Is it Right to Let ChatGPT Write?

Future Point Digital is a research-based consultancy positioned at the intersection of artificial intelligence and humanity. We employ a human-centric, interdisciplinary, and outcomes-based approach to augmenting human and machine capabilities to create super intelligence. Follow us at: www.futurepointdigital.com.

Probably like many reading this post, I have always enjoyed the process of research and writing. There’s something highly satisfying, even thrilling at times, about deeply exploring a complex topic or phenomenon and then writing about it in a way that makes it broadly accessible to very busy people who might not otherwise have the opportunity to do such research. For those of us who happen to fall into this category, adding value to society in this way is our raison d'être.

Now, however, it seems our algorithmically driven cousins, while not completely butting us out of this space, are relegating us to being glorified (perhaps not-so-glorified) prompters rather than writers. A little over a year ago, when I first immersed myself in ChatGPT and all things generative AI, I didn’t feel particularly threatened as a writer. When I asked the platform (ChatGPT-3, at the time) to write something on a particular topic (say, Quantum Cognition), and/or to rewrite something I had researched and written, its output ranged from poor to barely acceptable. It was also extremely prone to hallucinating or just making things up when it didn’t know the answer (the algorithms are optimized to complete the next word, sentence, paragraph…line of code, etc. based on statistical probabilities, not accuracy). In short, I wasn’t inclined to view the platform, as it was then, as anything other than a snapper-up of unconsidered trifles.

Flash forward to just over a year later, however, and I have had to seriously reconsider my relationship with our “chatty” friend. For instance, before Chat-GPT, if I wanted to write a white paper on a topic that I knew little about, I would first have to conduct an expansive literature review (preferably based on empirical, peer-reviewed research published in top refereed journals). This process could take weeks if not months, depending on the topic, and what I was hoping to achieve with the article. Now, however, ChatGPT-4 is capable of producing a white paper, on any topic, with output that is generally quite acceptable —in my view, at least. One can even prompt the platform to provide empirical support for every position or claim in the AI produced paper. (This, of course, has to be carefully confirmed, but in every instance, so far, I’ve found it to be quite accurate and sound). Even creative writing (a side hobby for me) is something the platform is improving upon quite rapidly —at least as a muse of sorts, and/or editor. (See the section below on creative writing for a demonstration).

Why the Quantum Leap from a Year Ago?

My guess is that anyone who has delved, even a little, into the AI space already knows the answer to this question, at least generally, if not specifically. I started to write this section myself, based on the AI acumen I’ve arduously, yet passionately attempted to refine since ChatGPT hit the public domain —but, why bother? ChatGPT-4 (if prompted correctly) can produce a more than adequate answer to the question and will likely capture points or provide considerations I might have missed.

[Note: see the sidebar below on proper citation of ChatGPT output.]

According to ChatGPT, the improvement in the quality of output from ChatGPT-3 to the current version is due to several factors related to advancements in technology, training processes, and algorithm optimization. Here's a brief overview of why you might notice such improvements:

1. Larger Training Data: The models have been trained on a more extensive and diverse dataset, encompassing a wide range of topics, styles, and contexts. This larger dataset helps the model understand and generate more accurate and nuanced responses.
2. Advanced Algorithms: The algorithms have been refined and optimized to better understand the context and intent behind queries. This includes improvements in understanding nuances, reducing irrelevant or inaccurate information (hallucinations), and generating more coherent and contextually relevant responses.
3. Fine-Tuning and Reinforcement Learning: The use of fine-tuning techniques and reinforcement learning from human feedback (RLHF) has allowed the model to better align with desirable output qualities such as accuracy, relevance, and coherence. Human feedback helps identify and correct errors, biases, and hallucinations, leading to more reliable responses.
4. Increased Model Size and Complexity: The architecture of the neural network, including the number of parameters, has evolved. Larger and more complex models can capture more detailed patterns in the data, leading to improved understanding and generation capabilities.
5. Focused Research on Reducing Hallucinations: There has been focused research on minimizing hallucinations—instances where the model generates plausible but false or irrelevant information. Techniques such as dataset cleaning, better training methodologies, and post-training evaluation have contributed to reducing such occurrences.
6. Continuous Updates and Iterations: The model undergoes continuous updates and iterations based on user feedback, performance assessments, and advancements in AI research. This iterative process allows for constant refinement and improvement of the model's capabilities.

Overall, the enhancements in model design, training methodologies, and algorithm optimization collectively contribute to the noticeable improvement in the quality, accuracy, and reliability of the generated content compared to previous versions (OpenAI ChatGPT, personal communication, February 19, 2024).

Sidebar: Citing ChatGPT Output Citing the output from an AI model like ChatGPT in a white paper depends on the citation style you are following (e.g., APA, MLA, Chicago). However, since interactions with AI models like ChatGPT are not published works, they are often treated similarly to personal communications or unpublished interviews. Here’s how you might cite this output in a few common styles: APA Style (7th edition) In the text, you would reference the information as a personal communication, since it's not retrievable data for your readers:(OpenAI ChatGPT, personal communication, February 15, 2024)Note: According to APA guidelines, personal communications are not included in the reference list because they do not provide recoverable data. MLA Style (8th edition) In MLA style, because the conversation with an AI is not a published source and not retrievable by readers, you would also treat it similarly to a personal interview:Text citation: (OpenAI)In your works cited list, you might format it as a personal interview:OpenAI. Personal interview. 15 Feb. 2024. Chicago Style (17th edition) For Chicago style, which is often used in historical and some scientific publications, you might note the information in a footnote rather than the bibliography for similar reasons:Conversation with OpenAI's ChatGPT on February 15, 2024.Because this is akin to personal communication, it typically wouldn't be included in the bibliography in Chicago style. Note: When citing AI-generated content, it's important to clarify the nature of the source for your readers, especially since the information cannot be independently verified in a traditional manner. Additionally, consider the context of your white paper and the preferences of your target publication or audience, which might have specific requirements or expectations for citing such sources (OpenAI ChatGPT, personal communication, February 19, 2024).

Demonstrating ChatGPT’s Non-Creative Writing Capabilities

Despite my initial dismay at being relegated to the role of prompt engineer versus a true content creator —or, in the case of non-creative writing, a purveyor of what I hope is useful knowledge —I have adjusted (more precisely, am adjusting) to this new reality by embracing our motto at FuturePoint Digital, that human intelligence + artificial intelligence = super intelligence.

I suppose I could find my happy place by sticking my head in the proverbial sand and pretending that none of this is happening. But then my writing would likely pale in comparison to the quality of those who have learned to harness the multiplier effect of human/AI interaction -in this case, content creation. This may be a poor analogy but, imagine if doping in athletics was legal. If you’re an athlete who really needed to compete professionally, would you abstain from doping? You could…but you’d likely not be able to compete at a professional level. (Ughh…admittedly, as I type this I’m having to direct my rational brain to tame my emotional brain, but I will do so rather than allow myself to become a living, breathing anachronism).

So, what does this human/AI interaction look like, vis-a-vis non-creative writing? (As referenced above, we’ll explore ChatGPT’s growing capabilities in creative writing in the subsequent section). Perhaps the best way to answer this question is to look at a paragraph I wrote for a forthcoming white paper on AI, Digital Twins, and Emotional Intelligence. I started with the following paragraph but got temporarily stuck finding the right words to complete the last sentence. So, I copied the paragraph into ChatGPT-4 and asked it to try to find a properly worded ending. Not only did it comply (admirably, I might add) but it took it upon itself to re-write the entire paragraph. I will let the reader be the judge of which is the preferable paragraph (no pride of authorship here):

Below is my initial paragraph, as I wrote it:

As we outlined in a previous post (Why the Value of Uniquely Human Skills Will Increase in Value in the Age of AI), we firmly believe at FuturePoint Digital that being quintessentially human will increase in importance and value as artificial intelligence becomes increasingly sophisticated. In short, the more we can understand about ourselves and others in a variety of situations, the better we will be able to capitalize on the symbioses between humans and AI-driven platforms. Traditionally speaking, we believe that the humanities and the social sciences, in conjunction with rich life experiences, have presented the best means of gaining this type of knowledge. Now, however, AI may present new and dynamic opportunities for enhancing and advancing our capacity to understand ourselves and others, and to make quantum improvements..(this is where I was momentarily stuck).

Below is ChatGPT’s Rewrite:

As we outlined in a previous post (Why the Value of Uniquely Human Skills Will Increase in Value in the Age of AI), we at FuturePoint Digital hold the conviction that being quintessentially human will grow in importance and value as artificial intelligence becomes more advanced. In short, the more we can understand about ourselves and others in a variety of situations, the better we will be able to capitalize on the symbioses between humans and AI-driven platforms. Traditionally, we have seen that the humanities and social sciences, enriched by a breadth of life experiences, offer the best avenues for acquiring such insight. Now, however, AI may present novel and dynamic opportunities to and advance our capacity to comprehend our own nature and that of others, leading to substantial improvements in how we interact, empathize, and collaborate across different cultures and viewpoints in increasingly complex social environments.

OK, so this example shows only subtle changes to my original version, but it did give me some ideas for how to complete the final sentence, and also made what I thought were some improvements to the preceding sentences. I then asked it to re-write specific sentences —ones that I wasn’t totally sold on.

In the end, I used pieces of the original and pieces of the re-write. The section, as rewritten, will be presented in this week’s post of the above article (keep an eye out for it), but I am satisfied that the final version of this paragraph, based on a mixing of ChatGPT’s output and my original creation, is superior to what either of us produced independently. I think this is a good demonstration of how this new dance is already going.

What About ChatGPT’s Creative Writing Capabilities?

In an article last year I attempted to demonstrate ChatGPT-3’s ability to create something novel, in the form of a vignette, based on a detailed prompt that I provided. The prompt included instructions vis-a-vis setting, time period, brief descriptions and the names of each character, the general theme I wanted it to convey, etc. The output demonstrated that Dostoevsky or Chekhov it was decidedly not! Though it was able to complete the task in a seemingly creative, albeit childlike, way.

Today however, the child seems to have grown into a young adult. While neither Fyodor nor Anton, were they alive today, would be shaking in their boots, I think they would find our chatty friend to be of great use, if nothing other than a well-informed muse, sitting on their shoulders. But I also think they would find the current version of the platform to be of great assistance ferreting out new phraseology for maximum impact, or even analyzing initial drafts and providing feedback, maybe even editorial advice, etc.

I hesitate to provide examples of my own feeble attempts at fiction writing, especially after invoking the names of Dostoyevsky and Chekhov, to demonstrate ChatGPT-4’s current capabilities in the realm of creative writing, but it may be more useful to see what it can do for an amateur writer, versus re-writing classic literature. Last year I wrote the first 30 or so pages of a historical fiction novel I’m slowly stumbling and fumbling through in my spare time. Not trusting my own judgment of the work product, I copied the 30-page draft into ChatGPT and asked for an analysis. Specifically, I asked it to evaluate the manuscript in terms of character development, narrative style and voice, dialogue, pace, setting, emotional impact, innovation and originality, potential contribution to the genre, etc. etc.

Including the full analysis and additional output is beyond the scope of this article, but suffice it to say that it was extensive, incisive, insightful, and highly useful in terms of feedback for improving the novel. One example that I will include here is feedback ChatGPT-4 provided in terms of the novel’s pacing. The platform noted an issue with respect to pace in the initial analysis, so I asked it to provide specific examples that might illuminate the point.

Below is ChatGPT-4’s response to my prompt to provide specific examples of how to improve the pace of my novel:

ChatGPT:

To enhance action and streamline description, you can focus on tightening the prose and ensuring that each sentence moves the story forward. here's an example of how you might revise a section for a more dynamic pace while maintaining essential details:

Original: "It was nearly dusk when 8-year-old Evan Morgan heard the familiar repeat of a 22-caliber rifle coming from the back 40 of the 160-acre farm upon which his family barely scratched out a living. He knew who was firing the shots and he couldn't run fast enough in his bare feet to be a spectator of the event before complete darkness fell."

Revised for action: "As the last light of dusk clung to the sky, Evan Morgan, barefoot and breathless, raced across the farm's fields. Each crack of the .22 rifle hastened his steps, a staccato urging him on to witness the sharpshooting spectacle before night fell."

Original: "Abe was tall and lean with sharp eyes and a piercingly observant countenance. He kept his eyes trained on the sky as Evan approached --again moving the rifle with fluidity and precision, as if he was somehow connected to the twisting, darting objects in the darkening sky."

Revised for action and pared-down description: "Abe stood tall against the evening sky, his gaze sharp as he tracked his aerial targets. The rifle was an extension of his will, each motion fluid, each shot precise, connecting him to the dance of dusk's last flyers."

In the revised sections, I’ve condensed the description to ensure it serves the action. Description still sets the scene, but it's more integrated with the character's movement and the unfolding events. This method keeps the reader engaged with the story's progress while maintaining the atmospheric setting you've built.

Additionally, try to show rather than tell wherever possible. For instance, rather than saying "Abe was the best shot in the county," demonstrate his skill through action: "Abe's aim was true, dropping another bat from the sky, confirming his unspoken title as the county's finest marksman."

Remember, these are just suggestions to illustrate a point, and any changes should be made in keeping with your narrative voice and the story you want to tell.

(Me again) ChatGPT’s ful analysis included 11 specific points of feedback to consider (again, too voluminous to include here), and then proceeded to rewrite the draft to demonstrate how it could be improved, vis-a-vis all of the recommendations. In a future post we will explore ChatGPT’s current creative capabilities in much greater detail, including its improved ability to create vignettes and even short fiction based on detailed prompts. Spoiler alert, it’s still not at the point in which talented creative writers should feel threatened (based on my assessment) but it’s certainly making undeniable improvements. (I even asked it to reduce the draft manuscript I provided to a truncated poem using iambic pentameter —which it accomplished in seconds).

Nevertheless, the concluding punchline, in my view, is that the human writer still must possess the requisite knowledge, curiosity, insight, creativity, and instincts to know what questions to ask, and what topics to explore, etc. This will require that humans dive ever deeper into what it means to be uniquely human. If one wants to write a compelling white paper, or even a novel, it is the human who must have the spark to light the fire. And it is the human who will need to control how the fire burns.

History also teaches us that human insatiability always finds ways to outrun the limits of technology, thereby propelling further innovations and advancements in a never ending symbiotic, circular dance. For those of us who find meaning in research and writing, either non-fiction, fiction, or both -advancements in generative AI platforms like ChatGPT may feel like a momentary “gut punch” of sorts, but I have no doubt humans will quickly find a way to harness the AI horse and…(whoops, I’m stuck again — ChatGPT, can you help me bring it home?) …ride it to new frontiers of creativity and discovery, redefining the boundaries between the creator and the tool, and in the process, expanding the very horizons of human potential.

How might FuturePoint Digital help your organization explore exciting, emerging concepts in science and technology? Follow us at www.futurepointdigital.com, or contact us via email at [email protected].

About the Author: David Ragland is a former senior technology executive and an adjunct professor of management. He serves as a partner at FuturePoint Digital, a research-based technology consultancy specializing in strategy, advisory, and educational services for global clients. David earned his Doctorate in Business Administration from IE University in Madrid, Spain, and a Master of Science in Information and Telecommunications Systems from Johns Hopkins University, where he was honored with the Edward J. Stegman Award for Academic Excellence. He holds an undergraduate degree in Psychology from James Madison University and has completed a certificate in Artificial Intelligence and Business Strategy at MIT. His research focuses on the intersection of emerging technology with organizational and societal dynamics.