Does AI get an A+? Assessing a ChatGPT editorial

The emergence of artificial intelligence (AI) as a tool for content generation has sparked debate within multiple fields. In particular, the generative language model "chat generative pre-trained transformer"—more commonly known as "ChatGPT"—is a user-friendly AI chatbot trained and designed to produce human-like written output based on user prompts and is (seemingly) smart enough to explain complex scientific ideas. ChatGPT can generate subject-relevant text, actively participate in brainstorming, produce creative work, and even write code. At this point, it is not a question of will this become the "new norm" as a content generation tool, but simply when. In the scientific publishing world, there are ongoing discussions on how to apply AI and platforms such as ChatGPT. Clearly, as a writing tool, it can be used (at the very least) to help revise and polish original works (i.e., as a kind of powerful spellchecker). On the other end of utility, it can be used to generate original articles altogether, bypassing the need for human intervention or oversight. Such applications work very well when the generated content is simply a summary of facts (i.e., like reporting the prior night's football scores). But what about more subjective output? In my role, I try to write one editorial per issue for Matter. It doesn't help our impact factor, but I like to think that I bring a fresh perspective to topics relevant to materials science, publishing, and research in general. I try to bring a little levity and entertainment, not only to get my point across but potentially to inspire and motivate my readers to think in new directions. A lofty goal for sure. Can I be replaced by ChatGPT? It is a concerning thought. I've played around with the tool before, and it can definitely output useful text on pretty much any topic. But is the work any good? Can ChatGPT serve as an editor? Only one way to find out… While it has been a while since I've graded student papers, I dusted off my mental red pen and decided to evaluate an essay produced by ChatGPT. One prompt, one output, no revisions. My prompt (produced 4/11/2024): "Write a 600-word editorial discussing the pros and cons of using ChatGPT to write an editorial for an academic journal." The unannotated version of the output is shown in Figure 1. Feel free to read in full before we delve into the prose and peel back the editorial prowess of ChatGPT… Starting with the title:Title: The Dual-edged Sword of ChatGPT in Academic Editorial Writing Cute. It included a little wordplay, taking the "pros and cons" prompt into a "dual-edged sword." Not the most creative, but definitely high-school-level wordplay and acceptable. It gets the point across. A good start. In the realm of academia, the utilization of artificial intelligence (AI) technologies like ChatGPT for editorial writing in academic journals has sparked both intrigue and skepticism. This innovative approach presents a unique set of pros and cons, stirring debates among scholars and researchers. Delving into this discourse unveils the dual-edged sword that ChatGPT wields in academic editorial endeavors. I have few issues with the introduction, which may be a little lofty ("sparked both intrigue and skepticism" is fine, if not a little whimsical), and we see the use of the "dual-edge sword" metaphor from the title (more on that later). Of note, AI Phrase Finder, which used a dataset of 50,000 ChatGPT responses, found that "delve" was in the top 10 words used by ChatGPT [Note: I snuck in a "delve" a little earlier for the keen readers]. Other common words used by the chatbot included "explore," "captivate," and "tapestry." "In the realm of…" is also a common phrase. Be wary. To kick off the main text, we are presented with three advantages of using the LLM to write editorials:Firstly, the merits of employing ChatGPT in academic editorial writing are notable. One of its primary advantages lies in its efficiency. ChatGPT can swiftly generate coherent and well-structured editorial content, saving considerable time for editors and authors alike. By harnessing AI, journals can potentially expedite the publication process, facilitating a quicker dissemination of scholarly knowledge. Hard to argue with the efficiency of AI. This prompt took a mere 30 s to produce an essay. I wish there was less talk of generalities, and perhaps also mentioning the efficiency of iterating an existing work (not just generating a new article). On the one hand, the word limit likely truncated additional examples and supporting arguments. On the other hand, this is likely the strongest point being made, so perhaps it warranted extended discussion. In terms of raw content creation, ChatGPT clearly has immense value. But can ChatGPT recognize its own strongest argument? As a follow-up to the initial prompt, I asked: Of the three pros discussed—efficiency, accessibility, and creativity—please rank them in terms of value. ChatGPT replied:Efficiency tops the list due to its direct and immediate impact on the publication process. Saving time for editors and authors can lead to faster turnaround times for manuscripts, accelerating the dissemination of scholarly knowledge. This can have tangible benefits for the academic community, enabling researchers to stay abreast of the latest developments in their field and facilitating quicker exchanges of ideas and findings. So ChatGPT does recognize that efficiency is the key value-add. Again, from a persuasive essay perspective, lost points for not stressing this in the paragraph. Moreover, ChatGPT has the capacity to enhance the accessibility of academic discourse. Its ability to generate content in multiple languages can broaden the reach of scholarly publications, fostering inclusivity within the academic community. This democratization of knowledge aligns with the ethos of academia, wherein the exchange of ideas should transcend linguistic barriers. The argument here is that ChatGPT can effectively be used as a translation platform. This is correct, but it would be equivalent to using an iPhone for phone calls only—you are missing out on many more powerful features. I could also use ChatGPT as a spellchecker or a thesaurus. However, we already have developed tools for that. Maybe ChatGPT is just being modest. Furthermore, ChatGPT can serve as a valuable tool for overcoming writer's block and stimulating creativity. Its capacity to generate diverse perspectives and insights might inspire authors and editors to explore unconventional ideas, thereby enriching the intellectual landscape of academic discourse. In this regard, ChatGPT can be viewed as a catalyst for innovation within scholarly writing. For the final advantage of ChatGPT, the platform self-espouses creativity. A little overselling in my opinion (what happened to modesty?). I will agree it serves as a valuable tool for overcoming writer's block. It can kick-start written works with a simple prompt (at the very least) and re-formulate notes into complete sentences and paragraphs. From a creative side, it can certainly generate relevant ideas—a key word or two may be sufficient to drive ideas in a new direction. But the prompts and requests must arise from the user, so it is not without some "creative dependency." Enriching the intellectual landscape of academic discourse is perhaps hyperbole, but "viewed as a catalyst for innovation"… I like that turn of phrase. A catalyst, of course, increases the rate of reaction by lowering the energy required to produce a product. ChatGPT can reduce the energy required for writing but not eliminate it. Now we move on to three cons of using ChatGPT:However, despite these notable advantages, the integration of ChatGPT in academic editorial processes also raises significant concerns. Foremost among these is the issue of authenticity and credibility. While ChatGPT can mimic human writing to a remarkable extent, there remains a palpable disconnect between AI-generated content and the nuanced thought processes inherent in human-authored works. This raises questions regarding the reliability and trustworthiness of editorials produced by ChatGPT. Ok. Before I mention the content of this paragraph, I have to mention the use of conjunctive adverbs: moreover, furthermore, however. Other choices could be accordingly, hence, indeed, nevertheless, etc. While conjunctive adverbs can link sentences and paragraphs together as one moves on from point to point, their overuse can make writing droning and seemingly automated. With that being said, the first disadvantage expressed by ChatGPT is the clear elephant in the room—a question of authenticity, credibility, and reliability. Giving some leeway in the interpretation of "authenticity," it is the "human vs. computer" question, wherein there is a near uncanny valley in AI-generated content and human creativity. While AI-generated content can output a logical discussion, it lacks certain insight and perspective. It is akin to asking a middle schooler to read Shakespeare aloud: they can read the words, but they don't understand the nuance of what they are saying. It just feels… off. A more cynical interpretation must admit the (un)reliability of AI-generated content, which is often prone to mistakes and falsehoods, also known as "hallucinations." As ChatGPT is excellent at sounding authentic, these plausible falsehoods can be easily overlooked, even by experts. Here, ChatGPT should admit that (sometimes) it is mistaken. Note that these hallucinations are a well-known issue and are being addressed/improved with each generation of LLM. Currently, explicit recognition that ChatGPT is not infallible (and doing a simple factcheck) should be demanded for any implementation. Additionally, the potential for bias in AI-generated content poses a formidable challenge. ChatGPT learns from vast datasets, which may inadvertently perpetuate biases present within the training data. This raises concerns about the objectivity and impartiality of editorial content produced by ChatGPT, potentially compromising the integrity of academic discourse. A key concern of AI-generated content is indeed bias, be it LLM or other applications (i.e., materials discovery). This bias, however, does not arise from the algorithms of ChatGPT itself, but rather a reflection of inherent bias due to the human-generated input training data. The presence of these biases in LLM output could magnify preexisting human tendencies for cognitively appealing and not necessarily informative, or valuable, content. I.e., left unchecked, AI-generated bias reinforcing human bias. Simple awareness of such bias (along with steps to mitigate) can be used to counteract any predispositions. Indeed, AI is not subjective as humans are, either consciously or unconsciously. AI cannot differentiate between authors with a high h-index coming from reputable institutes unless it is explicitly told (programmed) to do so. If anything, AI can uncover potentially embedded systematic bias in the academic publishing sphere. Furthermore, the reliance on AI for editorial writing may diminish the role of human expertise and creativity in academic publishing. While ChatGPT can mimic human writing styles, it lacks the ability to engage in critical thinking, discern context, and exercise judgment—a hallmark of human-authored editorials. The erosion of human involvement in editorial processes may undermine the quality and authenticity of academic publications. Wait… now a stated disadvantage is diminishing creativity? The third described advantage was "stimulating creativity" with a potential "capacity to generate diverse perspectives and insights." A single feature can be a blessing and a curse, but I doubt ChatGPT is nuanced enough to recognize that creative advantages and disadvantages are two sides of the same coin. Or the aforementioned double-edged sword. Just as this essay extracts three pros and three cons for using ChatGPT, which could be good to generate ideas, connecting the relationships and implications between the three pros and cons is lacking. I.e., there is a lack of critical thinking, context, and judgement. For AI-generated content, it is more of a "just the facts" presentation of examples. User interpretation is required. Thus, for pure information dissemination, ChatGPT is clearly valuable. But in an editorial context—i.e., where opinion, speculation, and perspective are not only warranted by desired—ChatGPT will devalue the format. In conclusion, the utilization of ChatGPT in academic editorial writing embodies a double-edged sword, presenting both opportunities and challenges. While it offers efficiency, accessibility, and creativity, concerns regarding authenticity, bias, and the diminishing role of human expertise loom large. As the academic community grapples with the implications of AI integration, a balanced approach that leverages the strengths of ChatGPT while mitigating its limitations is imperative. Ultimately, the judicious use of ChatGPT in academic editorial writing should be guided by a commitment to upholding the integrity and rigor of scholarly discourse. The conclusion is pretty standard. Hard to find flaws. It references the double-edge imagery from the title and intro, recaps the six described pros and cons, and calls for caution when using AI as a content-generation tool. The final line is also a nice takeaway, appealing to the nobleness of academia (without a hint of irony, clearly). Of note, the format of Chat's short essay is a standard Intro + 3 + 3 + Conclusion, where an introductory paragraph sets the scene, three "pros" are outlined, followed by three "cons", and a short summary to conclude. This aligns well with the communication ethos "tell them what you're going to tell them, tell them, then tell them what you told them" (typically attributed to Aristotle). Also, the essay/editorial came in at 488 words (the prompt requested 600). If you have word minimums, make sure to check!