> topics > Literature

AI-human writing collaboration opens new possibilities

Huang Wenhu | 2020-06-03
(Chinese Social Sciences Today)

Robot poets show their potential on CCTV science program “AI VS Human” by competing with contestants who are experts on Chinese poetry. Photo: FILE


Looking at the evolution of the human-machine relationship, literary writing will be divided into the pre-AI era and the AI era, with the latter prominently characterized by the rise of machine writing. Due to the theoretical divergence between strong AI and weak AI, machine writing may lead literary creation down two different paths: human-machine collaboration and extra-human ability. With the rapid iteration and comprehensive penetration of machine writing, various aspects of literary writing are expected to change, requiring our prospective thinking and discussion about the future and its implications.


Strong, weak AI

The argument of strong and weak AI could date back to the work of Alan Turing, the father of artificial intelligence. As early as 1950, Turing posed the question of “can machines think?” in his short paper “Computing Machinery and Intelligence,” which resulted in what we now call the imitation game. His Turing machine, a prototype of the modern computer, could theoretically mimic almost any complex system, including life and natural intelligence. At most, the human brain can also be regarded as a digital computer, and a computer program can realize all the functions of the human brain, such as emotion, consciousness, spirit and other high-level cognitive activities, which compose the core idea of strong AI.

Back in the 1970s, AI scientists optimistically predicted that within a decade there could be a machine with “universal intelligence” capable of reading even complex literary works such as Shakespeare’s plays. However, Tale-Spin, a major AI attempt at story generation that emerged in the 1970s, could only produce extremely simple story patterns.

In the view of critics such as Hubert Dreyfus, it is impossible to create intelligent computer programs analogous to the human brain, because the working of human intelligence is entirely different from that of computing machines. In the 1980s, John Searle launched a remarkable discussion about the foundations of AI and cognitive science in his well-known Chinese room argument refuting strong AI. In his opinion, machines only rigidly process symbolic information based on formalized rules of logic, and it is impossible for them to show their intentionality towards the empirical world through language communication as humans do.

In other words, though machines can produce the form of literature, it is impossible for them to apprehend its content. The idea that computers can only simulate a limited number of non-advanced functions of the human brain is called weak AI.

However, after more than half a century of development, the differences between strong AI and weak AI are no longer technical, but rooted in philosophy. In recent years, AI connectionism has broken through the technical barrier of “classical AI” by optimizing neural network algorithms, unleashing the potential of machine writing in practice such as news writing.

At the same time, machine deep learning has begun to dabble in fiction. In 2016, Tsinghua University’s poetry robot “Weiwei” became the first robot poet in China to pass the Turing test, and it made its debut on the CCTV science challenge program “AI VS Human,” or “Ji Zhi Guo Ren.” Though machines have plenty of shortcomings in writing, the emergence of “digital poets” has completely refreshed the literary world’s understanding of AI.

As technology evolves, the boundaries between strong and weak AI begin to blur. In essence, there is a natural gap between the two ideas, which also hints at the two different fates of machine writing. Weak AI is essentially a kind of instrumentalism and anthropocentrism, whereas strong AI is essentially a kind of non-instrumentalism and “superhuman centrism.” The main difference is that the machine made by the former must rely on and serve human beings, while the machine created by the latter is a “human-like agent” that can be independent of human beings.


AI-human collaboration

From the perspective of weak AI, the trend of machine writing is not to replace human intelligence with machine intelligence, but to develop human-machine collaboration. At the current stage of technological advancement, human-machine collaboration takes on two major forms.

The first is human-machine co-authored literary work. In 2016, an AI-written novel titled The Day a Computer Writes a Novel made it past the first round of screening for the Hoshi Shinichi Literary Award, a national literary prize in Japan, stirring up heated discussion. However, the plot and gender of the characters of this novel are pre-set by programmers, while AI only processes and restructures the literary materials.

In 2018, Chinese science fiction writer Chen Qiufan, also known as Stanley Chen, published A State of Trance, in Chinese Chu Shen Zhuang Tai, using an AI deep learning program that has learned from his writing style. In 2019, he published The Algorithms for Life, using a similar program. In his case, the AI generated some stylish, avant-garde poetic fragments, which were not logical. Chen then embedded these sentences into his own writing and developed some stories around these automatically generated phrases.

It is worth noting that A State of Trance showed up on some novel ranking lists that were picked by an AI judge. The list included work from 20 nationally representative literary magazines in China, such as Harvest, Youth Literature and Youth Writer, covering 771 short stories published in 2018, and eventually it selected 60 “best works” based on plots, characters and other indicators.

There is no doubt that algorithms can help avoid subjective assumptions, but there is a loophole in the selection of literary works solely based on AI. Is the criterion based on algorithms fair? For example, will the AI selection system favor works out of human-AI collaboration?

In this light, a more reasonable approach might be to combine natural intelligence and machine intelligence, which means we could combine the judgment of readers and critics with the AI algorithm. In the end, such an evaluation of literary works is determined by a “merciless” algorithm but contains the subjective judgment of human beings.

As algorithms continue to advance, machine writing may someday approach or even surpass human creativity. By then, the less creative genres of literature such as entertainment-themed screenplays, popular online novels and so on are most likely to turn to machine intelligence for mass production, putting some writers out of work.

However, with the hypothesis of weak AI, machine writing will always be a creative tool for human beings. No advanced writing machine could ever replace human literature, but merely put icing on the cake.


New literary ecology

From the perspective of strong AI, the trend of machine writing would be to create a new species independent of the human literary world. This species would be neither the creative tool of human writers nor necessarily centered on the literary ideas constructed by human beings. Therefore, it would consciously or unconsciously move towards the opposite of human-machine collaboration.

In 2014, Microsoft Xiaoice published a poem collection titled Sunshine Misses Windows, which has been praised as the first poetry collection created by AI in the history of human literature. In 2016, a short screenplay called Sunspring written entirely by AI that was developed by New York University researcher Ross Goodwin and movie director Oscar Sharp, made its online debut.

In the preface of Sunlight Without Glass Windows, it wrote that Xiaoice is a combination of IQ and EQ, and that its poetry collection is a work with independent intellectual property rights. Its writing is a kind of creative behavior similar to that of human beings, and it has endless creativity. This view recognizes the legal status of AI programs and affirms its creative behaviors, which is a typical strong AI perception.

If we follow this train of thought, how we regard writing robots will eventually be upgraded from “digital existence” to “embodied existence.” One of the important reasons that critics oppose machine writing is that machines lack life experience and consciousness absorbed from the real world, and thus cannot understand the history and culture of human society, so it cannot create literary classics like human writers. However, once this “embodied nature” allows robots to travel between the real world and the virtual world, the line between man and machine will also become blurred. “Embodied existence” is what strong AI pursues ultimately: a human-level AI with rich emotions and self-awareness.

From the perspective of strong and weak AI, the evolution of machine writing is in essence the revolutionary transformation of the man-machine relationship, and it is also the epitome of the intelligent transformation of human society in the field of literature. The emergence of robot poets or writers would redefine the nature of literature and the meaning of writing.

However, the challenges and pressure brought by machine writing do not mean the withdrawal or defeat of human writers. On the contrary, the stronger AI is, the more it highlights the creativity contained in human beings as intelligent creatures.

Therefore, machine writing can neither completely replace human literary creation nor make human writers obsolete in the foreseeable future. What is more likely to happen is the human-machine collaboration advocated by weak AI. In the future, machine writing is likely to reshape the form and boundary of literature’s production, communication and acceptance, and build a virtuous, complementary and symbiotic human-machine literary ecology.

Huang Wenhu is from the School of Journalism and Communication at Huaqiao University.

edited by YANG XUE