In this blog post, I’d like to explore the limits of human mental processes and artificial intelligence, and consider whether robots can possess intelligence and emotions similar to those of humans.
Humankind has always imagined beings resembling ourselves and sought to create them. This can be seen in various murals and sculptures, and in many stories, imaginary beings are depicted as having human-like qualities. In particular, the myth of Pygmalion—in which a statue transforms into a human—clearly illustrates that humans possess an instinct for self-imitation and creativity. In modern times, humans are creating not only creations with human-like physical characteristics—such as statues and dolls—but also robots capable of mimicking human behavior. Consequently, as robots replace humans in various tasks in the real world, their role is growing increasingly significant. Along with this, research on robots is being conducted very actively, and related technologies are advancing rapidly. But just how far can robots actually evolve? In the future, will robots possess intelligence and emotions similar to—or even surpassing—those of humans? Looking at robots in movies, the range of what we can imagine is gradually expanding, from robots primarily tasked with gathering and processing information—such as C-3PO in “Star Wars”—to the humanoids featured in “The Terminator.” Ultimately, films like “A.I.” and “I, Robot,” which endows robots with human emotions such as sadness, anguish, and joy, many people believe that robots will possess human-like artificial intelligence in the not-too-distant future. Predicting the future is certainly difficult. However, I do not believe that robots can perfectly mimic human mental activities.
First, I will discuss this from the perspective of the human mind. To do so, we need to consider the human condition. Let’s examine a few aspects of the human condition. First, humans constantly reflect on themselves and the world around them in an effort to find their identity. In other words, they ponder what it means to be human. Second, humans possess free will. Free will refers to the driving force that enables humans to act according to their own intentions and value judgments without being constrained by external limitations. Since free will, by definition, is not believed to be influenced by physical factors, it cannot be viewed merely as the result of simple physical mechanisms. The reason that enables humans to judge and decide matters for themselves does not come from outside the human being but arises independently from within the mind. Third, humans accept contradictory elements and synthesize them to create something new. For example, humans feel joy and sorrow simultaneously, and while they hope for eternal life, they are also creating weapons capable of destroying one another. If there were an entity that satisfied the above characteristics, we could reasonably call it a human. However, robots cannot meet these conditions. Robots, which are implemented through algorithms, cannot possess free will nor reflect on their very existence. They merely execute given commands. No matter how sophisticated the algorithm is designed, since it operates according to predetermined rules and learned structures, whether it can realize natural intelligence in the same way as human mental processes remains a separate question. Humans naturally perceive abstract emotions such as loneliness and joy, and possess a free system of thought rather than a uniformly structured one that merely responds appropriately to them. While humans, like algorithms, have their own habits and ways of living, they possess the potential to change them at any time. Humans accept and connect contradictions without becoming overwhelmed, thereby creating something entirely new. If we were merely bound by algorithms and driven by stimuli, we could not even conceive of identity, freedom, or dignity. Humanity lies in the mind, and the mind cannot be created without identity and free will. I believe that a robot lacking such a mind can never become human.
On the other hand, some argue that the human condition discussed here can be achieved by enabling robots to feel human emotions, claiming that if robots are equipped with systems similar to human hormonal systems to experience emotions, it will be possible to create robots similar to humans. However, for a robot to possess human capabilities, it must be able to engage in mathematical reasoning rather than mere calculation; it must be able to write and translate not just individual words but complete sentences in context; and it must possess a mechanism capable of recognizing a wide variety of new patterns, rather than merely recognizing preprogrammed ones. Furthermore, rather than simply responding to stimuli with preprogrammed commands as they do now, robots must act independently; they must build their own logic and identity by influencing—and being influenced by—other humans and robots; and they must be capable of creating original works. Even if robots possess emotions, it is doubtful whether they could possess these abilities. Although it has recently been revealed that non-human animals also possess various emotions, this does not mean we can conclude that such animals have the ability to reason, create, or reflect on their own identity in the same way humans do. Furthermore, the claim that human emotions are simply governed by chemicals such as hormones lacks clear evidence. While it is true that human emotions are closely related to chemical and biological processes, humans are capable of regulating their emotions to a certain extent and choosing their actions based on the situation. Therefore, it is difficult to assume that simply creating a system that implements emotions would be sufficient to reproduce the entirety of human mental processes.
Next, I will examine the technical challenges involved in creating robots that resemble humans. From a hardware perspective, robots clearly possess superior capabilities in certain areas compared to humans. These can be studied relatively clearly by mimicking the human body or mechanically implementing specific functions. However, for a robot to possess intelligence comparable to that of a human, research into artificial intelligence must also be conducted simultaneously. The difficulty in artificial intelligence research stems from the fact that creating an artificial version of any natural object requires a precise understanding of that object’s characteristics. However, when studying human intelligence—unlike when studying the human body—it is difficult to clearly define what intelligence is or where to begin research based solely on the information we currently possess. Current artificial intelligence technology has evolved based on various machine learning methods, including artificial neural networks; while these are inspired by the human brain, they do not replicate the entire human thought process exactly. It is estimated that the human brain contains approximately 86 billion neurons, with an enormous number of connections between them. Furthermore, the human brain processes information simultaneously through the complex interactions of countless neurons. Therefore, it is difficult to assume that human mental processes can be faithfully reproduced simply by increasing a computer’s computational power. In a large-scale artificial neural network experiment conducted by Google in 2012, approximately 1,000 computers and 16,000 CPU cores were used to train the system to autonomously recognize objects such as cats and human faces from 10 million YouTube video images. While this experiment represented a significant achievement in artificial neural network research at the time, it did not fully replicate human intelligence. Rather, this example demonstrates that even implementing a specific AI capability requires significant computational resources.
In contrast, there is a view that, given sufficient time and resources, robots capable of human-like thinking will eventually be created in the distant future. However, creating such robots may not be merely a quantitative issue but rather a methodological one. In other words, the current limitations in AI robot technology lie not only in a lack of resources but also in the fundamental question of how to understand and implement human intelligence and consciousness. It is inaccurate to view the 1970s as the starting point of AI research. The field of artificial intelligence was established as a distinct area of study following the 1956 Dartmouth Conference, though research on machine intelligence and computation had been ongoing even before that. Since then, artificial intelligence has advanced significantly through various approaches—including symbolism, expert systems, connectionism, machine learning, and deep learning—and today demonstrates capabilities so powerful that they are difficult to compare with those of the past. Therefore, it is also inappropriate to view the difference between past and present artificial intelligence as merely a distinction between serial computation and parallel processing. Let’s examine the “Chinese Room,” a thought experiment devised to demonstrate the limitations of using the Turing Test to determine whether a machine possesses artificial intelligence. Inside the room is a person who does not speak Chinese. This person has only a look-up table that tells them how to respond to questions posed in Chinese. Although the person inside the room does not speak Chinese, an observer outside might mistakenly believe that the person inside understands the questions and is answering them. If we consider the person inside the room to be an AI, the point is that just because an AI mimics human-like responses does not necessarily mean it possesses human-level intelligence or understanding. In fact, the process of programming AI itself shares similarities with this “Chinese Room” thought experiment. Some argue that the “Chinese Room” theory is a thought experiment based on AI from the past and therefore cannot be applied to current AI. In fact, technology and methods in the field of AI have changed significantly over time; in particular, today’s AI utilizes large-scale data, neural networks, and deep learning to demonstrate capabilities that are difficult to compare with those of past systems. Nevertheless, the philosophical debate remains as to whether a machine processing symbols or information is the same as the machine actually understanding or being conscious of them.
A robot can exhibit behavior similar to human mental processes. However, this does not necessarily mean that a robot understands the meaning of those actions in the same way a human does. Humans possess many qualities—such as spontaneity, randomness, and contradiction—that are difficult to program logically and that distinguish them from robots. While today’s AI has advanced far beyond its predecessors and can perform tasks similar to human mental activity—such as understanding and generating language or solving complex problems—whether such capabilities are equivalent to human consciousness, self-awareness, or free will remains a separate issue. Therefore, while robots may physically resemble humans, we need to carefully consider whether they can ever mentally resemble humans, and I believe it is difficult to replicate human mental processes in exactly the same way.
