Chapter 1 Understanding AI's Growing Influence

The landscape of online writing is on the brink of significant transformation due to the emergence of artificial intelligence (AI).

An AI could have composed this article, and you might not even realize it. Before diving deeper, let me share my background. I spent three years at an AI startup, where I contributed to developing a real-time sign language translator. My experience has acquainted me with natural language processing (NLP) and natural language generation (NLG) systems—essentially, AI that can comprehend and produce text.

Throughout my writing journey, I have explored the interplay between AI and written language, reporting on advancements from OpenAI's GPT-3—whose remarkable capabilities captured global attention—to recent innovations like Wu Dao 2.0, Microsoft's GitHub Copilot, and Facebook's BlenderBot 2.0.

A noteworthy event in Japan is the Hoshi Shinichi Literary Award, where both humans and machines can submit literary works. Since 2016, AI-generated novels have participated in the competition, with one almost clinching the prize. The title "The Day a Computer Writes a Novel" encapsulates a unique human trait: recursion, where the AI narrates the tale of its own creation. Despite advancing through the initial stages, the AI faced challenges, such as effectively conveying character descriptions, as noted by judge Satoshi Hase.

This was in 2016, before GPT-3 was even conceived. The foundational technology, the transformer architecture that powers modern language models, emerged in 2017. Fast forward to today, and language AIs have matured substantially. Natural language generation is at the forefront of AI research, posing a growing threat to many jobs reliant on language, particularly writing.

In this article, I will outline five reasons why AI is a potential risk to writers, examining the state-of-the-art in language generation, the proficiency of contemporary language models, and the primary methods through which individuals could harness these systems to jeopardize our livelihoods.

Chapter 2 AI's Versatility in Writing

AI's capabilities have expanded significantly, enabling it to produce text on almost any subject. The breakthrough came with the transformer architecture, a neural network that processes text efficiently. Researchers trained a model using vast amounts of internet data without the need for extensive labeling, which proved both time-saving and cost-effective. This model became known as OpenAI's GPT-3.

Upon its release, developers quickly recognized its hidden potential. The capabilities of the model exceeded what was described in OpenAI's initial publication. Users discovered it could generate essays, poetry, music, jokes, and even computer code. It could engage in discussions about analogies, philosophy, recipes, and even existential queries. Additionally, it proved capable of crafting advertisements, revising emails, and emulating the styles of renowned authors.

However, just a year later, GPT-3's abilities have begun to seem outdated. Numerous organizations have adopted OpenAI's model as a template to create even more advanced systems. In May, Google unveiled LaMDA, a chatbot capable of engaging in meaningful and factual dialogues. It even showcased a whimsical personality, as demonstrated by its impersonation of Pluto, expressing desires—an intriguing concept for an AI.

In June, the Beijing Academy of Artificial Intelligence (BAAI) introduced Wu Dao 2.0, the largest neural network to date, boasting 1.75 trillion parameters—ten times the size of GPT-3. This system powers Hua Zhibing, the first virtual student, capable of continuous learning, poetry composition, and coding, all while retaining its knowledge.

Then in July, Facebook launched BlenderBot 2.0, a new chatbot that can store and recall information while engaging in conversation. Unlike GPT-3, which lacks memory and becomes outdated quickly, BlenderBot 2.0 can form virtual connections, demonstrating its adaptability.

This is just the beginning of a new era. AI has witnessed numerous advancements since the inception of the transformer architecture, and in the next decade, these systems are expected to improve even further, fueled by more data, enhanced computing power, refined architectures, and features such as multimodality and prompt programming.

AI has become adept at manipulating language, with written expression being its forte. However, just because AI can generate text doesn't mean it does so with quality.

Chapter 3 AI's Writing Skills Compared to Humans

While nearly everyone can write, producing quality writing is a different story. AI has now reached a level of writing that rivals good human authors. A notable example arose in July 2020 when Liam Porr, who runs the newsletter "Nothing but Words," published an article on productivity, which many believed was penned by a human. However, it was entirely crafted by GPT-3, showcasing the model's ability to mimic human writing styles convincingly.

Despite the article's lack of genuine creativity, it successfully deceived readers into believing it was authentic. This incident highlights the potential of GPT-3 to generate content that resonates with human audiences. In fact, studies indicate that human accuracy in detecting GPT-3-generated articles hovers around 52%—barely above chance.

Imagine the implications if we were inundated with AI-generated content, often without our awareness. Such a scenario may already be unfolding.

The genuine concern lies not in AI's ability to write but in its capacity to produce text that aligns with the standards of proficient human writers.

Chapter 4 The Limitations of AI in Writing

One significant limitation of language models is their lack of comprehension. While AI has mastered the mechanics of language, it does not grasp meaning. Its outputs are generated through probability calculations, devoid of genuine understanding. This distinction is crucial, as the value we derive from written text is often tied to context and intention—qualities AI lacks.

However, there is a silver lining: AI does not need to understand language to produce coherent and structured writing. For instance, GPT-3's article may lack intention, yet readers can still extract value from it. This paradox challenges our perception of value in writing, as we often associate intention with worth.

Although AI can produce subpar writing, humans have the ability to edit and refine. We can assess our work critically, seeking feedback and making revisions. Unlike AI, which lacks self-awareness, we can learn from our writing experiences.

Writing automation is primarily about efficiency. The most successful writers often advise, "Just write." Even a poorly constructed first draft can be improved through editing.

With GPT-3, I could generate ten distinct self-help articles in a matter of minutes, selectively curating and refining the best parts for publication.

Chapter 5 The Accessibility of AI Writing Tools

Developing high-quality AI writers is an expensive endeavor, often limited to major tech companies. For instance, training GPT-3 cost OpenAI approximately $12 million. To sustain their operations, OpenAI partnered with Microsoft, which invested $1 billion to support their pursuit of artificial general intelligence.

Nevertheless, accessing GPT-3 has become relatively inexpensive for users. Currently, subscription rates start at €100 per month, allowing users to generate around 3,000 pages of text—far more than an individual could produce in a year. If users can monetize the content generated, the initial investment becomes trivial.

The primary barrier to utilizing state-of-the-art language models is simply the monthly fee. This ease of access enables individuals to generate significant quantities of content without extensive writing skills, presenting a disconcerting reality for established writers.

As AI writing tools become commonplace, the perception of writing as a valuable skill may diminish. While AI may not entirely replace human writers, the market value of our work could decline, even if the quality remains constant.

Chapter 6 Finding Hope Amidst Change

AI's impact on the writing industry is undeniable, but it extends to virtually all sectors. Current projections suggest that 40-50% of jobs may be automated within the next 15-20 years. The reality is that even creative and knowledge-based professions are not immune.

While the potential loss of writing jobs is concerning, it is crucial to recognize that AI's influence is widespread. This presents an opportunity for policymakers to devise comprehensive strategies to support affected workers.

OpenAI's CEO, Sam Altman, recently stated that AI may disrupt white-collar jobs more significantly than blue-collar roles. Writing, coding, and administrative tasks are all vulnerable. However, it is improbable that AI will fully eradicate the online writing industry. Technology tends to induce changes in unpredictable ways, and complete industry overhauls are rare.

The most practical response is to embrace AI as a collaborative partner. Just as we have integrated other technologies—cars, computers, smartphones—into our lives, we can find ways to work alongside AI. By leveraging our unique strengths and combining them with AI capabilities, we can enhance our value in the marketplace.

Ultimately, the best course of action is to continue honing our writing skills, focusing on producing content that resonates with audiences. This article explores potential futures, but the takeaway is to remain proactive and adaptable in the face of change.

Subscribe to my free weekly newsletter, Minds of Tomorrow, for more insights on Artificial Intelligence! Feel free to connect with me on LinkedIn or Twitter! :)