Strong Examples of AI Writing
|
|
|
Figure 1. Four stages of the writing process. In general, we recommend that authors draft important documents themselves because AI drafts are technically imprecise and because during the drafting stage, you often uncover insights about the content. In addition, authors should carefully review any external document that AI has written or rewritten because AI writing is technically imprecise.
|
 Figure 2. Depiction of manager's Inbox (photo by rawpixel from Freerange Stock).
|
Revision of Emails
A manager of more than 300 engineers and 300 staff members uses AI to revise her emails. She drafts the emails to convey what she intends to say. However, given the long list of emails in her Inbox (Figure 2), she then asks AI to revise each email to "make it sound better." In some cases, she adds more instruction to the prompt, such as to "convey empathy for" a particular circumstance that the recipient has. According to this manager, she always proofreads what AI has revised, but often does not make changes. Overall, she likes the ways that AI rewords her emails. In addition, AI has saved her significant time in the revising stage of the writing process. |
|
Drafting Summary of Research Projects
As requested by upper-level management, a manager had to write a summary of fifty research projects at a large institution. This summary was to emphasize the value of those projects, which focused on energy, from both a technical and a societal perspective. The deadline for the summary was only 36 hours. As a first step, the manager gave the principal investigators 32 hours to write descriptions of their respective projects. With only 4 hours to go until the deadline, the manager used AI to synthesize the individual descriptions into a single coherent summary. For a sense of what that synthesis entailed, see Figure 3. After several iterations, the manager submitted the summary to management. This summary was well received by management, who touted this summary as the standard for the institution. |
Figure 3. Symbolic representation of the diversity of energy projects that AI synthesized for the summary.
|
Figure 4. START Laboratory at Penn State. START stands for steady thermal aero research turbine. This laboratory was one of the laboratories discussed in the expert's marketing analysis.
|
Market Analysis of Research Laboratories
To help her in preparing proposals, an expert in gas turbines requested AI to provide a market analysis of U.S. laboratories that researched gas turbines. In particular, the expert asked to know what major laboratories existed, what their focus was (compressor, combustor, turbine, or so forth), what the scope of their research in that area was, and what technical readiness level the laboratories were. In addition, the expert asked what research gaps for gas turbine research existed overall among these laboratories. Shown in Figure 4 is one such laboratory. The expert, who is one of the leading authors in this field, was deeply impressed by the analysis. From her perspective, the analysis mentioned all of the laboratories that she expected and did a good job of describing the scope of the laboratory in which she works. She estimated that a marketing team would have taken two weeks to do this analysis, which took her only a few minutes. |
|
Drafting Summary of Technical Design Report
In a design course that my team teaches, students write an initial design report (3 to 4 pages) that defines the design problem and establishes the customer needs, metrics, and target specifications. Depicted in Figure 5 is one such design problem: to design a leash clasp for service-dog owners who have low hand dexterity. For the past six semesters, after the students submitted the report and received feedback on both the content and writing, we have had students use AI to draft a summary of the report that is no more than 150 words. Typically, students have written only a one-sentence prompt and attached the revised report. Not surprisingly, when ChatGPT 3.5 and 4 came out, the quality of the AI summaries was low (compared with summaries written with traditional methods). Although understandable by a wide audience, the AI summaries did not target the report's primary audience: a technical manager who is interested in both the design and the design process. For that reason, the summaries lacked the technical details that the manager expected. However, ChatGPT 5.0 was much improved. After just 15 minutes of iterations, the summary matched the best summaries written with traditional methods. |
Figure 5. Photo that depicted one design challenge in the course. This photo appeared in the main text for a report that AI provided a summary (photo by Matthew Klint (CC BY-NC 4.0).
|
Photo from newsonline.library.vanderbilt.edu
|
Revising of Research Papers
A number of researchers are using AI to revise their research papers. In one case, an international graduate student drafts her papers and then uses AI to revise the drafts to make better connections between details and to correct errors of error. After using AI, she proofreads the paper for technical precision. Overall, this young scientist feels that using AI not only saves her time, but also significantly improves the writing style and removes distracting errors of grammar, punctuation, and usage. |
|
Tightening of Proposal
An international research scientist used AI to tighten a proposal that he had drafted. In this use of AI, the scientist was specific in the writing of the prompt and went through multiple iterations to reduce the length of the proposal from 25 pages to 20 pages. In addition, the scientist carefully proofread the proposal after using AI to make sure that no essential information was removed. In the end, the scientist felt that using AI saved him significant time. |
Photo by ulleo on Pixneo
|
*AI was not used to research, draft, or revise this webpage.