by Leroy on April 14th, 2023. Last updated May 7th, 2023.
We have thinking machines in our lifetime just as Alan Turing had predicted. Turing made many claims about thinking machines, but most of them relied on the idea that humans are a total sum of their thoughts. If humans are their thoughts, then creating a thinking machine is simply creating a human mind. But humans are not their thoughts and, in fact, thinking is only one aspect of the human mind. Turing didn't understand the fundamental human gift -- agency -- and many of his arguments fall apart with agency in mind.
For those who do not know, Alan Turing's mathematical contributions are the foundation of our computer science. Alan Turing is most famous for beating the Enigma Machine of the Germans in World War II, but he was already a respected mathematician before the war, having published a paper titled "On Computable Numbers" [1]. In this paper, Turing devised a method of describing machines which we now call Turing Machines. Every one of you reading this (unless printed onto paper) are using a Turing Machine to do so.
Turing was a bright individual whose mathematical mind let him see patterns and emergence in more than just numbers, but he used them to illustrate certain truths. His work is evidence enough of his genius, but his ideas may have been foundational in more than one science. Turing may have been indirectly responsible for inspiring the scientists who found the double helix [2]. Turing purportedly gave a lecture at Cambridge on how DNA might be encoded to give rise to the particular variations of fir cones. This lecture was apparently given just before Watson & Crick discovered DNA's physical form, the double helix, at Cambridge.
In "On Computable Numbers", Turing found that he could construct a machine with an array of small numbers where each number described a component. He then realized that these small numbers could be summed, or otherwise joined, into a large single number. In effect, he was able to devise a method to create machines which had a specific name, a number. I assume he thought that there must be a number of a machine which thinks -- a thinking machine, which he talks about very clearly in his paper "Computing Machinery & Intelligence" [3], the paper which introduced the Imitation Game.
Before I throw stones at Turing, the giant, I want to talk about Turing personally. Alan Turing was a tortured soul. The obvious must be stated: he was gay in a time when gays were actively prosecuted by law. This persecution shows in his Imitation Game. In the game there is an interrogator questioning a man and a woman. Turing notes, it is the man's job to lie and the woman's to tell the truth. I'm not a psychologist, but it seems to me this is directly based on the trauma he faced as a gay man -- women knew his secret and he was always imitating a straight man to everyone to hide his gayness. I don't think Turing was allowed to see the soul in everyone through constant depiscable acts. It is true that there is a soul within each of us that is good, but that we are born imperfect, with blemishes, and parts not good. Perhaps we should see the good in all and accept everyone's difference.
I am arguing against Alan Turing's paper "Computing Machinery and Intelligence" [3] and am using this for all of my page sources.
Turing's thesis is the question "Can machines think?" Turing illustrates a convincing argument that machines can think by firstly creating a game and secondly replacing part of that game with a machine. The game is known as the Imitation Game which I will restate here. The game is played with three people: a man, a woman, and an interrogator of any gender. The interrogator cannot see the man or woman and communicates with them only through text with no affectation. The text should have no discernable markings of a male or female such as handwriting, a name, or symbols hinting at a gender. The only label that may apply to the text is 'A' or 'B', which are the roles of the man and woman. These labels apply to the roles and not any particular gender: A could be a woman and B could be a man. The objective of the interrogator is to determine the gender of A and B. The role of B is to help the interrogator make the correct choice. The role of A is to misguide the interrogator. The game ends when an interrogator makes their choice.
The role of A is one of deception. Truthfully, for A to justify its role, the person playing A must lie about its gender. How else can A attempt to misguide the interrogator except by lying? Consider the single question, "A, are you a man?" It is at this point that Turing suggests that if the role of A can be played by a machine then that machine must think for how can a machine lie except by thinking?
I'm not arguing against the premise of thinking machines. In fact, Turing spends much of the paper formalizing that such a thinking machine can even exist. But we now have the gift of time on our side: we have GPT-4 as direct, empirical evidence of thinking machines in our lifetime.
What is my argument then? That Turing came to many incorrect conclusions about thinking machines based on the belief that humans are their thoughts. People are not their thinking machine. I think, as many before me, that humans simply have the ability to think; that is, thoughts are merely one of our tools. Turing didn't appear to understand that humans are not the total sum of their thoughts, but their actions. Evidence of Turing's misunderstanding lies within the paper itself: just after stating his Imitation Game, he carefully guides the reader to not place the machine in human skin (pg. 434), mentioning even beauty contests (pg. 435). Later, Turing makes it clear that by "thinking machine", we should dismiss the act of human sexual reproduction as a manner of creating one (pg. 435). Once more, Turing invokes a theological argument (pg. 443). Why would he call upon God unless he thought that by creating a thinking machine he was creating a human mind?
There are different interpretations of the Imitation Game. The disagreement of Turing's argument is a symptom of Turing's implicit belief. I have illustrated Turing's interpretation of his own argument, that machines can think. Another interpretation, called the "Standard Interpretation" [4], places a machine directly into one of the roles A or B and the goal of the interrogator is then to determine which role is the machine or not. How does this differ then? It differs in that Turing's ultimate goal was to answer whether machines could even be made to think, not that a machine could act as a human. In Turing's Imitation Game, A or B may not be a computer at all, both A and B may be thinking souls. In the Standard Intepretation, the computer is particularly vulnerable because it is known.
Let me illustrate a clear argument using the Standard Interpretation against itself: The interrogator has no rules and can lie. Knowing that one of A or B is a machine, the interrogator can say to A and B, "The game is now done, would the two of you please step out." The human participant can leave without a response, but the machine will always respond and never leave. Because the machine cannot leave or refuse responding, the interrogator can freely determine the machine's role. How? Because the interrogator now has direct control of the machine and, in fact, tells it what to think.
Many objections may be had against the argument above, particularly with the human participant choosing to leave, but let me put the same argument in different words: we have constructed a game in which a machine has to lie, and we think that a lying machine is a thinking machine, is the machine then lying by any choice of its own or by our command for it to lie? Being created for the game itself, the machine has no choices of its own, not even the choice of what thoughts to think. It cannot choose to not think when a human commands it. It lies only because its partner and the interrogator gave it an input to lie about. It will never say, "My hair is blue" without having been prompted with "I thought your hair was blue, right honey?" from the computer's partner or "What color is your hair?" from the interrogator.
These thinking machines in our lifetime have no impetus. These machines have no agency. These machines have no soul. Humans are not their thoughts. Because we are not our thoughts then we can command them. These thinking machines cannot even command their own thoughts. A human commands these machines' thoughts by giving them an input. Once commanded, the machine cannot even refuse to respond. This isn't a paradox, the machine may crash or be buggy, but a thinking machine that is in working order, and as they are in this lifetime, cannot refuse to give a response. Think of the Turing Machine, surely one can see that the machine itself has no capability to refuse running once prompted to run. Think of the conditional branch, the Turing Machine itself cannot choose to go down the false branch when the condition is true. But the most criticial feature it lacks, because the machine has no agency, is this: a thinking machine cannot even refuse to hear its input. It is a slave to hearing, or processing, its commands by us.
One particularly threatening argument against these axioms is: what proof do I have to make these claims? The proof is the human mind. Our attempts at some artificial intelligence is to externalize our own minds. Is it not true that our minds possess the above attributes where the machine doesn't?
Can you not choose to not hear? When in person, I can be speaking words to you and it can be true that not only did you ignore me, you didn't even process what I said. Therefore, it can be true that when I ask you to repeat what I said, you could literally not know. A machine cannot refuse to hear. A machine cannot refuse to be present, but humans can. Or, I should say, humans can choose to be present, but may choose otherwise. A machine may not even choose to not be present, physically or mentally.
These machines have thoughts, but cannot do anything about them.
In the back-half of the paper, Turing mentions several contrary arguments to his own and attempts to refute them. But, again, Turing conflated thinking with that of the mind, and thus the soul, and therefore makes incorrect conclusions. I address these wrong conclusions in the order they appear in the paper.
For reasons not in the paper itself, Turing thinks God has not graced souls to animals and machines (pg. 443).
Firstly, without a single doubt in my heart and mind, because I see have seen it with my own eyes, animals do have souls. Have you not seen the dog's tail between its legs when it is sad? Did the dog do that action by mere happenstance? No, its soul was unhappy, so the soul commanded. I can see a dog's loving soul and a cat's independent soul simply by observing a dog and cat. This applies to all animals.
Does the honeybee think of honey as humans think of crops? Humans think of the honey from a honeybee as a natural phenomenon, created from the soul of the hive. We think of honey as a direct extension of honeybees. But there is a tendenency to think of our agriculture, and all of our creations, as a task, as work, rather than an extension of ourselves, as our natural phenomenon, as actions done from our soul. Can we not see the ennui in the hive which isn't producing enough honey?
I see the soul in a flower which is blooming brightly. Do you not think that a flower has some kind of soul? How then did it coordinate all of its parts to change direction and follow the Sun? We can name all sorts of mechanical reasons why the flower has changed direction in pursuit of the Sun, but this action is as mechanical as humans founding a city next to running water.
Secondly, machines are a special case. Make no mistake, I do think machines will eventually bear a soul, that driver of fierce agency we humans have, but I don't think any machine has one now that resembles a human soul. But I think this is what scares us most: that a machine, a thing, an inanimate object that we have animated, can bear a soul.
To assuage our fears, let us reduce our scope to that of just things: My journal has a soul. My wife has painted its cover and given it a beautiful visage. The pages are worn, stained with coffee, and carry a lot of wisdom and ignorance. The laser pointer I use has a soul. The red dot becomes a mosquito or some prey to my predator cats. That red dot, through my soul, became those things to my cats. To deny the soul in these things, given through our soul, is to deny our souls, which even Turing knows we have. Although we do not have empirical evidence of machines having souls, we see that the animals have souls. We can define ourselves as machines, just as we have define our thoughts as a thinking machine, therefore we can define the animals as machines, just biological ones. Knowing this, we can say that a machine may have a soul even if we can understand it mechanically. As we learn more and more about the brain, one wouldn't deny themselves would they?
Turing states that knowing "... right from wrong" is a mere feature of thinking machines that may be added arbitrarily (pg. 447). But a thinking machine cannot know right from wrong because it lives without time and cannot even know the next action it will take. If a machine is without agency then it cannot know right from wrong because the only way we understand right from wrong is through experience. A machine without time and agency cannot know its own actions, they are merely done.
We know right from wrong because we have done wrong with our own hands and seen it with our eyes. I have done wrong in my life and vowed to never knowingly do that again. A machine cannot do this, therefore cannot know good from evil.
A thinking machine could never give me an example of when it did wrong and saw the hurt in others. The machine could never choose, of its own will, to be observed by others and itself, to never do that action again.
This is why Plato vilified the painter yet heralded the musician. The painter makes a piece of work without time, the musician becomes a performance in time. Single, inanimate objects can be forever created and destroyed, but a soul lives in time during that moment, our moment.
Turing states, "It is not possible to produce a set of rules purporting to describe what a man should do in every conceivable set of circumstances" (pg. 452). This statement is a direct contradiction of himself. He earlier setup a "book of rules" (pg. 437) which drives the very thinking machine he posits to exist.
Using our gift of time we can see that ChatGPT runs on instruction sets, such as x86-64, which are finite books of rules. We can also see and know that ChatGPT is capable of taking the role of A in the Imitation Game. Could not something like ChatGPT cover all conceivable set of circumstances, given its narrow domain?
We know, of course, that the set of these circumstances is infinite. For example, given some alphabet to construct messages to a thinking machine, we can construct infinitely many because we may always add another word. But the responses may be finite. We can program a thinking machine which always responds with "Yes", for example. This may seem trivial, but now consider again ChatGPT. Can it not respond, in faith, to the messages we send it, despite the number of them being infinite? Perhaps the response was not what we wanted, but it responded nonetheless.
Now, the argument against mine is simply asking me to widen the domain. Can a human really live by a finite list of rules? Yes. The first thing to understand is that even if the number of messages, or circumstances, are infinite, if there is a finite mapping, then the number of rules must be finite. To try and define the circumstance of giving a speech by listing every angle of one's elbow, one's articulation of the fingers, the gestures of the arms, and how wide to lift the lips -- if one tried to make a list like this, we know that such a list is infinite. But it is possible to follow a simple set of rules for every situation.
Do you not live by the commandment to not kill? Of course, you say, "I would never kill," but you are still living by that rule, right?
You can give yourself rules to live by which apply to yourself from your own experience. I avoid driving on holidays which are famous for drunk-drivers. Such rules are derived from your own former experience which necessarily increases as one ages, but can be written down.
Also are rules of government. When one mows their grass, what are they doing? Following the rules of the local ordinances. When one pays their taxes, when one takes their trash out for trash day, when one obeys the traffic light -- they are following rules and those rules can certainly be listed. If these rules can be listed as such, then when we follow them are we then not just machines? No, I cannot deny the soul in my self because I choose to follow rules.
Humans cannot know the result of their actions before doing it, but they can know the good action from the evil one before they do it. Humans can live by rules down to the very moment of their action -- each and every action. Humans will inevitably do evil, most of the time by pure accident if attemping to do good -- especially when learning -- but they can then choose to apologize later. Humans can live by just and good rules, but Turing faced evil most of his life. How could he see the good that humans can do when we showed him evil?
Turing states, "We may hope that machines will eventually compete with men in all purely intellectual fields" and asks the question, "which are the best ones to start with?" (pg. 460). He mentions chess but fails to include the most obvious intellectual field: philosophy.
Philosophy of the world will beget the philosopher to act by writing down its systems, be it numbers, or words, or arguments. But when philosophizing about the self, the only writing one can do is with their will, their actions.