“When OpenAI launched ChatGPT, with zero fanfare, in late November 2022, the San Francisco–based artificial-intelligence company had few expectations.”
The inside story of how ChatGPT was built from the people who made it1
MIT Technology Review’s article “The inside story of how ChatGPT was built from the people who made it“1 starts with the quote above. As the article goes on to discuss, the creators of ChatGPT were not expecting it to become a topic of conversation for nearly every field it could potentially touch along with the general public. In looking to analyze ChatGPT’s website from the perspective of writing in digital media, I first had to narrow down my focus. A quick Google search came back with over 1.4 BILLION results.2
Even focusing on ChatGPT’s impact on education only narrows the results to 125 million.3 I offer these results not as an example of how I approached my research for this analysis but as an example of how much information is available for a program that’s only been open to the general public for seven months. . So rather than cover how AI like ChatGPT fits into the history of writing in digital media or its impact on how writing is taught, I will focus on how it impacts my future; technical writing.
I’ve been asked multiple times if AI will replace potential jobs in a field I’ve yet to start working in, and I can safely say no. Many articles have already been written about the potential role of programs like ChatGPT in technical writing. Most see it as a tool. Technical Writer HQ posted an article, “11 Documentation Experts on the Future of AI“,4 where they asked documentation writers their thoughts on the influence of AI. The common theme through their responses is that AI is a tool. It can be a helpful tool that future writers must learn and understand, but it won’t replace them.
As Jennifer Achaval is quoted in the article, “Humans are needed to provide context and a distinct perspective that AIs are not (currently) capable of.” as the article notes, “She believes technical writers and editors review the article thoroughly to ensure the high quality of AI tools. She also thinks, “Make sure it meets current industry standards and makes sense to the intended audiences – whether that audience consists of subject matter experts or laypeople.”4
The article summarized another writer, Dylan Small’s views on AI, “The limitation of AI lies in the fact that it can’t replicate human experience because certain key elements, including conducting face-to-face research, taking meaningful pictures, and engaging in human interaction, are all deeply connected to human experience. These processes demand some degree of interpretation, the ability to read the context, and physical proximity that current AI systems simply don’t have.”4
As the article concluded, AI can’t replace the uniquely human skills of interpretation and use of language that good technical writers provide.4 Good technical writing is about more than just writing instructions on how to use software. It’s about considering the audience using those instructions and writing them to make it easy for the user to follow. Along with making sure the instructions follow the company’s own branding or style, something AI can’t do unless it’s been given that information.
This leads to another reason AI won’t replace technical writers, copyright. An article by Sarah O’Keefe, AI in the content lifecycle, comments on the US Copyright Office’s recently published rules on copyrighting AI-generated material5 by saying that “Let’s say that an organization has a large amount of reviewed, approved “known good” content, which is (of course) copyrighted. If someone feeds that information into ChatGPT and synthesizes a summary, is the summary copyrighted? The Copyright Office says no.”6 I can’t think of any company that would be comfortable giving up their copyrights on any of the content they’ve created for use, even if it’s just a standard operating procedure manual to be used by their own employees, much less manuals or websites designed to educate the users of their products.
It’s also been well-documented that AI, such as ChatGPT, sometimes gives inaccurate information.7 OpenAI’s help section notes that “ChatGPT is not connected to the internet, and it can occasionally produce incorrect answers. It has limited knowledge of world and events after 2021 and may also occasionally produce harmful instructions or biased content.”8 This means that anything written by ChatGPT must, at the very least, be fact-checked by a human to ensure the information is accurate.
Let’s move on and look at some writing samples from ChatGPT. The first page9 you see once you’ve signed up for the site (free to use with some restrictions, learn more at ChatGPT) lists what the AI is capable of and what some of its limitations are.
If you ask ChatGPT a simple question, such as “how to make a peanut butter and jelly sandwich”10, you get a pretty good set of instructions (even with a typo in my prompt), as shown below.
Writing instructions like this is more challenging than it sounds. A YouTube video by Josh Darnit titled “Exact Instructions Challenge – THIS is why my kids hate me. “11 is an excellent example of how challenging this can be.
Ask ChatGPT something more complex, such as “How do you use Facebook”12, and you will get the answer shown in the screenshot below. While its answer isn’t wrong, it lacks a few things that most people agree are crucial to explaining software use.
generic. The audience could be anyone or no one, but documentation is written to be read. The article “Audience Addressed/Audience Invoked: The Role of Audience in Composition Theory and Pedagogy.” proposes a model of writing based on two theories, the second of which is based on the idea that “…the writing process is not complete unless another person, someone other than the writer, reads the text also. The second assertion thus emphasizes the creative, dynamic duality of the process of reading and writing, whereby writers create readers and readers create writers. In the meeting of these two lies meaning, lies communication.”13
An audience that feels the writer is talking down to them (because the information is too simple) or one that doesn’t understand what the writer is saying (because they’re being too technical or not explaining well enough) isn’t going to read any further. Writing for a specific audience requires a writer to understand their primary audience, how complex the writing should be, and how formal or casual. As the article notes in regards to the audience, the writer’s “…knowledge of this audience’s attitudes, beliefs, and expectations is not only possible (via observation and analysis) but essential.”14
If you ask ChatGPT how it knows what audience to write for15, it says it doesn’t. It relies on you, the user, to give it an idea of the audience the writing is for, but the user still needs to understand the audience to help guide the AI in shaping writing for that audience.
Writing, even technical documentation, is meant to be read and used by humans. That requires the writer to determine and understand who the audience for a piece of writing is likely to be, their level of knowledge on the subject, how formal or informal the writing needs to be, and even what form it should be. The writer must also understand and follow any other requirements necessary for the writing, such as a company’s style or standards, along with ensuring the information in a document is accurate and well organized.
AI programs such as ChatGPT are excellent tools for writers, but they won’t replace humans as writers. As Elima notes in “ChatGPT for Technical Writing Purposes” on Click Help,
“…writing comprises creative art and personal experience that machines may never fully replicate. While AI’s text generation capabilities will keep improving, it will most likely always be viewed as a supplement to human writers rather than a substitute.”16
ChatGPT for Technical Writing Purposes16
or
Works Cited
- Heaven, Will Douglas. “The Inside Story of How ChatGPT Was Built From the People Who Made It.” MIT Technology Review, 6 Mar. 2023, www.technologyreview.com/2023/03/03/1069311/inside-story-oral-history-how-chatgpt-built-openai.
- Chatgpt – Google Search. www.google.com/search?q=chatgpt. Accessed 22 June 2023.
- Chatgpt and Education – Google Search. www.google.com/search?q=chatgpt+and+education. Accessed 22 June 2023.
- Fechter, Josh. “11 Documentation Experts on the Future of AI.” Technical Writer HQ, June 2023, technicalwriterhq.com/interview/documentation-experts-on-the-future-of-ai.
- O’Keefe, Sarah. “AI In the Content Lifecycle.” Scriptorium, June 2023, scriptorium.com/2023/06/ai-in-the-content-lifecycle.
- “Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence.” Federal Register, 16 Mar. 2023, www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence.
- May, Jonathan. “Analysis: ChatGPT Is Great at What It’s Designed to Do. You’re Just Using It Wrong.” PBS NewsHour, 5 Feb. 2023, www.pbs.org/newshour/science/analysis-chatgpt-is-great-at-what-its-designed-to-do-youre-just-using-it-wrong.
- Introducing ChatGPT. openai.com/blog/chatgpt.
- “ChatGPT.” OpenAI, chat.openai.com.
- “How do you make a peanut butter and jelly sandwhich?” prompt. ChatGPT, 24 May version, OpenAI, 22 June 2023, chat.openai.com.
- Josh Darnit. “Exact Instructions Challenge – THIS Is Why My Kids Hate Me. | Josh Darnit.” YouTube, 26 Jan. 2017, www.youtube.com/watch?v=cDA3_5982h8.
- “How do you use Facebook?” prompt. ChatGPT, 24 May version, OpenAI, 22 June 2023, chat.openai.com.
- Ede, Lisa, and Andrea A. Lunsford. “Audience Addressed/Audience Invoked: The Role of Audience in Composition Theory and Pedagogy.” College Composition and Communication, vol. 35, no. 2, National Council of Teachers of English, May 1984, p. 169. https://doi.org/10.2307/358093. Accessed 31 May 2023.
- Ede, Lisa, and Andrea A. Lunsford. “Audience Addressed/Audience Invoked: The Role of Audience in Composition Theory and Pedagogy.” College Composition and Communication, vol. 35, no. 2, National Council of Teachers of English, May 1984, p. 156. https://doi.org/10.2307/358093. Accessed 31 May 2023.
- “How do you know what audience you’re writing for?” prompt. ChatGPT, 24 May version, OpenAI, 22 June 2023, chat.openai.com.
- Elmira. “ChatGPT for Technical Writing Purposes.” clickhelp.com, Feb. 2023, clickhelp.com/clickhelp-technical-writing-blog/chatgpt-for-technical-writing-purposes/?utm_source=medium-com&utm_medium=kesi-medium.
or