Posey's Tips & Tricks
Can AI Reconstruct the Past?
An experiment with Microsoft Copilot shows AI can turn fragmented notes into a surprisingly coherent and plausible first-person narrative -- though it still falls short in length, voice accuracy, and fully capturing lived experience.
Last weekend, while doing a bit of spring cleaning, I stumbled across a packet of information left over from one of my commercial spaceflight exercises. Being that I am no longer with the program, I have been working on updating my memoirs to include some of the experiences that I didn't initially talk about, and my newly discovered packet was choked full of notes pertaining to a training exercise that I have never written about before.
As I began sifting through the various documents (many of which were already in digital formats), I got an idea for a fun experiment. Copilot for Microsoft 365 is designed to help users find information within a collection of documents, but it also excels at authoring new content. This led me to wonder: could Copilot take my fragmented notes and instructor handouts and transform them into a compelling first-person narrative, as if written by someone who actually experienced the training? I also wanted to see if Copilot could mimic my writing style.
Now before I go on, I want to be completely clear. I am in no way planning on letting Copilot write my next book for me. This exercise was just an experiment. While there is a chance that I might get Copilot to help me turn the collection of documents into something cohesive, I plan to use Copilot only as a starting point to establish the chapter's overall structure. I am going to do all of the writing myself, as I always do.
So with that said, let's talk about my experiment. To get started, I opened the Microsoft 365 portal, created a new notebook and then uploaded the documents. From there, I wanted to give Copilot Chat a bit of direction, so I types a brief explanation of what the various files were. As an example, one of the documents was a course overview PDF. Another file contained the instructor's PowerPoint presentation. There was also a file listing all of the procedures to be used during the training exercise, and a file containing detailed notes that I had taken during the training, documenting some of my own experiences. I explained to Copilot exactly what each file was.
Next, I instructed Copilot to create a detailed first person narrative reflecting what the training was like, from the perspective of one of the trainees. I also told Copilot to create the text using my own writing style.
Upon doing so, Copilot created a document that was exactly what I had asked for, but not what I wanted. My eventual goal was to write an entire chapter based on this particular experience, but Copilot gave me a three paragraph summary, not a book chapter. Additionally, the text only very loosely mimicked my writing style. On the plus side however, Copilot had indeed taken information from all of the documents and turned all of the random notes into something cohesive.
At this point, I decided that I needed to give Copilot a bit more help. I instructed Copilot to make the text longer and more detailed. I also gave Copilot another chapter from my book, and told Copilot to use the same writing style as had been used in the sample chapter.
This time, the resulting text was a lot closer to what I was looking for. Copilot created about four pages of text. This is still way too short for a book chapter, but it's a lot better than the three paragraphs that Copilot had initially written. Of course, I never actually told Copilot to create a book chapter. Besides, I am going to be rewriting all of the text myself, so it doesn't really matter that Copilot did not create a super long document.
The document's style was a closer reflection of my own writing style, but it still wasn't perfect. I could indeed tell the difference between my own writing and text that Copilot had written, posing as me.
The thing that I was most curious about was whether Copilot could take a collection of notes and convincingly tell the story of what it was like to go through the training. I honestly wasn't quite sure what to expect. After all, I was basically giving Copilot a collection of technical documents and telling it to turn those documents into a memories.
The resulting narrative was actually pretty good. There were a lot of details about the experience that Copilot had to fabricate, because those details weren't mentioned in the notes. Even so, I found that when Copilot was forced to fill in the gaps, it guessed correctly about half of the time. Even when Copilot got the details wrong, the text that Copilot generated was at least plausible. It didn't make any wild leaps that led to completely illogical conclusions.
The resulting text was far from perfect, but it was still way more compelling than I ever expected it to be. Overall I was impressed with what Copilot had come up with.
About the Author
Brien Posey is a 22-time Microsoft MVP with decades of IT experience. As a freelance writer, Posey has written thousands of articles and contributed to several dozen books on a wide variety of IT topics. Prior to going freelance, Posey was a CIO for a national chain of hospitals and health care facilities. He has also served as a network administrator for some of the country's largest insurance companies and for the Department of Defense at Fort Knox. In addition to his continued work in IT, Posey has spent the last several years actively training as a commercial scientist-astronaut candidate in preparation to fly on a mission to study polar mesospheric clouds from space. You can follow his spaceflight training on his Web site.