In September 2020 a controversy arose in the literary world. The Espasa Prize for Poetry was won by Rafael Cabaliere, a Venezuelan poet, unknown until then, and on whose Instagram account there were only phrases full of common places. The surprise at this qualitative leap in Cabaliere’s work, and the fact that it was unknown to the public and critics gave rise to several hypotheses. Among these, two stand out: Cabaliere is a bot with artificial intelligence (AI), used by Espasa to win the award; and the other is that it was the sex robot Henry, released two years earlier and could recite simple poems. In both cases it would be an AI capable of learning quickly and generating increasingly complex poems.

Whether these hypotheses are true or not, artificial intelligence has been able to outperform humans in many activities for decades, being more precise and faster. In 1996, the Deep Blue program beat chess champion Gari Kasparov. This program evaluated possible plays much faster than its human opponent, but it had still been taught to play by other humans. In 2018 the computerized AlphaZero system went further. Watching matches of the Asian game Go, he taught himself and defeated champion Lee Sedol. If machines are capable of teaching themselves, is it possible that they can do such a creative task as writing literature? Could it be true that Cabaliere is a computerized system?

Various artificial intelligence systems capable of writing a story have been developed since the beginning of the 21st century, with varying degrees of success. The most advanced of these systems today is The Generative Pre-Trained Transformer (GPT-3), launched in 2020 by the Open AI laboratory in San Francisco. It has 175 billion parameters and has been trained with nearly a trillion words collected from the Web and digitized books. The user only has to start writing a paragraph and the system itself would take care of completing it. Although the style is clean, the program has shown shortcomings. The GPT-3 returns contradictions because, in reality, the generation of words is not guided by an understanding of the text. In addition to this, it presents biases such as the reproduction of stereotypes and sometimes writes and sexist or racist ideas. It really is a program with a large memory and playability, but unable to understand and generate something truly authentic.

Cuban writer Daniel Burguet explains that books have been written about the structure that a story must have in order to be attractive to the public. “That is an A-B-C, the story is cut into three moments and three turning points are introduced and thus you make sure that the story is dynamic. So you can disassemble a Hollywood movie by transforming it into a scheme. In addition, the possible dramatic functions do not reach 40 and the archetypal characters are 12. Series like Friends or The Big Bang Theory have around 7 archetypal characters. If you train an AI who can handle this scheme, she might be able to generate a story with these elements. But even so it would be unable to break the very rules that it has learned, this being the main difference with humans, who are capable of generating something totally different from what they have learned. ”

For these reasons AI is used for more schematic tasks such as writing news, editing texts or translations. It is one thing that AI can learn and quite another is the ability to imagine and invent something totally new. Writing narrative and literature until just a few years ago was something unique to humans. Recently, software, such as the aforementioned GPT-3, have come to write novels and poetry books, but they are still unable to break stereotypes and clichés.

Computing and AI are advancing rapidly. Computerized systems are increasingly complex, fast, and efficient, but they are unlikely to fully replace human writers. Narrative and poetry need a high level of imagination and feelings that software is not capable of emulating. The Cuban science fiction writer Erick Mota argues that the talent and originality of a writer is difficult to match. “AI can rival the human level when it comes to writing, but writing is something else, it requires an emotional compulsion. The need to write or the writer’s gaze, I don’t think that can be copied. Even humans who want to be writers can’t. ”

In the not too distant future, computerized programs will replace human writers in tasks such as writing notes or editing texts, as they have already done in the search for information. This can be seen in an article published by the British newspaper The Guardian, written entirely by an AI, in which it is possible to see deficiencies in the narrative quality, but it presents a clean and congruent style.

Even when thinking about the future, it is still unimaginable that AI could have feelings, the imagination and creative capacity to achieve a transcendental work. What is very likely is that the software will end complementing the work of a writer. During the 19th century, many authors employed helpers called “blacks or ghostwriters,” people who completed a work based on the author’s draft. Without creating the plot, the people, or the setting, they would complete the descriptions, dialogue, and correct the style. Authors such as Alejandro Dumas Senior, Edward Stratemeyer, and Manuel Fernández y González at some point employed ghostwriters. This work could be taken up again in the 21st century by artificial intelligence. Far from being a threat to writers, computerized programs would be a help. The editor’s work would be replaced by the AIs.

Returning to Cabaliere’s case, it is unlikely that the poem, due to its complexity and sentiments, was ever written by an IA. The hypothesis that it is an exquisite corpse of various poets, or a joke made by a writer, would be more logical. Computerized programs, more than a threat to humanity and writers in particular, represent a help and will not be able to replace human beings in this work.

Credits: EL Soto