3 Ways Artificial Intelligence Won’t Help You Write Well
Artificial intelligence (AI) tools, such as ChatGPT and Bing, have garnered a lot of attention over the past 6 months. Many researchers are excited about the potential of these tools to help them do the challenging work of writing. But although these tools can generate text to help you write something, they cannot help you write well…yet.
In science and medicine, writing well means that you do more than just generate text. Writing well means you craft text into a story that readers can easily understand and enjoy (readability), that shows leadership and builds trust with readers (credibility), and that reflects your knowledge and interpretations (think-ability). These factors—readability, credibility, and think-ability—are part of what makes writing persuasive. And these factors are not addressed by AI tools.
Readability
AI tools work by pulling language from sources that they have been trained on. In scientific and medical writing, these sources are often published literature written by humans. However, most researchers have not had adequate training in writing. Instead, they learn by mimicking what they read in the literature, which is plagued with poor writing. Mimicking poor writing only breeds more poor writing, perpetuating the problem. And because AI tools also learn by mimicking from the literature, they will continue to perpetuate the problem as well.
An important feature of good writing is readability, or the quality of being easy and enjoyable to read. Readability is often measured by readability scores that are calculated based on numerical factors, such as word length and sentence length. When prompted, AI tools can measure readability, and even attempt to revise text to improve readability, based on readability scores. However, readability scores have important limitations that AI does not have the capacity to consider.
For example, readability scores do not measure the power of the story and how it connects with readers, both emotionally and intellectually. These two characteristics are important parts of persuasive writing. Also, readability scores do not measure context, readers’ prior knowledge, readers’ interest in the topic, or the complexity of the concepts. In other words, readability scores do not measure the curse of knowledge.
What is the curse of knowledge? The curse of knowledge is the cognitive bias that occurs when we—unintentionally and unknowingly—assume someone else knows what we know. In other words, we struggle to imagine what it is like to not know what we know. This curse contributes to bad writing. According to Steven Pinker, author of The Sense of Style: The Thinking Person's Guide to Writing in the 21st Century, “The curse of knowledge is the single best explanation I know of why good people write bad prose.”
AI can generate text for you, and even calculate the readability score of that text. But you will still need to do the hard work of overcoming the curse of knowledge and ensuring that the writing is readable for your intended audience.
Credibility
Credibility is a foundational part of science. As a researcher, your credibility influences your reputation as a leader in your field. And the credibility of your research and writing builds trust in your work, in you, and in science.
How do you show credibility? First, you base your ideas on previous work. Although AI can help you formulate ideas based on previous work, AI may generate the same “original” ideas to other researchers who give similar prompts. In other words, those ideas may not actually be original and will not include your unique insight and perspectives.
Another way you show credibility is by citing credible sources. Although AI tools can generate sources, these sources have been wrong and even made up. Also, AI tools are biased and do not have the ability to distinguish good and bad sources. As a result, you need to verify any sources that are generated, and you need to analyze each source to determine whether that source is credible. An AI tool cannot review a publication and tell you whether you believe that the research had sound methods, a robust analysis, or valid interpretations. You only get that information from reading and analyzing the research in the context of your unique knowledge and thinking.
You also show the credibility of your work—and you—in many other ways. For example, you need to describe your methods in detail, present your data clearly, use appropriate statistics, and support your interpretations and conclusions with credible data. Because these aspects of your research are often novel (unpublished and out of AI’s reach), you must write about them, not AI.
AI can point you to sources, but those sources may be incorrect or false. Be sure to vet any sources, because by citing a source, you are telling the world that you believe that source is credible. And the credibility of that source will reflect on your credibility.
Think-ability
Many people believe that writing is hard. But this perspective masks the real challenge. Writing is not hard because the writing itself is hard. Writing is hard because the thinking needed to write is hard.
Although many of us struggle with putting pen to paper (or fingers to keyboard), we often fail to recognize that this struggle is how we refine our thinking. Wrangling words helps us to clarify our ideas and understanding. In essence, the struggle is really an opportunity to refine our thinking and to stimulate new ideas that advance science.
AI tools can write for you, but they cannot think for you. If you fully rely on AI tools, you can strip away valuable opportunities for you to think critically. And if you strip away your opportunities to think critically, you also strip away your opportunities to think uniquely. And your unique thinking is what helps you create original work that distinguishes you in your field and in the world.
AI Won’t Help You Write Well…Yet
AI tools are rapidly evolving. Over time, they will improve and likely overcome some of the challenges described in this article. But no matter how AI tools evolve, we need to use them as assistants (not creators) and approach them as we do research: we need to be constantly aware of their advantages and limitations as they advance. We also need to do the hard work of thinking, verifying the information, and refining the writing that AI generates so that we can craft readable text that shows our credibility and reflects our unique thinking.
Want free tools and templates to help you enhance your scientific and medical writing? Get access to our free writing toolkit!
