Many people think that science will eventually be able to explain everything that happens in nature, and that technology will be able to reproduce it. Perhaps that is so, but even then, that day lies far into the future. Probably a more likely scenario is that the further science and technology advance, the deeper the mysteries of the world will grow. Even with topics that we believe science has solved for good, when you take a closer look, you'll find that plenty of problems have slipped through the cracks or been swept under the carpet. Furthermore, these are often the issues that are closest to us and most important in our daily lives. Take hunches or intuitions or premonitions, for example. They may have rational-sounding explanations, but our gut feelings tell us something is not quite right after all. Such examples are not at all uncommon. When you think about it, there are lots of things that modern civilization has forgotten all about. Maybe the time has come to stop for a moment and try to remember. The seeds of forthcoming science and technology are impatiently waiting to be discovered among the things we have left behind.
When it comes down to it, deep learning is just a kind of artificial selection, I think. You show the system various examples, and it outputs answers according to the boundary conditions. You then mark the answers as right or wrong, and improve the judgment criteria. In that respect, the process is very similar to the process of “stabilizing the genome by selecting the phenotype it produces.” It is better to think that you are evolving the computer rather than assembling algorithms. In that way, we should be able to think in parallel about the problems indicated by AI and the question of when animals first acquired intelligence.
The way animals respond to light and sounds and other stimuli from the external world might be considered the prototype of “intelligence.” At first, a reflecting sphere of simple neurons create a simple logic circuit, misconnected units are eliminated, and as the process goes on, increasingly subtle circuits are created, a second and third stage of logic is added, until the system can respond even to various complex conditions. The human brain is said to consist of high-order neurons, but in the end it boils down to a mass of interneurons. As to whether there is a distinction between reflexes and thoughts and intelligence, and from what stage you can call it intelligence, the boundaries are gradual. Strictly speaking, I don’t think there is any boundary that slugs crawl along or humans move along.
Whenever a cabinet minister or celebrity makes a gaffe, they regret it wondering “why did I ever say such a thing?” It happens to all of us, all the time. Such utterances are almost a reflex, and it doesn’t look like the person is thinking at all. The phrase “I can’t believe I did that” means that there was no time for the consciousness or ego or whatever to monitor what he or she was doing. It may very well be part of their personality, but in any case both personality and intelligence could almost be considered extensions of the reflexes in a broad sense. From this perspective, it is doubtful that AI’s “intelligence” really is intelligent. On the contrary, it’s the same with human “intelligence.” The workings of the cerebrum and neurons of Andrew Wiles when he solved Fermat’s Last Theorem are profoundly interesting, but on the other hand there is the guy at the station kicking the ticket vending machine and yelling “Gonna keep my change, are you? Motherfucker!” at it. Both are using human brains. It’s even extremely uncertain how much we humans really use our brains or intelligence in our daily work. Human creativity is often mentioned as something machines can’t do, but if that was the only thing considered human work, most people be incapable of working, including me. Certainly, if AI keeps developing, eventually it will be able to take over most of the things that humans can do, I believe. Humans will become beings served by machines. One day, we might even become like the Eloi of the year 802,701 in H.G. Wells’s The Time Machine.
The Turing Test, which is used to determine whether the other party is a human or a machine, depends heavily on the intelligence of the human doing the testing. Some people are easily deceived, others are tougher. In the end, what we judge is whether the other party has some sort of self that feels the way ours do, whether they experience a process going on in their heads. In short, if we think we have established communication with the other party, it is only an illusion to begin with, even if the other party is actually human, and what the test passes judgment on is ultimately whether the illusion holds or not.
What we experience as intelligence is communication-based mutual understanding, two-way comprehension, reciprocal agreement. Paradoxically, “deception” works the same. If the other party can deceive, they must have the corresponding level of intelligence. In that case we have a program that estimates the recipient’s reactions to its messages, and works its mechanisms in a process to actually fool the recipient. If we call that intelligence, then the genetic program of mimicking insects is also a form of intelligence. There may be no real-time communication, but there is a kind of a logical system for survival and its purpose is achieved when the result is transmitted to the receiver. The development program, so to speak, of mimicking insects, is actually very similar to intelligence algorithms. There is a clearwing moth in Japan called Sesia yezoensis. Although it’s weak, it has the temerity to mimic a wasp. It’s a moth, so it’s head is not as big as a wasp’s. Instead, the small-headed moth puts its front legs next to its head to make it look larger. The result is that when it flies, it really resembles a wasp to an astonishing degree. They are practically indistinguishable. Even the buzzing of the wings is the same. Thanks to the developmental program or neural network-like thing that controls its behavior and morphology, we reliably receive a certain message and are taken in by the moth’s deception. It may be a one-way deception, but something like communication is established here. That’s why I think mimicry and the adaptability of various animals that have emerged over the course of evolution might be considered a form of intelligence. The animals don’t have that “intention,” of course, but it sure looks that way. It may not be intelligence in the sense that they are not guided by a subjective self or ego, but whether the other party is conscious or not is something only the other party knows. It is the same with other people. We assume they are conscious, and only shy or angry.
In the manga Ghost in the Shell there was a scene where a Tachikoma (a thinking armored vehicle) brings out Zeno’s paradoxes to tease another AI: “Don’t you even understand the paradox of self-reference? What a primitive AI you are.” The question whether an AI can understand self-referentiality is similar to the question whether an AI can understand jokes. Even if it’s possible for an AI to learn jokes, it will probably never get the human tendency in the background to play around with logical systems. We can’t talk about intelligence without mentioning the “very human irresponsibility” of jumping between logics.
Do humans really use their “intelligence” that much in their daily lives in the first place? On reflection, most people spend the whole day without thinking at all, or without thinking profoundly. Even researchers go to their laboratory, start up their computers, read their mail, and respond to what they are told. They sometimes write documents, but what they are going to write has already been decided in any case. It may be a chore, but it doesn’t require any intelligence. You do have to use your brains in order to move forward efficiently with your work, but the work itself is just a repetition of simple procedures, and before you know it, it’s evening and time to go home. If you get hungry in the middle of these activities, you eat or make a cup of coffee. Even when you make a cup of coffee, your body moves almost by reflex. You already know where everything is. Once in a while you run out of filters, and then for the first time you have to think. That is when your intelligence is activated. In other words, as long as things are moving along as usual, we don’t think at all. That is why we sometimes get anxious that we forgot to turn off the gas or lock the front door when we went out. We don’t think about such actions each and every time. The same is true when we talk. We don’t speak by carefully selecting grammatical patterns and styles all the time. To put it another way, how many moments are there in human behavior when we really use intelligence? Both research and most other jobs just follow the steps in the flowchart. This doesn’t require intelligence but rather attention in order to avoid omissions. For sure, I really have to use my head when I’m thinking up new experiments or writing papers, but large parts of the work is still routine. A paper is basically a report and could be written matter-of-factly, but after all I want to win over my readers and I want the paper to make an impact. Desire comes into play. And that is when your creativity is tested at last.
According to the behavioral ecologist Mariko Hasegawa, humans have a weirdly high intelligence, much higher than we need to live our lives. The brain’s functions are abnormal, she claims. For other parts of the body, the story is simple. In order to break a femur, for instance, you need more than double the force of the maximum stress that acts upon the bone as part of the living body. Without such a structure, sprains and torn muscles would occur all the time. So that strength is really necessary. Even this kind of double capacity is cutting it close. Unfortunately human intelligence can’t be explained in the same way. As I just mentioned, we don’t really need a lot of intelligence to just live our lives. The brain functions can’t be explained by a need for double capacity. In fact, ordinarily the brain doesn’t do very much. Going to the supermarket, deciding what to buy today, and adding up the cost is about as hard as it gets. That is the arithmetic of ordinary people; even most scientists don’t use much more mathematics than that. I myself, for instance, never use calculus or partial differential equations. And yet, with a bit of knowledge people can somehow understand Einstein’s theory of relativity. The interesting thing is that this ability is completely unnecessary for daily life. According to Professor Hasegawa, the reason why our powers are so much higher than we need to live our lives is that they have actually evolved due to us racking our brains thinking about how to attract the opposite sex. When you think about it, the things people worry about the most in their lives are politics and human relationships, fights with their rivals, and acquiring sexual partners. We make our partners angry, regret it, realize what we did wrong, and conceal our intentions. Such experiences refined our brains. This became a target for selection. It seems that our intelligence was more about human relations and communications than simply survival. We could also consider the brain as a generator of delusions that fulfill our desires, or appear to fulfill our desires. The strangely high level of human intelligence depends to a large extent on the delusions that fulfill our uniquely human desires. We create other people inside ourselves, make analogies and simulations, and try out various things.
It is often claimed that it was thanks to our intelligence that we were able to invent bows and arrows and slay mammoths, but human intelligence is on a completely different level. If that was all there were, nobody would have come up with a solution to Fermat’s Last Theorem, ever. Human intelligence far exceeds double capacity, to an extent that at first glance seems redundant. Although humans don’t behave intellectually under ordinary circumstances, they are actually abnormally smart.