Literary Writing in the Age of AI Monarchs

The original text was published on Dong Fu Kee website

It is not an exaggeration to say that ChatGPT dominates the world. Almost all fields involving the use of language, including media, business, culture, education, academia, and even computer programming, were immediately conquered by it, or stubbornly resisted. It is far from the ultimate perfect language model, and there are still various shortcomings and deficiencies, so that people of insight can easily sneer at it, but its appearance will change the way of language use in future generations, and in turn affect human language ability, and Criteria for the evaluation of language works. As a machine that imitates and produces natural language, it will make language no longer natural and become completely artificial. The line between human and machine writing will become blurred and even cease to matter. It completely subverts the human cognition of language.

As a literary creator who uses language as a medium (or ontology), he naturally bears the brunt and is directly challenged by ChatGPT. Some people rolled up their sleeves and actively faced the battle; some were worried and at a loss; others were confident and dismissive. No matter what attitude we adopt, literature that is already a minority will eventually become passive and fall into a state of being beaten or retreating. Unless AI language generation is just a flash in the pan and disappears soon (extremely unlikely), the territory it occupies will only get wider and wider. The use of language in the future is likely to become the world of machine language. Literature may not be replaced, or there is no need to replace it at all, because it is something with very low demand, but literature is likely to be washed to the barren cliffs at the edge of the world by the huge waves of AI, and disappear from human vision . This is not alarmist talk.

🦊Favorite this article Writing NFT 🦔

Linguist Noam Chomsky's criticism of ChatGPT is too harsh, but it is not aimless. To blame the language generation program for having no thinking ability and moral judgment, and to scold it as a "banal evil" is to regard it as a living entity with real intelligence. Only intelligent life forms can be "mediocre" and "evil". Don't say machines, even animals cannot be matched with such predicates. In fact, ChatGPT is just a tool, and it is people themselves who mistakenly regard tools as truth providers, whether they are developers, promoters, users, or critics. However, tools are not neutral either. The design and application logic of tools imply value orientation, or no orientation. A value without orientation is also a value, and this is the so-called "evil of mediocrity". The AI language machine is a Pandora's box that has just been opened, and it is still unknown what will fly out. Discussing its possible benefits and disadvantages early on can help turn what could lead to evil into hope for the future of mankind.

What Hang Shiji said is that the operation of a large language model (Large Language Model) like ChatGPT runs counter to the universal grammar (universal grammar) preset by the human mind. Hang Shiji's linguistic theory is in line with Kant's transcendental philosophy, and it is exactly the same as Jung's deep psychology. All three can be classified as deep structuralism, the idea that the human mind has innate forms that can be reduced to specific structures or laws. According to these a priori forms, we recognize and understand the world in a posteriori or empirical domain, and use language to construct meaning. In contrast, there is no pre-set cognitive structure behind the LLM, and there is no need to follow the grammar to make sentences. It only needs to "read" a large amount of data according to the self-attention mechanism, and then match the order of words according to the probability of occurrence.

This is what Italo Calvino said in his article "Cybernetics and Ghosts". He first explained from the aspect of innate structure that human language is a word combination game played according to limited rules, and it is an activity of "putting one word after another". Citing the theory of cybernetics, he believes that such combined activities can be performed by machines, and even boldly declares that "the writer itself is a well-functioning literary machine." However, he changed his style in the second half of the article, and pointed out from the perspective of mythology that the unconsciousness suppressed by taboos can only be released in specific lives through experienced individuals living under a specific historical time and space. To say the unsaid at the junction of inner form and outer experience is literary creation.

The way AI generates language completely overturns the assumptions of linguistics (and even philosophy and psychology), and Kang Shiji believes that this is the reason for the poor quality of its output products. Because there is no prior cognitive program, but just pieced together according to chance, the things written by ChatGPT are mediocre, empty, low-intelligence, and ambiguous. But the reason why it behaves like this is not only because the designer imposed restrictions on its value orientation, but also because what it reads is mainly public information on the Internet. This kind of information represents the most average and average aspect of human language, in other words, it lacks personality. But it is not necessarily impossible to write high-quality text with personality. It is technically possible to connect ChatGPT to a specific database, such as all the writings, teachings and speeches of Hang Shiji himself, and then it can also generate text with Hang Shiji's sharp views and sharp language style. Of course, this is not the result of real learning and thinking, but a high degree of imitation.

Natural language and natural language generators (that is, human language and GPT language) are diametrically opposed in terms of production principles and operating processes. Natural language is produced from the source, and it is created out of nothing, so it is creation. The earliest use of language in history was like this, and the language of later generations continued to evolve and update, and did not stop creating. Although as far as individuals are concerned, language must be used according to rules and through learning, and most people just follow the existing usage most of the time, the possibility of innovation (new combinations, breaking rules, saying things that have never been uttered) did not disappear. Literature is born out of this possibility. On the contrary, the language generator is produced from the result (database of human language usage), and it is something out of something, so it is reinvention. Because of the combination approach based on common probabilities, performance tends to tend toward uniformity, normality, or even mediocrity.

The question is whether we are satisfied with this level of writing, and human evaluation criteria are the greatest threat to literature. For literary people, what AI generates cannot be called literature, because it is only imitation without creativity. But for ordinary people, it is likely to feel that AI has already written well enough, and there is no need for deeper, more tortuous, and more subtle human writing. If the requirements are lowered, AI can definitely replace human writers. The general public does not have sufficient literary appreciation ability, and under the feeding of AI language, judgment will deteriorate even more. As a result, "genuine" literature is ostracized by "false" literature, and art is replaced by technology. But that's not the technology's fault. In ancient Greece, the concept of art originated from the word "techné", which means technology. Without technology, there is no art; art is artificial production relative to nature. So, if AI can improve on language technology, why can't it be included in the field of art?

Let us temporarily put aside the debate about who is smarter and who is more stupid, and compare the difference in technical effect between the two purely from the point of view of the Turing test. Many people try to identify what AI cannot do in literary creation, that is, to confirm that literature is irreplaceable through the method of elimination. We are compelled to redefine literature in order to resist the invasion of AI into the field of literature. Under the strong imitative power of GPT, style is the hardest hit area of literature. Style is the unique stamp of every writer. "Style is man" (Buffon's words) is an unbreakable literary creed that we have always believed. But if the style can be easily imitated by AI, or even faked, the value of the unique personality and life experience behind it will be in jeopardy. Not only style can be imitated, but even theme, structure, thought and emotional character can be imitated. It is not unimaginable to replace "War and Peace" with the current Russo-Ukraine War, and let AI Tolstoy write a new work, with a writing style that sweeps thousands of troops, and self-criticize from the perspective of the Russians. But who actually wrote this work? Without a flesh-and-blood Tolstoy, who wrote "New War and Peace" while suffering from pain and guilt, what is the point of this work, even if it is well written?

🦊Favorite this article Writing NFT 🦔

We should also note that there is an asymmetry in this comparison. As literary creators and lovers, we have full confidence in literary judgments, but we must admit that our knowledge of AI technology is limited. If we're not careful, it's easy to underestimate its capabilities and rashly conclude that there are things it can't do. (Of course, its capabilities and threats may be exaggerated on the contrary.) There are at least three reasons for the misjudgment:

1. We novice users have never learned how to issue high-quality commands to GPT. If you know how to write prompts, the results can be very different. The reason we get mediocre results in experiments may be because we are issuing mediocre instructions. Prompt engineering has now become a science, and people who claim to be experts continue to post new teachings every day. The next generation may no longer know how to write their own articles, but they can become masters at writing prompts. The great writer at that time will be the great prompter. The focus of language learning will shift. Prompt becomes a meta-language, the language of the language, or the language of the output language. The situation is quite strange.

2. Those of us ordinary people without computer skills do not know how to write applications that hang on GPT, so we can only use a very small percentage of its potential. The issue of ChatGPT providing wrong information is most often criticized and ridiculed, but this issue is actually very low-level and can be easily resolved. As long as it is connected to a specific database, it can easily find accurate information. Travel sites and restaurant reservation sites are already actively testing applications for this. As for language quality and writing skills, as long as a large number of excellent contemporary literary works that are protected by copyright around the world and cannot be read by ChatGPT are opened, and it takes a week or two or less to absorb them, their literary knowledge and taste can be greatly improved , and produce higher quality literary language. Although such suggestions are more imaginative than practical, it is entirely possible to conduct intensive training for specific purposes or categories.

3. ChatGPT may have more and stronger capabilities that have not been disclosed or released, and it will continue to improve and strengthen. There is no reason to assume that what it cannot do today, it will never be able to do. It will likely do so quickly.

Some commentators refer to two major elements of literature—humor and irony—and argue that AI is inherently unachievable. Humor and irony are indeed more than superficial matters of style, that is, they cannot be achieved by mere imitation. Both involve duality, the awareness of subtle differences and oppositions in context, expressed in ambiguous yet obvious paradoxes. Humor is particularly dependent on cultural context and context, and requires practical experience to appreciate and use it. This hits the nail on the head of current AI — its lack of a physical, empirical presence. It is just an abstract, pure system. So we go back to the earlier critique - GPT is a freak with neither prior structure nor posterior experience. Or it can be said that its a priori (programs and algorithms) are also a posteriori (big data); both occur in a virtual, digital "space-time", that is, "non-time-space", which is different from the real world There is no direct interaction until the user calls it out.

In the final analysis, our current dissatisfaction and criticism of GPT may be due to only one reason-it is not a human being, but a machine. Even if one day it is technically perfect enough to write extremely high-quality, highly realistic works, we will not recognize what it writes as literature, just because it has no humanity, consciousness or soul. Therefore, what we cherish and stick to, or what we can say is persistent, is the entity of "author". It is meaningless to imitate Cao Xueqin and write the last forty chapters of "The Story of the Stone", because the "real" Cao Xueqin has died, and the last part of the "real" that he wrote has been lost. It is also meaningless to virtualize a new AI "writer", because no matter how well it writes technically, it is not a "real person". Half a century ago, Roland. Barthes shouted that "the author is dead", but half a century later, even postmodernism has long since died, and the concept of "author" is still alive and well. Our faith in the real person of the author is not diminished. The reason why we love literature is not only because literary works are superb, but also because such superb works are created by real people who have feelings and thoughts one after another. This is the basis on which literature can generate empathy.

Another thing that AI cannot replace is the generation of "meaning". Continuing the above point of view on "author", human beings are the source of all meanings, and AI cannot become a sufficient "author" for a day, and cannot create "meaning" for a day. GPT has no creativity, thinking skills, moral judgment, and therefore no ability to establish meaning. Human beings give their existence meaning, relying on three pillars - art, knowledge and morality. GPT cannot do it for these three aspects. (There is also a fourth pillar—faith; but in faith, the meaning comes from the transcendent God, which has nothing to do with AI. Unless the power of God is manifested through AI, it will become a veritable deus ex machina.) According to Hang Shiji’s criticism (Motivation is equivalent to Kant, Jung, etc.), the discovery of meaning is based on a priori structure, whether called categories, practical reason, archetypes, or universal grammar. This internal transcendental structure common to all human beings, through a real individual, that is, a historical and empirical person, comes into contact with the world, thereby producing the meaning of existence. The reason why human beings are different from machines is that human beings are a compound existence of prior knowledge and experience. (This is also what Calvino meant.) And AI is neither—neither a priori nor experienced, a dead thing without a metaphysical essence.

Following this line of thought, we can envision two future directions: 1. To ensure that AI does not get involved in these areas that are exclusive to humans, and is only used as a tool for specific purposes that are clearly distinguished without weakening human dignity and interests. . This means that AI cannot act as a proxy for artistic creation, intellectual inquiry, and moral judgment. I'm not sure that such a restriction is beneficial and feasible. 2. Committed to improving the AI system, so that it can finally acquire these three abilities and become a compound existence of prior and experience, that is, become a historical and empirical individual. This form of AI is indistinguishable from real people, and thus entitled to near-human rights. The former is a consensus proposed by some technological and cultural leaders to protect the rights and interests of human beings; the latter fully meets the requirements of Kant-Jung-Calvino-Honsky, but it is hard to say whether intellectuals are willing to admit the importance of AI at that time. "Personality" status. It's certainly a sci-fi scenario.

Language generation programs are sure to bring about dramatic changes in human society, and the changes have already begun. Even if literature is not immediately replaced, changes in the overall language environment will affect the survival of literature. Based on the two points of "author" and "significance", literary creators can still retain their reasons for existence for the time being. But in the near future, literature may be overwhelmed by the flood of machine language, becoming something more marginal than marginal, gradually being ignored and abandoned, or being preserved and appreciated like rare animals. The name of "literature" may be taken away by mass-produced machine literature, while the old literature of living people will become historical relics, and those who continue to create in the old way will become living fossils.

Under the challenge of GPT, literature has entered a stage full of crises, and it is no longer taken for granted and self-evident. Optimistically, this might be a good thing for literature. If we can face and respond well, it may be an opportunity to promote the emergence of literature in a new era. In the age of AI, literature cannot stay out of it, nor can it be alone; if you don't control it, you will be controlled by it. What we have to do is not to take back readers or markets from AI, but to take back language itself. And to take back language, we need to take back AI. To seize AI does not mean that everyone should use AI to write, but to teach AI what literature is, so that it can get closer to literature, understand literature, and even support and spread literature. AI language is a mirror of human language. Whether it produces high-quality or poor-quality language depends on what it learns. If you don't want it to write bad things, you have to teach it good things. However, the most important thing is to educate human beings about what literature is, because AI is also created by humans and used by humans. And to do this, we must continue to work tirelessly to write real literature. This is the responsibility of literature to the future.

🦊Favorite this article Writing NFT 🦔

Literary Writing in the Age of AI Monarch (ChatGPT Edition)

Related articles I've written before:

writing machine

the author is not dead

The point is not the conscious, but the unconscious

Is it the AI's fault, or the human's?