Many have experienced the difficulties of writing — sleepless nights hunched over the keyboard, futile searches for the perfect word, and the seemingly impossible task of translating formless ideas into comprehensible paragraphs. The process can often feel exhausting and take many months, years, or even decades to master. However, with the rising applicability of generative AI, new online tools are looking to change that.
Boo.ai describes itself as “a beautifully simple writing app that gives you superpowers,” Lex is “a modern word processor that enables a radically new way to write,” and Sudowrite can help “write a novel from start to finish. In a week.” Using language models such as Claude 2 by Anthropic and variants of GPT-3, GPT-3.5, and GPT-4 by OpenAI, these AI writing tools can brainstorm ideas, generate plot lines and character arcs, find the perfect synonyms, and write and edit thousands of words in seconds. With writing appearing easier and more accessible than ever, the question arises of whether such enhancing tools should be promoted or encouraged. In order to best answer this, a thorough consideration of the act of writing, its value, and how the usage of AI can change that is necessary.
Firstly, writing is valued because it is the result of a difficult and often painful process. Writer’s block is a phenomenon familiar to everyone who has tried to put thoughts to paper — even the greats, such as Tolstoy, have suffered from it. The triumph of the artist is pushing past various obstacles to produce something that truly represents them, exhibiting great vulnerability and thoughtfulness in the process. What generative AI hopes to offer is a shortcut, an easy way out of these challenges. But these shortcuts hinder the growth of the artist, leading writers to be more occupied with choosing between various premade options than coming up with their own.
At the same time, through the usage of these tools, a new issue emerges: In a work produced jointly by a human and a computer model, how much of the intellectual property belongs to the human, how much to the model, and how much to the model’s programmer? Drawing the line here seems even more difficult.
Secondly, when engaging with generative AI, it is important to consider the materials used to train it — books, written wholly by human writers. An investigation published last month by The Atlantic showed a dataset of more than 191,000 pirated ebooks, called Books3, that were used without the permission of the writers to train generative models by Meta, Bloomberg, and others. According to the article, books were stored in the dataset as “large, unlabeled blocks of text.” Out of the total amount of books used in the dataset, the majority published in the past 20 years, 183,000 were associated with author information. Written by authors such as Margaret Atwood, Zadie Smith, Stephen King and more, and spanning across numerous genres, these books were used to train AI to better communicate and produce long, natural-sounding answers.
Upon learning the news, many authors were not ecstatic. Min Jin Lee, whose books “Free Food for Millionaires” and “Pachinko” were used in the dataset, wrote on X (formerly Twitter), “Here to report a theft. I spent three decades of my life to write my books. The Al large language models did not ‘ingest’ or ‘scrape’ ‘data.’ Al companies stole my work, time, and creativity. They stole my stories. They stole a part of me.”
In order to produce human-like answers, the AI models were fed human-produced writing. Thus, if AI-enhanced tools are used in the future to produce new books, these newly-published titles can be in turn used to train further, more advanced models. However, that would put the industry in a loop of relying on AI-generated art to produce more AI-generated art, providing no progress for the craft.
This accounts for a third contradiction regarding why we value writing, and art in general. Throughout the centuries, the written word has been used as an expression of freedom and core values, birthed by personal experience, and it has inspired movements and genres. Books have served as markers of history. With the emergence of these centralized, uniform tools for creating, this ability of books to shape the progression of history would be lost forever.
Furthermore, books aren’t the only medium affected by generative AI. One of the main concerns of the SAG-AFTRA strike is that artificial intelligence will lead to less employment and unfair use of the actors’ images to create fake performances. Held jointly with the Writers Guild of America strike — which ended less than a month ago — SAG-AFTRA is still struggling to finish negotiations. And last year, many visual artists were shocked to discover that their paintings had been used, without consent nor compensation, to train Stable Diffusion, a text-to-image model released in 2022. Thus, artificial intelligence is finding its way into numerous facets of our creative lives.
There is one thing that I believe AI-generated art will never be able to encapsulate — the relationship between the artist and the receiver. We read knowing that someone exists behind the words, that our thoughts align with someone else’s. As beautifully stated by C.S. Lewis, “We read to know we are not alone.” This connection between writer and reader can only be facilitated by an author’s intentional choice of words, themes and symbols. We enjoy the writings of Márquez, Austen, Morrison, and many others not only because they are great on their own, but because there was someone behind them who labored over the best way to tell the story. There can be nothing intentional about generative, derivative art.
If, however, AI becomes an inevitable part of our current creative lives, then defined rules, such as those proposed by the Authors Guild regarding consent, compensation, transparency, and labeling of AI-generated content, should be correctly implemented. The inability to do so will only hinder the progress of art and self-expression.
