find the original paper here: https://academic.oup.com/mind/article/LIX/236/433/986238. Give it a try!
Turing, A. M. (1950). I.—Computing Machinery and Intelligenec. Mind, LIX(236), 433–460. https://doi.org/10.1093/mind/LIX.236.433
Russell, S., & Norvig, P. (2009). Artificial Intelligence: A Modern Approach (3rd ed.). Prentice Hall Press.
Oppy, Graham; Dowe, David, “The Turing Test”, __The Stanford Encyclopedia of Philosophy (Fall 2020 Edition)__, Edward N. Zalta (ed.), URL= https://plato.stanford.edu/archives/fall2020/entries/turing-test/.
Grier, D. A. (2001). The human computer and the birth of the information age. Joseph Henry Lecture of the Philosophical Soc. of Washington, 42-43.
Warwick, K., & Shah, H. (2015). Human misidentification in Turing tests. Journal of Experimental & Theoretical Artificial Intelligence, 27(2), 123–135. https://doi.org/10.1080/0952813X.2014.921734
Ewald, W. and Sieg, W., 2013. David Hilbert’s Lectures on the Foundations of Arithmetic and Logic 1917-1933. Springer Berlin Heidelberg.