This essay was prepared for this year’s conference on phenomenological philosophy of technology. It had been brewing for several months, but in the end I still failed to turn it into a complete paper, and handed in some essay-like prose instead. There were too many people attending this time, and Teacher Wu was not planning to have every participant present a paper; some papers would be included in the conference proceedings but not assigned a presentation, so I was even more willing to make do.
The main reason I could not write the paper, of course, was that I simply did not have enough time; I had read the literature along the wrong track, and had ground through material on “intentionality,” only to discover in the end that “intentionality” was a very deep pool, beyond my ability to master.
I was by no means lazy about this topic. On the contrary, it is so important that I did not want to cite a bunch of other people’s work indiscriminately, because that would very easily let myself be swept off course by their lines of thought. In particular, the “post-phenomenologists” since Ihde have already supplied too many useful concepts, and yet precisely because these concepts are so useful, they need to be handled with caution. Take “technical intentionality,” for example: I very much want to use this concept, but it implicates too much, and leaves me full of misgivings. (Soon I will write an article briefly discussing my thoughts on “technical intentionality”; whether I will actually use this concept is something I still have not decided.)
Though this rambling essay is rather perfunctory, it should still have buried within it many of the key points around which I will later develop my argument. Once I grasped the core concept of “learning,” I gradually felt that I could unify the four strands of history of science, philosophy of science, history of technology, and philosophy of technology…
1. Starting from phenomenology to discuss the problem of artificial intelligence
When this conference first confirmed registrations, it was probably around the time AlphaGo defeated the Go champion Lee Sedol, and I immediately thought I should follow the trend and write a paper on artificial intelligence. But after mulling it over for several months, this paper still proved hard to squeeze out. The key problem was that although artificial intelligence has always been a hot topic in the philosophy of technology, its “discursive authority” has basically belonged to analytic philosophy, which has essentially always situated the philosophical problems of artificial intelligence within the domain of philosophy of cognition or philosophy of mind.
Of course, there have also been phenomenologists involved in related discussions. In the United States there is Dreyfus’s *What Computers Can’t Do*, and in China Xu Xianjun and others have also done some related work. But the main contribution of these works has still been to introduce phenomenological ideas and concepts into the context of analytic philosophy; the overall framework for discussing the issue has basically remained within the horizon of philosophy of cognition and problems such as mind-body dualism. In addition, because the hugely influential John Searle appropriated the key phenomenological term “intentionality” into the context of philosophy of mind and made it the core concept in his discussion of artificial intelligence, he has created still more confusion for phenomenologists who try to participate in the relevant debate; if one is not careful, one falls into some kind of trap.
For example, if you go and argue over whether matter is primary, then no matter whether you answer yes or no, you have already fallen into the “trap” of foundationalist metaphysics. And if we argue over whether computers can or cannot do this or that, I am afraid we are also walking into a dead end.
I am also dissatisfied with this situation. A phenomenological philosophy of technology ought to keep its attention on the history and frontier of technological development. Yet I cannot accept Dreyfus’s route, and I am even less willing to take Searle’s route. Instead, I hope to start anew and discuss the problem of artificial intelligence from a different perspective. Put simply, I want to bring the problem of artificial intelligence into the domains of ontology and history. This is not to say that the various discussions in philosophy of cognition or philosophy of mind are wrong; rather, it is only to say that we should not be confined to these perspectives. In these fields analytic philosophers do indeed have their distinctive strengths, but phenomenologists have their own strengths as well, and there is no need, nor should there be any need, to collide head-on with analytic philosophy.
2. The Turing Test is a direction, not a verdict
If we are to reposition and sort out the problem of artificial intelligence, it may be best to begin with the definition of basic concepts. First of all, the concept of “artificial intelligence” (Artificial Intelligence) was from the very beginning clearly discussing a kind of “ability,” not a kind of “entity.” In other words, artificial intelligence is not “artificial mind.”
Although the phrase “artificial intelligence” was formally proposed in 1956, the “Turing Test” designed in 1950 by the father of the computer, Alan Turing, undoubtedly set the tone for the problem of artificial intelligence. Turing first posed the question “Can machines think?” and pointed out that since words like “think” are semantically ambiguous and would inevitably drag us into wrangling over terms, he proposed replacing this question with an operational “imitation game,” substituting for it the question of “whether machines can win this game” — that is, under a certain controlled environment, can they imitate human behavior without being exposed.
The concept of artificial intelligence and the related discipline that subsequently took shape were also deeply marked by this operationalist character. Artificial intelligence technology has always focused on how to enable machines to imitate and realize human abilities; the problem of artificial intelligence has always focused on external behavior and performance.
The test game Turing designed is very simple: let the machine and a human separately converse with the judge. During the conversation, factors such as accent and facial expression are eliminated (that is, communication takes place only through text). Then, if a considerable number of judges cannot tell which one is the real person and which one is the machine, the machine can be said to have passed the test.
What does it mean when a machine passes the Turing Test? Does it mean the machine can think? Does it mean the machine has a mind? Turing did not say. In fact, it was precisely by means of this test that he replaced the question of whether machines can think. How these questions are to be answered involves semantic analysis, and Turing’s strategy was: I am merely putting forward this operational test as the practical goal of technological development, so that engineers will naturally have a direction in which to work. As for what it ultimately means, let the rhetoricians and philosophers worry about that.
So the Turing Test had already succeeded the moment it was proposed. Although it takes the form of a decisive experiment, in essence it points out a path of gradual development. Turing himself was relatively arbitrary about certain details of the test: for instance, how many subjects must be unable to distinguish in order to count as passing? Turing casually gave a requirement of 30%. As for how long the exchange must continue without being exposed, there was even less of a fixed standard. If after only one question and answer the subject is asked to distinguish, then perhaps computers from the 1960s could already pass the test; but if the time limit is unlimited and one may converse for a lifetime before making the distinction, then it obviously loses all operability.
In fact, so long as a machine can fool one or two people within a few sentences, the outcome of the Turing Test is already determined. Because from 1% to 30% to 50%, from two sentences to ten sentences to ten thousand exchanges, these are all merely differences of degree; just as when AlphaGo defeated the European champion, whether it could defeat the world champion was only a matter of time. Even if technological development often encounters bottlenecks, these bottlenecks are mostly technical rather than principled.
In 2014, a chatbot finally fooled 33% of the judges at a Turing Test conference and won, but this event itself was already of little consequence. The development and application of artificial intelligence technology had long since far exceeded the scope of imitating humans in conversation.
3. Intelligence is the ability to learn
What is worth discussing and criticizing about the Turing Test is not its result, but its presupposition. The key point of the Turing Test, as Turing himself said, is that it is “for the purpose of drawing a fairly sharp line between the physical and intellectual capacities of a man.” By isolating the body, Turing treats the ability to communicate using linguistic symbols alone as the standard of “intelligence.”
Even among many other commentators, the ability tested in Turing’s imitation game is further regarded as a uniquely human ability. This comes from the most classical philosophical tradition: first regard humans as “rational animals,” then define “reason” as “the ability to use concepts,” and finally define “concepts” as the use of specific written symbols. In this way, the Turing Test is testing the fundamental ability by which humans are human, and it is not hard to understand why people would feel that a machine’s victory in this test would be some kind of blow to human “dignity.” Once phenomenology breaks through this essentialist and rationalist anthropological presupposition, there is naturally no need to panic so much about artificial intelligence.
In fact, humans have all sorts of “abilities,” and in most cases it is hard to say that a person’s various abilities are purely intellectual or purely physical. For example, riding a bicycle looks like a bodily ability, but it also requires a certain mind to learn; people with intellectual disabilities, like people with physical disabilities, are all hard-pressed to learn how to ride a bicycle. Hunting, gathering, farming, forging, and so on are all abilities at which human beings excel, but they are hard to subsume simply under the two sharply distinct concepts of intelligence and physical strength.
These various abilities of human beings are not independent of one another; they overlap one another, and some abilities are the basis of others. For example, the ability to walk upright is the prerequisite for more specific abilities such as running, jumping, dancing, and kicking a ball. “Intelligence” is undoubtedly also the prerequisite for many abilities, and perhaps one could even say that it is the basic ability that makes all human abilities possible. But what exactly is this ability?
In everyday discourse, the context in which the word “intelligence” appears most often may be “smartphones.” And from the usage of the phrase “smartphone” we can find some clues for understanding what “intelligence” is.
The opposite of “smartphone” is “feature phone,” so what exactly does a smartphone have more of than a feature phone, such that it can be called “smart”? Is it one more “function”? Not at all. When the iPhone first came out, the built-in functions it included were not necessarily more numerous than those of Nokia phones from the same period. The reason a smartphone is smart is precisely not that it provides more complete functions, but that its functions are “incomplete”; that is, it is “innately insufficient” and therefore leaves the user room for continually installing new functions. What makes a smartphone superior to a feature phone is not its perfection but its “defect,” its expandability.
If human beings surpass animals by virtue of their excellence in “intelligence,” then this advantage also does not lie in specific functions such as sharp teeth and claws, but precisely in their “expandability.” As Stiegler says, humans are able to supplement themselves because they are deficient (and technology is human supplementation). And as Heidegger also said, the distinctive feature of human beings (Dasein) is that “it is in the being of this being that its being is an issue for it.” The special strength of human beings, like that of smartphones, lies precisely in their non-presentness, their openness toward the future.
The “smartness” of the smartphone is embodied in its expandability, and the “smartness” of human beings is nothing other than their ability to learn.
4. Can artifacts have “intelligence”?
In any case, “intelligence” is indeed a special ability of human beings relative to other animals. Although some animals also use “tools” in a flexible, opportunistic way, when it comes to the ability to learn and expand, human beings are clearly in a class by themselves.
But can artifacts possess “intelligence” comparable to that of human beings? Or rather, in what sense is it legitimate to talk about the “intelligence” of artifacts?
From the actual usage of the phrase “smartphone,” we can indeed talk meaningfully about the “intelligence” of artifacts. But the question is whether this example is merely an accidental metaphor, or whether the intelligence of the smartphone and human intelligence were originally bound together by some sort of consistent inner connection?
First, we should avoid begging the question. If we predefine that only humans can talk about intelligence, then of course we can immediately conclude that “artifacts are not humans, so artifacts have no intelligence.” In that case, before talking about intelligence, one must first judge whether the object being discussed is a “human being”; but the question of what a human being is often already contains the definition that humans are rational beings.
As artifacts, computers are very different from human beings in their material substrate, but matter alone clearly is not enough to determine whether something has intelligence. For example, a brain-dead vegetative patient is not much different from a healthy person in terms of the material composition of the body, but one can hardly say that such a person still has intelligence. And an alien being (we can reasonably imagine) may have a bodily structure entirely different from ours, but before we ever crack open its skull to inspect the material composition inside, we would probably already be able to tell whether it is an intelligent being.
If “intelligence” must necessarily attach itself to some “mind,” and artifacts have no “mind,” then that would indeed exclude artifacts from “intelligence.” But what is “mind” anyway? No one has ever first identified whether others have a mind and only then discussed whether they display intelligence. The traditional dualism of mind and matter posits the existence of an insulated, independent, unconditional “mind,” but this questionable and ungrounded presupposition is precisely what phenomenology seeks to overturn.
Once we give up all preconceived definitions regarding human essence, speaking of artifacts as possessing “intelligence” comparable to human intelligence ceases to be such an unacceptable thing.
5. “Artificial” and “intelligence” share the same origin
If talking about the intelligence of artifacts is legitimate, then the next question is: from when exactly can artifacts be said to have intelligence?
As I said earlier, the Turing Test is not a verdict but a direction of development. If a machine could fool everyone in every time and place, then leaving aside the question of whether others possess minds, at least in practical terms this machine could of course be regarded as intelligent. But if the machine only fools some people in certain times and certain places, then since imitation of humans by machines is itself a matter of degree, from what degree of imitation onward can a machine be said to have intelligence?
In my view, artifacts did not begin to have intelligence in 2014, nor did they begin in 1950; they began in prehistoric times. Technology and intelligence have always been two sides of the same coin. Intelligence calls for self-expansion and extension, and technical artifacts are precisely this expansion and extension itself. Technology is the extension and solidification of “intelligence,” through artifacts human beings realize themselves and complete themselves. So “artificial-intelligence” from the very beginning shares the same root and source, and mutually constitutes itself.
I said earlier that intelligence is nothing other than “the ability to learn,” and technology might rather be “something that can be learned.”
How is learning possible? Plato already raised the paradox of learning in the *Meno* — learning is turning what one does not know into what one does know, but if I did not originally know it, then when I obtain it, how do I know that it is precisely the thing I was looking for? Plato’s explanation is that I can only learn what I already knew, except that I had forgotten it; thus learning is recollection.
But how is forgetting possible? Stiegler gives his explanation: forgetting means leaving memory to linger outside us, and that thing outside us that extends and bears our memory is technology.
Every artifact is the extension and fixation of some ability. For example, in every hammer there is sedimented the knowledge of human beings over thousands of years regarding hammering. Every person who can hammer with their fist may learn to use an iron hammer. The process of learning is the internalization of these external things into one’s own behavior or habits, while also externalizing one’s own body — for example, I may adjust the weight and length of the hammer according to my bodily habits, or install software and upgrade hardware on the computer according to my preferences. At any moment I may also feed back my experience of use to other people, including the maker of the artifact and the teacher who taught me how to use it. After receiving the feedback, they in turn may transmit new things to me or to the next learner… The whole process of learning to use technology is a two-way process of interaction and adjustment, moving from inside to outside and from outside to inside, and the shape and material of the hammer at my side are themselves one of the products that have sedimented after countless generations of use and learning.
While every technological object externalizes or extends some bodily skill, it also solidifies, or rather condenses, some set of human wisdom. For example, a simple hammer not only amplifies human strength, but also embodies a series of kinds of wisdom about how to grasp it, how to control it, how to exert force, and so on, along with wisdom concerning nails and repair activities, and even wisdom about the properties of materials such as iron and wood… People “package and seal up” wisdom from every conceivable aspect within each technological artifact.
6. Whose is the ability strengthened by technology?
Precisely because human beings can extend “intelligence” to what is “outside” themselves, these abilities can be transmitted, accumulated, and improved. But do these abilities that linger in “external things” still belong to human beings, or can they be said to belong to those artifacts?
Human beings have the ability to run, but horses run faster than humans, and human-made trains run faster than horses. When trains won out, ordinary people did not feel humiliated, nor did they hastily parse the different meanings of “running” so as to deny that a train’s racing along and a human’s running could be compared. Rather, they regarded the superiority of artifacts in certain respects as an expression of the great power by which human beings conquer nature. Technology was originally an extension of human beings, a completion of deficiency, an expansion of originally feeble human functions; so the various abilities displayed by human-made objects being stronger than those of a naked individual was, in the first place, only as it should be.
The problem is that the powerful abilities displayed in the operation of other artifacts are still regarded as belonging to human beings, not to the artifacts, because these artifacts are ultimately human works, the “crystallization of human wisdom.” Yet this wisdom has already “crystallized” in the artifact; why can we not say that the artifact possesses wisdom?
Regarding the question of whose wisdom exactly is “crystallized” in artifacts, people have always had a somewhat contradictory attitude toward it — already in Plato’s discussion of how “writing damages memory,” this problem had been laid before philosophers. With the aid of writing technology, human memory seems to have been strengthened, yet this strengthened capacity seems to belong to paper and pen rather than to the person. And because human beings rely on such external things, once separated from paper and pen, their memory capacity is instead weakened.
If a scribe who relies on paper and pen can display stronger memory, and if this additional capacity is not credited to the scribe “himself,” then naturally these abilities should be credited to writing technology.
VII. Artificial Things as the “Crystallization” of Human Intelligence
Yet people are often unwilling to credit capacities of intelligence such as memory to artifacts. Since this extra capacity neither belongs to the user of the technology nor to the technology itself, it can only be attributed to the creator of the technology.
What needs to be posthumously acknowledged is obviously not limited to the direct manufacturer of a technological object. What gathers and condenses in a single artifact is often not just the wisdom of one person. For 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. And this is even more true of the various complex modern technological objects produced on factory assembly lines: the successful operation of a car depends on the collaborative contributions of almost every link in the entire modern industrial system.
Thus, what is condensed and manifested in technological objects is not the wisdom of any one or several specific people, but can only be said, with some strain, to be the crystallization of “human” wisdom.
However, the attributes of the individual and those of the whole cannot often be mixed together; we may not always be able to subsume them under the same category. For example, we can build a house out of bricks and tiles, but talking about the shape of bricks and tiles is one thing, and talking about the shape of a house is another. The human body is composed 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 really the same thing as talking about the wisdom of a particular person?
The wisdom involved in making technological objects, the wisdom involved in selecting and using technological objects, and the wisdom displayed through the operation of technological objects also seem to be categories on different levels.
If an old pedant displays his encyclopedic memory, what is displayed is “his” wisdom; but when I display the same breadth of learning with the help of reference books or a search engine, is what is displayed “my” wisdom? We should note that being good at searching is itself a kind of practical wisdom. If the technological tools on which I rely were also given to the near-sighted old pedant, he might not necessarily be able to display more wisdom. So when I display erudition with the help of Google, I do indeed display “my” wisdom, yet this wisdom seems to belong to a different level from the wisdom displayed by the old pedant. Then if these erudite pieces of information no longer need someone skilled at searching to present them with his own technique, and Google itself can listen to other people’s requests for instruction like an old pedant and help solve their problems, then whose wisdom is it exactly that is displayed in this process?
We insist on saying that this is “human” wisdom, but aren’t the old pedant’s erudite knowledge also all derived from the accumulation of other people? Only through years of study has he gathered the wisdom of countless people into his own mind; publishers gather the wisdom of countless people into printed books, while Google gathers the wisdom of countless people into databases. If we insist that only when these “human wisdoms” are expressed through the old pedant’s brain and mouth can they be attributed to the old pedant’s wisdom, then why, when the same things are expressed through Google’s database and screen, can we not say that they are Google’s wisdom?
Countless people have accumulated all kinds of knowledge; the old pedant has studied and memorized a great deal of knowledge; I know the old pedant and am good at communicating with him; when my little friend poses me a difficult problem, I ask the old pedant for help and thus obtain a wise answer. — In this series of events, every link involves “wisdom” on a different level. But on the other hand, the protagonist at each link could be replaced by a machine. So why can’t we attribute the corresponding level of wisdom to the machine?
Technology crystallizes human wisdom into artifacts. This process of gathering and solidifying has already changed the attribution of “wisdom.” For example, when the wisdom of countless people is gathered into the old pedant’s mind, what is manifested is wisdom belonging to the old pedant; when the old pedant helps us solve difficult problems, the first thing we are grateful for and praise is obviously the old pedant’s wisdom, not his parents’ or teachers’—much less do we speak of so-called “human wisdom.” The wisdom condensed in technological artifacts, when manifested, should likewise be credited to the technological artifact.
VIII. The Gradual Independence of Machines
So, from the very beginning, artificial things possess “intelligence” independent of any specific person. But before the rise of computer technology, and especially in ancient technology, “artificial intelligence” had never become a problem. This is because although ancient technology already had a certain degree of independence, the exercise of its “capabilities” always remained inseparable from the human being as operator. The memory capacity displayed by paper and pen was not wholly credited to the writer, but there was always a specific hand at work. Thus its independence was not especially conspicuous.
Things like writing and weapons—these ancient technologies already had a certain degree of independence, that is to say, in addition to being controlled by human beings, they also had a certain inertia that was “not transferable by human will.” Hence Qin Shi Huang had books burned and Confucian scholars buried alive, and had weapons from all under heaven collected and cast into bronze figures. For as long as these books and weapons circulated in the world, they possessed a certain power. Although these powers ultimately had to be exercised through the will of specific operators, in a certain sense those operators’ wills were themselves brought about by these technological objects.
Of course, the “independence” of these ancient technologies was not especially conspicuous, because after all they still needed people to control them; otherwise they were just a pile of dead matter. Of course, the power of technological objects does not necessarily need to be exerted only when they are in motion. Things like city walls, buildings, tombstones, and so on, simply standing there unmoving, already guide and even govern human behavior and thought; nuclear bombs lying in a cellar are even more capable of influencing the global order. But after all, in the eyes of ordinary people, these artifacts always remain “passive” in relation to human beings.
When did technological objects begin to acquire a certain “initiative”? The most emblematic answer is the development of mechanical technology. From earlier waterwheels, windmills, and the like, to the mechanical clock in the late Middle Ages, and then to steam engines, textile machines, and factory assembly lines. The new feature of these machines is this: aside from a few relatively inconspicuous steps such as winding a spring or adding raw materials, machines in the course of normal operation are relatively independent; they detach 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, what human beings play is not the role of operator, but the role of power source. In the large-scale, mechanically linked operation of factory assembly lines and the like, the role human beings play is not that of a wise creator; on the contrary, they become even more machine-like than machines, selling their labor like livestock, repeating monotonous tasks like mechanisms. It was precisely from this time onward that fear of and resistance to technology also became increasingly pronounced.
Enlightenment thinkers proposed the claim that “man is a machine.” Whether this claim is reasonable or not can be set aside for the moment, but the very proposal of it first and foremost implies this reality: “machines are like man.” People did not necessarily arrive at the belief that man is a machine through a detailed analysis of the structure of the human body; on the contrary, people more often discovered the image of man in machines based on their experience of machines, and only then reached the conclusion that man is a machine. Machinery was no longer just a pile of dead matter; technology gradually came “alive.”
IX. Machines Learn to Learn
The recent rise of artificial intelligence technologies represented by “deep learning” has made machines seem to have completely broken free of human control.
A key issue hinted at in Searle’s “Chinese Room” thought experiment is this: when we can grasp the process of “consciousness” externally, when such a process loses its interiority—that is, loses its mystery—we tend to think that this consciousness is fake. For example, when a person appears to “understand Chinese” by relying on an external, visible, graspable behavior—looking up words in a dictionary—we think he only pretends to understand. But if he “understands Chinese” by relying on certain specific regions inside the brain, and if the mechanisms of those regions’ operation are still obscure, cannot be externalized, and cannot linger or be replicated outside the individual, then we take him to understand genuinely. In fact, ironically, it is precisely when we “cannot understand” how he “understands” that we believe he truly “understands.”
By contrast, the process by which a machine exerts its capabilities is from the outset externalized; it is visible and graspable. So we think it is nothing more than actually running through a set of routines already preset by others, still with no initiative or independence to speak of.
Yet with the development of new artificial intelligence technologies such as “neural networks” and “deep learning,” machines have begun to become capable of autonomous learning, rather than merely copying frameworks pre-established by human beings.
For instance, Google’s artificial intelligence has already “learned” to recognize “cats” from tens of millions of images. This differs from earlier face recognition and the like. In the past, human programmers would write a complete program for how to recognize things, formulate the standards for what counts as a face, and let the computer “follow the map to find the horse.” In the end, the machine itself formed the “concept” of a cat, and used a method of identifying cats that programmers had not known in advance.
This is also the key improvement of AlphaGo compared with earlier artificial intelligences like Deep Blue. Now there is no longer any need for human beings to preprogram how one should play the game; instead, the computer is allowed to independently summarize, from countless game records and countless actual matches, the strategy of how to play. Theoretically speaking, the programmers who designed AlphaGo do not need to be good at Go at all; they could even know none of the basic rules of Go, and let the computer discover the mysteries of Go on its own, just as one would let a computer learn to recognize cats from countless images. AlphaGo this time has not yet learned entirely from zero (it is said that it will do so later), but the computer has already displayed many aspects beyond human understanding. The frightening part is not that AlphaGo won; it is that people have begun not to know on what basis it won. 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 speed of calculation, then now AlphaGo is crushing human opponents in those areas at which people have traditionally prided themselves. For example, so-called intuition, overall perspective, creativity, and so on. At first, Lee Sedol tried to disrupt the computer’s footing with novel moves that broke established conventions, but in the end it turned out that it was the computer that was better at breaking conventions.
X. Symmetrical and Asymmetrical Relations Between Human Beings and Technology
It is not that only after the rise of technologies like “deep learning” did machines begin to be able to “learn.” On the contrary, the very possibility of deep learning technology lies in the fact that artificial things were already able to “learn” in the first place. Artificial things have, from the beginning, been the condensation of “intelligence,” and “intelligence” is nothing other than the capacity to learn, or rather, “blankness” and “deficiency.” From the start, artificial things and human beings have been a mutually constitutive relationship, each the outer and inner aspect of the other.
Can artificial things “learn”? It seems that artificial things lacking autonomous consciousness cannot learn “on their own.” But if we allow them to “with the aid of” human beings, then of course they are also learnable; artificial things too seek, in their own ways, to be “completed.” But on the other hand, can human beings really learn “on their own”? The reason human beings can learn is also that they must “rely on” external technological objects and technological activities. “Learning” was never something happening purely inside the soul. Human self-perfection and the self-perfection of artifacts have always been mutually constitutive.
Since wisdom means deficiency, does the development of wisdom mean the strengthening of deficiency? In a certain sense, yes; that is precisely why it is said that “writing damages memory.” As people continually extend their intelligence into technological objects, they become increasingly dependent on these technological environments; people dwell within technology. From the outset, human beings’ relation to technology is not merely one of control, but always also one of entrusting themselves to technology. While breaking through the limitations of the body through technology, they also acquire new deficiencies and are constrained by technology.
In this respect, however, the relation between human beings and technology seems no longer so “symmetrical.” The development of technology makes people increasingly dependent on technological environments, but conversely, technological environments seem increasingly less dependent on human participation.
Once strong artificial intelligence comprehensively surpasses human beings in the capacity to learn, thereby becoming able to dispense with human beings’ own driving force for technological innovation, the balance between human beings and technology may ultimately be broken: human beings will be unable to survive without machines, but machines, having parted from human beings, will still be able to run on their own.
XI. The Fate of the History of Technology
This asymmetrical relationship is historical. In prehistoric times, human life depended less on artifacts; as civilization evolved, human life depended on more and more artifacts, and the connections among artifacts became increasingly tight. On the other hand, the more ancient the age, the more human beings depended on their individual bodies, and the more tightly connected the bodily faculties were.
Insofar as technology is an extension and limitation of human beings, ancient technology and modern technology are the same. The difference is that in ancient times the human body was often a whole, with the various senses not yet split apart from one another, while the technological environment had not yet closed itself into a whole; each technology extended a particular capacity and pointed to a particular purpose. All kinds of artificial things only became a whole in human life in the end. In modern times, by contrast, the technological environment has closed itself into a “Gestell,” with the entire technological environment, represented by factory assembly lines, connected end to end and forming an integrated system, while the human body has instead been split apart. Vision was first peeled away under print culture; in factory assembly lines, workers’ hands were peeled away; the labor of the worker’s hands became unrelated to his emotions and desires, and was governed by the rhythm of the machine.
With the advance of industrial civilization, the “dismemberment” of the body and the “integration” of artificial things have become increasingly conspicuous. “Artificial intelligence” in the narrow sense, that is, the “computer,” was born precisely against this backdrop. The history of the computer illustrates the dissolution of the body and the independence of artificial things: nearly the whole human body has been sealed off outside the screen, and the computer seems to want only to ingest some bodiless soul into itself.
This “arrow of time” in the history of technology arises from human finitude. Compared with the vast external environment and the long history of technology, each person’s body and lifespan are exceedingly limited. People of course can, through learning, continuously internalize into themselves the wisdom that has settled beyond their own finite lives, while at the same time leaving behind new sediments outside themselves. The wisdom contributed by people will condense into various artificial things and accumulate day by day, allowing future generations to have a thicker and richer “environment.” However, future generations cannot possess correspondingly longer lifespans in which to learn. When a person dies, all the learning of a lifetime, except for those things fed back into the technological environment, will vanish like smoke; but a technology almost never dies, and is instead constantly inherited and improved.
The evolution of the human body always falls behind the progress of technology; that is to say, the increase in human beings’ capacity to learn always falls behind the accumulation of things that can be learned. Then the technological environment becomes increasingly detached from human control and, in turn, comes to govern human life. This is also the fate that human beings cannot escape.
The realization of “artificial intelligence” is neither a completed tense nor a future tense, but a progressive tense; the realization of artificial intelligence is technology’s own self-realization itself. The entire history of technology can be understood as the actualization of “artificial intelligence.”
This inescapable destiny may not necessarily be the doom of humanity. Just as human beings now must rely on the Earth to survive, but without human beings the Earth still turns as usual, this asymmetrical relationship between human beings and the Earth obviously does not imply the extinction of humanity.
Translated from the Chinese original with AI assistance. The original text is authoritative.
Leave a Reply