Can Computers Think or Do We Think That They Think?
The possibility of the ability that the machines can think appears to be carrying the interest of people. One of the most important scientists of all time Alan Turing discusses this subject in his paper called Computer Machinery and Intelligence (Turing, 1950). He supports the claim that machines can think under certain circumstances (Turing, 1950). Since he has written his paper there were major changes in computer science, now the profession of machine learning is being advanced, there are even subbranches of machine learning like deep learning. In the current age, it is now possible to call these machines AI. AI seems to support Turing’s claim about the possibility of thinking machines, getting closer to his claim. Contrary to this sense, in this paper, I will be having the opposite view, that it is not possible for machines to think. First, I will present my argument, and provide support for it for my proof. After my proof, I will present some possible objections to my claims. And finally, I will conclude my paper. Now I will present my argument.
Humans can think
To think, one must be creative.
AI is not creative.
Therefore AI cannot think.
Before I start evaluating the premises, I will have the assumption, that the procedure of learning is a prerequisite for thinking, or one can think in the procedure of learning. What I try to say is that one cannot think about a subject before s/he is aware of the concepts about it. For example, as an objection one can say that, I can think about stellar objects, that I do not know any specifications about them, or even I do not know which planet comes after earth, but I can still think about them. The way that I argue is that this person is aware of the concepts, like the existence of them, or the possibility of the existence of them, therefore s/he is able to think about stellar objects, and it can follow that if the learning procedure initiates, s/he will be thinking more about stellar objects. Now I can continue with the premises.
Now I will look at the first and second premises of my argument. I will evaluate them together, as it is the best way to express my standpoint, hence for my evaluation I will merge these two claims into one “Humans are creative when they think.”. I have used the word “creative”. This word can be interpreted differently among different individuals; therefore I will now explain the meaning of this word in my claim. When I say that humans are creative when they think, it applies, that they include emotions, as well as make connections between the information they have learned. I further include that humans are creative when they are learning, because whilst learning, humans think. The idea that I am presenting can be further narrowed down. Whenever a human thinks about a concept that s/he has learned before or s/he is currently finding out, the person is creative, as s/he makes some connections about this new or developing knowledge and includes their emotions. A related example of this phenomenon can be, the following. When a human is walking on the street, s/he should be walking on the pavement by the regulations, and assume that s/he is aware of this, which I have stated as the preliminary of learning. This does not necessarily follow, that this person will indeed obey the rule whenever s/he is walking on the street. Maybe there is a dangerous person walking on the pavement, that s/he thinks should be avoided, or sees a pothole and violates the rule, to avoid any harm coming to her/him, or maybe the person gets joy when s/he violates the rule and violates it. Now I will move to my third premise.
Now I will be evaluating my third premise; this premise will mostly emphasize why the claim made for humans in the previous part is not applicable to AI. Before I start, I want to state a possible objection, that one can argue that AI can learn. From a technical viewpoint, it can be said that AI can learn, that there exists the profession of machine learning, even more, specialized deep learning; and they can give feedback about the provided information. At the first glance this seems like the AI can learn and followingly think, but following the previous claims AI cannot think, because it lacks creativity and emotions, and it cannot make the required connections to be able to learn and think. To explain my claim I will now take a deeper and more technical view of the concept of machine learning. Let us imagine a mathematical function, for example let the function be “ f(x) = x^2”. In this function the provided input to this function, that is the value I put in this function will provide me with an output. For example, if I put “2” in this function, I will get “4”. I can keep on providing this function with different inputs, and I will get different or sometimes the same result. In a same way, I am not limited with this function, but I can write different functions like “f(x) = 2x”, or even very complex functions like “f(x) = x + (ln(x)^(x+1/x)) . (x^2 + 5x + 6)”, or even ones with multiple inputs like “f(x,y) = xy”. In a similar way, I can create machine learning models having different sizes of inputs, outputs, inner layers, complexity, etc. like mathematical functions. And after the model is provided with data, I can provide the model with inputs, and get outputs based on my input, like in the mathematical functions, for example, I can provide a picture of a dog as an input to the model, and it gives me an output based on the input that the image is an image of a dog or not. I think the last phase of machine learning is what makes people think that machines can learn and think, or in the future when they become more complex they will be learning and thinking, but as illustrated with the function example, the complexity provided is related to how the model is built by someone else, other than the model. To test my claim about the relationship between thinking and creativity, I want to use my example with the mathematical functions again. As I stated that the structure and functionality of AI can be represented with a single mathematical function. Now I want to make a compression between a human and AI. As stated, humans can make connections about different concepts in their minds, the concepts can be thought about different mathematical functions, which are constantly passing outputs to other functions as inputs. Similar to the dog example where the image of a dog in input and seeing it is the output, humans can say that that is a dog, and pass that output to another function, that they must be running away from that dog. But for the case of the AI, as it only consists of a single function, furthermore its structure is only a single function, it cannot go further than only identifying a dog. One can object, that in the example of the self-driving cars whenever a car identifies another car and stops accordingly, that functionality is actually not implemented in the model, but is an external functionality provided by sensors that is distinct from the model, coded individually by humans. This example can also prove that the previous example about walking on pavement will not apply to the AI, that seeing a pothole is similar, even the same as seeing a car like in the previous example. Now I will move on with the possible objections to my claims.
One can argue, “If humans are making connections in their learning procedures to be able to think, how did they collect their first knowledge, how did they start thinking?”. The way I respond to this is this. Of course, humans have to have some initial information to be able to form more complex functions in their minds. This is why humans are able to think more complexly as time advances. They build up information via thinking and forming new functions in their minds. Whereas AI lacks the ability to have more than one function.
One other objection can be that “What will happen if the AI has the functionality of creating other models?”. This response fails because the way to do so is to provide the structure of other functions as an input to the creator function like the composite functions in mathematics, putting aside the possibility to do this technique in computer science, even if it could be done, what is ended up with is another single function.
One last objection that I can think about can be that indeed if a single AI has only one function, we can use multiple AIs together to resemble the functionality of a human brain, that now the AI can make connections. This response also fails, because still, we would be in need of another AI to guide how to pass information to other functions. Let us assume we have one and the AI was able to think, then again we would need another guide function created by AI, which I objected to in the previous possible objection. Having talked about the possible objections, now I will move on to the conclusion.
In this paper, I claimed that AI cannot think, and presented my argument to prove my thesis. I have claimed that creativity is the prerequisite for thinking, in which making connections and emotions are important. I had the important assumption about the relationship between learning and thinking, I have also claimed that AI cannot learn. I have evaluated how humans are creative, but AI is not. The examples to support my claim about the pavement and mathematical functions were important. With these elements I have proven my claim and presented some possible objections, and how they fail.
Turing, A. M. (1950, October). COMPUTING MACHINERY AND INTELLIGENCE. Mind, Volume LIX(Issue 236), pp. 433-460.