AlphaGo’s Go battle with Lee Sedol is now underway, and artificial intelligence has won 3:0; two honor matches remain, but the outcome is basically beyond doubt already.
Whatever the final score may be, the historical significance of this match is beyond question.
The significance of this match is mainly symbolic. In fact, since AlphaGo already defeated the European Go champion last year, proving that its Go-playing strength had reached the level of a professional player, its total victory over humans was only a matter of time. Of course, we might also say that ever since Deep Blue dominated chess, the AI conquest of Go had only been a “matter of time.”
But at the time, even people who were optimistic about artificial intelligence technology did not expect that this “time” would be so short. Many people thought that at least decades, perhaps even a century or more, would have to pass before this day arrived, and some even believed that it would never come. And the first major significance of AlphaGo lies in showing the astonishing “speed” of AI’s development.
Artificial intelligence has long been a hot topic for philosophers of technology as well, and Dreyfus’s famous critique of AI, What Computers Can’t Do, is a classic work in phenomenological philosophy of technology. In fact, this book contains many profound ideas; the problem is that philosophers are often too presumptuous, trying to make definitive statements and predictions about the development of technology, and that is liable to come back and slap them in the face. Dreyfus’s face was basically swollen up from the beating, although if he had been able to change the title of the book to What Computers Have Difficulty Doing, it would have been much better.
Personally, I hold a relatively optimistic view of the prospects for the development of artificial intelligence, but I have no intention of making any predictions. Whether artificial intelligence can be realized, and when and in what form it may be realized, are questions for technical experts; the proper way to pose the philosopher’s question is: what does artificial intelligence mean?
To discuss the meaning of artificial intelligence, it is not necessary to assume that artificial intelligence is destined to be realized. This is rather like how we can discuss the meaning of God and the kingdom of heaven, or discuss the meaning of utopia and communism, without necessarily believing in them.
“Why does artificial intelligence make people afraid?” is a good question. Of course, many people are not afraid at all. Some of them firmly believe that strong AI is impossible to realize; others believe that the development of AI will only benefit humankind, and that there is nothing to fear.
Both of these attitudes are somewhat naive. First, the trend of AI technology continually developing has almost become inevitable. Perhaps the eventual turn of the tables, the “Skynet” scenario in which it masters and dominates humanity, will not become reality, but in many more cases the “realization” of AI is a process in progress, not a finished state; AI is already being continually realized. Second, even if you believe that the result of AI will surely be beneficial to humankind, that does not mean it cannot be frightening. We can say that the development of firearms is a good thing, but that does not mean you should not be afraid of a musket; you can say that atomic energy overall benefits humankind, but that does not mean you need not fear the atomic bomb and nuclear leaks. Optimism is one thing; ignorance is another. Only the ignorant can be fearless. New technology may indeed be able to benefit humankind, but the prerequisite is that humankind is able to understand and bear its dangers. To dodge the frightful aspects of it, or even to denounce everyone who shows a wary attitude as ignorant, that is the greater ignorance.
The development of technology and the history of science have a similar structure (this is a view I hope to elaborate further). It is not the case that embracing innovation unconditionally is always a good thing, nor is clinging to the old always bad. We cannot give up vigilance and reflection just because the development of artificial intelligence is the general trend.
The history of technology is also a history of ideas. Every technological transformation involves, to a greater or lesser extent, an ideological transformation, and artificial intelligence is certainly no exception. As the name suggests, the core concepts involved in artificial intelligence are “artificial” and “intelligence,” and these two concepts are precisely bound up with human self-understanding.
We generally believe that the reason human beings stand out above all things is nothing more than “a keen mind and deft hands,” that is, brain and hands. In many situations, brain and hand are unified: human beings rely on the wise brain to understand all things, and then use dexterous hands to control and transform all things. Especially in the modern age, when science and technology propel each other forward, the expansion of technology has also greatly inflated people’s confidence and sense of superiority. People seem to feel that they can remake heaven and earth and do anything whatsoever.
But intelligence and power encounter a “contradiction” here with artificial intelligence: can intelligent human beings create, with their own hands, an intelligence that surpasses themselves? This question is almost a blasphemy, like asking: can the omnipotent God create a being that stands above Himself?
“To make a being that stands above oneself” is precisely the original and ultimate pursuit of human technological history. Technology is an “extension” of human beings; every technology is an externalization, fixation, and strengthening of some human capacity.
A hammer is an extension of the fist; using a hammer to smash something is certainly more forceful than using one’s fist, and there is nothing strange about that—or rather, that is exactly what “technology” is supposed to mean. Of course, the development of technology in turn causes our bodily functions to atrophy or be diminished. For example, the more weapons develop, the more hand-to-hand fighting ability declines; brute-force close combat is either dismissed as barbaric and backward, or becomes art or game, and is no longer practical.
This characteristic of technology—its externalization and fixation—reveals a kind of contradiction: on the one hand, the development of technology has unquestionably enhanced and expanded human capacities; on the other hand, because such enhancement is often borne by technological objects external to the human body, once human beings are deprived of dependence on these “external things,” it can also be said that the development of technology has weakened and diminished human capacities.
So, for human beings, does technology enhance or weaken, inflate or diminish? When Plato discussed how “writing damages memory,” this difficulty had already been placed before philosophers, and until today, the question has only become sharper.
Things like writing, weapons, and the like—these ancient technologies—already possessed a certain degree of independence. That is to say, besides being controlled by humans, they also had a kind of inertia that was “not shifted by human will.” That is why Qin Shihuang burned books and buried scholars, and confiscated the weapons of the realm in order to cast bronze figures. Because so long as these books and weapons continued to circulate in the world, they possessed a certain force. Although these forces ultimately still have to be brought into play through the will of specific operators, in a sense the will of those operators is itself also brought about by these technological objects.
Of course, the “independence” of these ancient technologies is not very conspicuous, because after all they still require human beings to manipulate them; otherwise they are just a pile of dead things. Of course, the power of technological objects does not necessarily need to be exerted only when they are in motion. For instance, city walls, buildings, tombstones, and the like: by simply standing there unmoving, they guide and even govern human behavior and ideas, and an atomic bomb lying in a cellar is even more capable of influencing global order. But after all, in the eyes of ordinary people, these objects are always “passive” in relation to human beings.
When did we start counting technological objects as acquiring a certain “initiative”? The most emblematic development is mechanical technology. From the earlier waterwheel, windmill, and so on, to the mechanical clock of the late Middle Ages, and all the way to steam engines, textile machines, and factory assembly lines. The new characteristic of these machines is that, apart from steps like winding them, adding raw materials or energy, and so forth, they are relatively independent in the course of normal operation; they are detached from human beings and run on their own according to their own rhythm.
In textile machines and assembly lines, human participation is still a necessary link, but in these mechanical activities, humans do not play the role of operators, but of power sources—the key point in what Marx called the Industrial Revolution. Under the operation of large mechanical systems such as factory assembly lines, human beings are not the wise creators at all; on the contrary, they become more machine-like than machines, selling their labor like livestock and repeating monotonous motions like machines (what Marx called alienation). It is precisely from this point onward that fear of technology and resistance to it became increasingly pronounced.
The Enlightenment thinkers put forward the proposition that “man is a machine.” Whether this proposition is reasonable or not can be set aside for the moment, but the very raising of this proposition first of all suggests the reality that “machines are like humans.” People may not necessarily have arrived at the belief that man is a machine through a careful analysis of the structure of the human body; on the contrary, they more often arrived at the conclusion that man is a machine by starting from their experience of machines and discovering the image of human beings in them. Machinery was no longer a pile of dead things; technology gradually came “alive.”
The intelligence of animals evolving into human intelligence took a long process, yet the development of technological objects from automatic machines to artificial intelligence seems extremely rapid. This is because the evolution of technology does not proceed only through blind “natural selection”; from the very beginning, it is continually infused with human intelligence.
Each technological object, while externalizing or extending some human skill, also fixes or condenses a certain body of human wisdom. For example, a simple hammer: its construction not only magnifies human force, but also contains a whole series of kinds of wisdom—how to grasp, how to control, how to exert force, and so on—as well as wisdom concerning nails and repair activities, and even wisdom concerning the properties of materials like iron and wood… People “package and encapsulate” all kinds of wisdom within each technological object.
When technology is in operation, it is not only a display of force, but also a display of wisdom. But whose force does technology display, and whose wisdom is it?
People seem more willing to accept that “force” should be credited to technology itself. For example, if a weak girl can use a shotgun to kill a strong wrestler, we will marvel at the power of the shotgun, and will not say that the girl herself is very strong. Yet when we speak of “wisdom,” we always refuse to credit it to technological objects.
A person who can at any moment consult printed books may appear more learned than an isolated old pedant; we would think that he is borrowing the wisdom of others. We still refuse to think that the printed book itself is wise, and only say that others can transmit wisdom through printed books.
The wisdom gathered and condensed in an object is often not just that of a single person. Take a piano, for instance: the craftsman knows how to make it, but need not know how to play it; the performer may not know how to tune it, and the tuner may not know how to appreciate the music. Modern complex technological objects produced by factory assembly lines are even more so: the successful operation of a car depends on the coordinated contribution of almost every link in the entire modern industrial system.
Therefore, what is condensed and manifested in technological objects is not the wisdom of one or several specific individuals; at best, one can only say rather awkwardly that technology is the crystallization of “human” wisdom.
However, the attributes of the individual and the attributes of the whole are often not to be confused. We cannot always use the same categories to encompass them. For example, we can build a house out of bricks and tiles, but talking about the shape of bricks and tiles and talking about the shape of a house are two different things; the human body is made up of cells, but talking about the lifespan of the human body and talking about the lifespan of individual cells are two different things. So when we talk about so-called “human wisdom,” is that still the same thing as talking about the wisdom of specific individuals?
The wisdom involved in making technological objects, selecting and using technological objects, and the wisdom displayed through the operation of technological objects also seem to belong to different levels of category.
If an old pedant displays his ability to remember a great deal, then what he displays is “his” wisdom; when I use reference books or search engines to display the same erudition, is what I display “my” wisdom? It is worth noting that skill at searching is itself a kind of practical wisdom. If the technological tools I rely on were also given to the old pedant with failing eyesight to use, he might not necessarily be able to display any more wisdom. So when I use Google to display erudition, I too am indeed displaying “my” wisdom, yet this wisdom seems to belong to a different level from the wisdom displayed by the old pedant. Then, if those erudite bits of information no longer need a skilled searcher to use his own techniques to present them, if Google itself can listen to others asking it questions like an old pedant and help answer their doubts and resolve their confusion, then whose wisdom is it, exactly, that is displayed in this process?
We stubbornly insist that this is “human” wisdom, but surely the old pedant’s learned knowledge also comes from the accumulation of other people’s knowledge? It is just that, through years of study, he has gathered the wisdom of countless people in his own mind; the publisher gathers the wisdom of countless people into printed books; and Google gathers the wisdom of countless people into a database. If we insist that only when this “human wisdom” is expressed through the old pedant’s brain and mouth can it be counted as the old pedant’s wisdom, then why can’t we say that when the same thing is expressed through Google’s database and screen, it is Google’s wisdom?
Countless people accumulate all kinds of knowledge; the old pedant studies and remembers much knowledge; I know the old pedant and am good at communicating with him; when my little friend throws me a hard question, I seek help from the old pedant and thereby obtain a wise answer.—In this series of events, every link involves “wisdom” at a different level, but on the other hand, the protagonist of each link could be replaced by a machine. So why can the corresponding level of wisdom not be credited to the machine?
Technology is an extension and condensation of human capacities. What is meant here includes both physical power and intelligence. If we can credit force to technology, then we should also credit intellect to technology.
“Artificial intelligence,” in a sense, is itself a tautology. So-called “technology” is nothing but “artificial intelligence,” that is to say, “intelligence condensed into artificial objects.” This “condensation” has already changed the ownership of “wisdom.” For example, when the wisdom of countless people converges in the old pedant’s mind, what is displayed is wisdom belonging to the old pedant. When the old pedant helps us resolve difficult problems, what we first thank and praise is obviously the old pedant’s wisdom, not the old pedant’s parents and teachers, and even less do we speak of some so-called “human wisdom.” The wisdom condensed in technological objects should likewise, when it is displayed, be credited to technological objects; at least the primary credit should go to them.
Among the Greeks, wisdom stood alongside courage, justice, and temperance, and was understood as one of the “virtues” of human beings. But the term “virtue” was by no means the exclusive property of human beings. In ancient Chinese, terms such as “dao,” “nature,” and “goodness” all included things in general; the Greeks’ “virtue” originally also referred to some thing’s characteristics, quality, function, and so on—for example, running is a virtue of horses.
Running is one of the specialties of horses, but cheetahs, antelopes, long-distance runners, and many other things can also be good at running. Wisdom is one of the virtues of human beings, but does it necessarily belong exclusively to human beings?
Horses surpassed humans in speed, and trains then surpassed horses; people felt no humiliation. So why does it make people lose their “dignity” when machines surpass humans in intelligence?
This seems to be some kind of cycle of retribution. When the power of human technological invention surpassed animals and nature, people’s confidence swelled, and they took themselves to have conquered nature and stood above all things. It is precisely because arrogance and conceit are treated as so-called “dignity” that when humanity’s hegemonic position may be shaken, people feel that their dignity has been wounded.
Of course, many people may accuse me of muddling concepts, because “intelligence,” unlike general capacities such as force or speed, seems to involve the existence of “mind” or “self-consciousness,” and therefore a thing without self-consciousness cannot possess intelligence. But this is actually another question. In fact, the only being whose “self-consciousness” we can be sure of is “I” myself. As for whether other people also have minds, that itself is inferred by sympathy. We first see other people displaying “intelligent” behavior similar to mine, and only then do we think that others possess a mind similar to mine. It is not that we first need to determine whether others have minds, and only then can we speak of whether they display intelligence. Regarding the problem of other minds, only this behaviorist-leaning understanding is reasonable; mind itself cannot be seen. We merely posit “mind” as the ultimate responsible party for a series of behaviors.
The form and matter of computers differ greatly from the human brain, but these differences themselves cannot a priori deny that computers may be regarded as having minds. For example, the brain structure of a person in a vegetative state may be more or less similar to ours, but given that he can no longer manifest any conscious behavior, we can say that he no longer has self-consciousness. And if we assume that aliens visit us, then before we can crack open their skulls and thoroughly study their structure, we would probably already have had to interact with them on the premise that they possess intelligence and consciousness.
Aliens, I’m afraid, are something we are unlikely to encounter; yet artificial intelligence is already right before our eyes. The “Turing test” is no longer a holy grail out of reach; in many senses it has already been achieved. Today’s computers are already able, in many specific scenarios, to mimic human beings to the point that people cannot tell the difference. For example, pretending to be a 13-year-old boy for five minutes, or playing the role of online customer service, or acting as a chess player, and so on. Among philosophers, Searle’s “Chinese Room” thought experiment is one of the few objections worth taking seriously, but it has little to do with the issue we are discussing here—whether artificial intelligence actually has a “mind” has nothing to do with whether it is frightening.
One key issue implied by the “Chinese Room” is this: when we can grasp the process of “consciousness” externally, when that process loses its inwardness—that is, loses its mystery—we then tend to think that such consciousness is virtual. For example, when a person’s apparent “understanding of Chinese” depends on external, visible, graspable behavior—consulting a dictionary—we take him to be a fake understander. But if his “understanding of Chinese” depends on certain specific regions inside the brain, and the operating mechanism of those regions remains obscure, impossible to externalize, to remain or be replicated outside the individual, then we take him to be a real understander. In fact, ironically, it is precisely when we cannot “understand” how he “understands” that we believe he really “understands.”
And the consciousness process of artificial intelligence or computers is externalized from the very beginning; it is visible and graspable. So we can understand why it seems to understand, and therefore we think it is fake understanding.
Leaving aside for the moment whether the logic above is reasonable, the frightening thing is that people have also begun to find themselves not quite able to understand the mechanism behind artificial intelligence’s “why can it understand XX?” This is a feature of the new-generation AI model based on “neural networks”: machines are beginning to be able to learn autonomously, instead of simply copying frameworks predetermined by human beings.
For example, Google’s artificial intelligence has already “learned” to identify “cats” from tens of millions of images. This is different from face recognition and the like in the past: in the past, human programmers would write out a complete program for how recognition should be done, establish standards for what counts as a face, and let the computer “look up the answer by following the clues.” In the end, the machine itself formed a “concept” of cat, and recognized cats in ways the programmers had not known in advance. (Since a machine can spontaneously form the concept of cat, can it also spontaneously form the concept of “self”? I think this is possible, but very difficult; if I have the chance, I will discuss the special nature of the concept of “self” separately. But even without self-consciousness, “Skynet” could still dominate human beings, just as factory assembly lines already can dominate workers.)
This is also the key improvement of this AlphaGo compared with earlier artificial intelligences such as Deep Blue: now there is no longer any need for humans to preprogram how it should play chess; instead, the computer is allowed to summarize for itself, from countless game records and countless actual matches, what strategy to use. In theory, the programmers designing AlphaGo do not need to be skilled at Go at all; they could even know nothing about the basic rules of Go, and, just as the computer learns to recognize cats from countless images, let the computer discover the mysteries of Go on its own. This AlphaGo had not yet learned entirely from scratch (it is said that it will do so later), but the computer has already displayed many aspects beyond human understanding. What is frightening is not that AlphaGo won, but that people have begun not to know just what it was winning by. Many of its moves exceeded the judgment of all human masters.
If, when Deep Blue defeated Kasparov, we said that the computer’s advantage was nothing more than its speed of calculation, then now AlphaGo has been crushing its human opponents in precisely those areas in which people have traditionally taken pride. For example, so-called intuition, a sense of the whole situation, creativity, and so on. At first Lee Sedol tried to disrupt the computer’s footing with novel moves that broke fixed patterns, but in the end he found that it was the computer that was better at breaking patterns.
Human intelligence includes many abilities, such as reasoning, imagination, judgment, expression, and so on, but among them the most central and important, I think, is “learning.” If the key breakthrough of the Industrial Revolution lay in the rise of the “tool machine” (Marx pointed out that the tool machine replaced the human directly acting upon the labor object), allowing machines to replace humans in the “operation” phase, then the key breakthrough of this information technology revolution, or of artificial intelligence, lies in the rise of “deep learning,” that is, in the “learning” phase, where machines too begin to replace humans. From this point on, we can no longer regard machines as “mechanical,” or as merely repeating in a rigid way the patterns humans have preset in advance; on the contrary, machines will get ahead of humans in breaking with convention.
From now on, the Turing test will become increasingly irrelevant, because the Turing test is still only about how to make machines imitate human beings. But in fact machines can be far more sophisticated than humans, and need not form intelligence by imitating humans at all; instead, they can bypass humans, need not take humans as models, and can learn directly from things, developing intelligence in their own way. Perhaps one day, letting machines take part in the Turing test and imitate humans will be as dull as letting human beings imitate a dog.
In the matter of Go, this may already be the case. A few years ago, it was hard for us to tell whether a certain player was an amateur human or an artificial intelligence. But after this year, we can tell again: the machine’s sophistication, rather than its stupidity, has instead become the difference between machine and human.
I’ll stop here for now; this discussion has become a bit muddled, and I’ll sort it out slowly later. Artificial intelligence is something my writing on the history of technology must address, because it is the “end point” of the whole history of technology, and it also embodies the deepest unity and paradox between human beings and technology, while fitting my own key lines of thought—“externalization,” “learning,” and so on—even more closely.
Translated from the Chinese original with AI assistance. The original text is authoritative.
Leave a Reply