Alan Turing on "Can Machines Think?"

Alan Turing is a famous machine learning and artificial intelligence (AI) contributor. In a recent book I have been reading, "The Quest for Artificial Intelligence", the author refers to Alan Turing's 1950 literature paper, "Computing Machinery and Intelligence", which addresses some of the critics of machinery intelligence. Of course, as a scientist, I read it thoroughly. Here is my opinion on the matter.

Link to Image Source Licence.

The paper starts with a section called "The Imitation Game", which sets the tone for what AI is made to do: imitate. Such sentiment is repeated throughout the literature paper, using words such as "mimic" and "made to do". In the very first paragraph of the very first section, Alan Turing writes:

I propose to consider the question, "Can machines think?" This should begin with definitions of the meaning of the terms "machine" and "think". The definitions might be framed so as to reflect so far as possible the normal use of the words, but this attitude is dangerous.

Alan Turing proposed that the words "machine" and "think" have a different meaning to those accustomed to our everyday language. On page 2, Turing writes:

May not machines carry out something which ought to be described as thinking but which is very different from what a man does?

This speaks to the core of AI perception at large. AI marketing is focused on using words related to human functionality. Still, even Alan Turing says this won't apply and that a different definition or word needs to be used. In his paper, Alan proposes that the word "think" change meaning so that some other word describes the general idea or understanding of human thought. But that is not feasible in the real world.

Whilst words play a different role in different contexts, saying machines think implies that a machine has the same functionality as a person who thinks. But Alan is not saying that! Instead, he is saying that what the machine is doing can be called thinking, but the process is different from how a man thinks—thus removing that personification that haunts the AI hype.

Moreover, Alan Turing says on page 3 that "the best strategy [of the machine when playing the imitation game] is to try to provide answers that would naturally be given by a man." But how would a machine be able to do what collective humanity can't except via yet undiscovered relations?

On the same page, Alan Turing writes:

We wish to allow the possibility that an engineer or team of engineers may construct a machine which works, but whose manner of operation cannot be satisfactorily described by its constructors because they have applied a method that is largely experimental.

To this, I pause! How can someone build something without understanding what they built? They might initially not understand precisely what went right, but documentation and reconstruction/repetition of the experiment ought to shed light on this, it is called The Scientific Method. I believe what Turing is hinting at here is the ability of a person calling themselves "engineers/data scientists" who prompts a model to build another model without understanding how the initial machine works, rather than constructing a machine from scratch.

Alan Turing continues on the same page by saying:

The short answer is that we are not asking whether all digital computers would do well in the [imitation] game... but whether there are imaginable computers which would do well.

Again, I pause. What is meant by imaginable is not clear.

On page 5, Alan Turing makes a powerful statement:

The reader must accept it as a fact that digital computers can be constructed... and that they can in fact mimic the actions of a human computer very closely.

This is true. That is Artificial Intelligence, a software constructed to mimic the actions of a human. Alan Turing continues to explain:

If one wants to make a machine mimic the behaviour of the human computer in some complex operation one has to ask him how it is done, then translate the answer into the form of an instruction table. Constructing instruction tables is usually described as "programming".

We all are aware of this, and despite playing with words on whether machines "think", Alan Turing has identified what needs to happen for AI to exist.

Later on, in Section 6: Contrary Views on the Main Question, Alan Turing addresses contrary views to "Can machines think?" and in his responses, one notes again the idea that what we mean by "thinking" today is not what will be meant by "thinking" in the context of machines. He writes

The original question, "Can machines think?" I believe to be too meaningless to deserve discussion. Nevertheless I believe that at the end of the century, the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.

Oxford Dictionary defines "think" as

- to have a particular idea or opinion about something/somebody; to believe something - to use your mind to consider something, to form connected ideas, to try to solve problems, etc.

In other words, the word "think" is still primarily and universally used as always and thus, the verb "to think" and "machine" should not be used together.

The fourth argument in Section 6 relates to "The Argument from Consciousness", in which Professor Jefferson wrote "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 is had written it. No mechanism could feel pleasure at its successes, grief when its valves fuse, be warmed by flattery, made miserable by its mistakes".

Alan Turing's counterargument was that the only way we can know that a person can think is by being that person and that the acceptance that everyone thinks is just a "polite convention". In my opinion, this is not an intelligent argument. While I hand it to Alan Turing that not every person feels the same way, it all boils down to foundational facts. Pain is described more or less in the same way, even if its tolerance differs from one person to another. Alan Turing himself claims on page 13, "Possibly a machine might be made to enjoy this delicious dish, but any attempt to make one do so would be idiotic." In his own words, Alan Turing admits that for a machine to enjoy a sense of taste, it must be "made to" and such effort would be "idiotic".

Continuing on the fourth argument of section 6, Alan Turing explains that

The [imitation] game is frequently used in practice under the name of viva voce to discover whether someone really understands something or has "learnt it parrot fashion".

I couldn't agree more, but then he asked what critics would say: "If the sonnet-writing machine was able to answer like this in the viva voce?" I am unsure whether "answer like this" means in a flowing manner or intuitive. In today's AI, there is ample evidence that it can answer conversationally but not intuitively. Proof of this is Alan Turing's statement on page 13 "When a false proposition is typed we say that the machine has committed an error of conclusion". How beautiful is that statement! "Error of conclusion", not"Hallucinations."

I think one of the strongest counterarguments that hold to this date is what Alan Turing refers to as "Lady Lovelace's Objection", which states

The Analytical Engine has no pretensions to originate anything. It can do whatever we know how to order it to perform.

I copied the italics as referred to in Alan Turing's paper "Her Italics". Alan Turing quotes Hartree (1949), who explains that such current state 'does not imply that it may not be possible to construct electronic equipment which will "think for itself".' Note that "think for itself" is in parenthesis as quoted by Hartree, thus again showing that the verb 'to think' is perceived to have a different meaning than to what humans do. Evidence of this is Alan Turing's statement saying

The Analytical Engine was a universal digital computer, so that, if its storage capacity and speed were adequate, it could by suitable programming be made to mimic the machine in question.

Again, I pause and encourage readers to re-read "suitable programming be made to mimic the machine", which machine is referring to the same one in "Can machines think?".

Alan Turing repeatedly showed in his own words that a machine's ability to think and learn can only be confirmed if the definitions of "think" and "learn" have different meanings from how humans think or learn. Turing continued to counterargue that claiming "machines can never do anything really new" compares to claiming "there is nothing new under the sun."

I pause! How much did science progress, almost seemingly out of thin air? But can a machine think outside the data and logic it is trained on? There is ample evidence that AI will continue to obey Newton's first law of motion so that without new data and retraining, it will reproduce the same output.

Turing compared the learning of a machine with education, which is true. Education plants a seed, but as nature shows time and time again, that seed has the power to grow, adapt and be destroyed. If we leave it alone, the cycle may repeat itself or not. But I also want to emphasise and celebrate the groundbreaking discoveries that other people have made, which made digital computers possible. Whilst they might have been inspired by other work, they have taken that education above and beyond. After all, reading "The Quest for Artificial Intelligence" proves that curiosity drives innovation.

I begin concluding the article by sharing that Alan Turing anticipated the concept of "Reinforcement Learning" and others. He also predicted that:

Most of the programmes which we can put into the machine will result in it doing something that we cannot make sense (if at all, or which we regard as completely random behaviour.

This is already happening with AI, and are dubbed as "hallucinations". In light of the above, Alan Turing suggests that

Intelligent behaviour presumably consists in a departure from the completely disciplined behaviour involved in computation, but a rather slight one, which does not give rise to random behaviour, or to pointless repetitive loops.

Again, Alan shares that in favour of machines that "can think" or are "intelligent", one must depart from the current definitions of such words.

Over the years, languages have evolved, which is part of our social evolution. Nevertheless, claiming "machines can think" and putting such a statement out in the world where only a fraction of the world's population knows the ins and outs of the technology and an even smaller fraction understands that the word "think" has a different meaning has led to a dark cloud hovering over the AI because it is expected to behave using verbs universally understood as meaning one thing, but have been applied to AI for doing another thing. AI marketers strive on such opportunities, claiming AI "speaks", "talks", "thinks", "hallucinates", "draws" and "designs". However, Alan Turing's legacy has given a rich philosophical foundation for methodically discussing words and phrases before attaching them to technology, specifically AI.