Artificial Intelligence (AI) is pervading nearly every aspect of our lives. Businesses use it, educators use it, students use it, creative artists use it. AI has been described as “invisible and inscrutable.” The “invisible” part is easy to understand; we don’t see the data processing going on behind the scenes at lightning speed. The “inscrutable” part means we can’t always trace how AI came to its answers or identify all the data it uses to reach an answer.
Out with friends recently, one related how he needed to provide a biography to introduce him at a conference where he would be speaking. The article that ChatGPT spit out in a matter of seconds was stellar, but not a description of him. His response was, “Wow, I’d like to be that guy.”
Arriving home, I decided to see what ChatGPT (which is just one variant of AI) would write about my land surveying career, since I am asked for a biography every time I teach. While there were some factual parts (yes, I began my career at a time when our profession was predominantly male), there were many exaggerations and outright falsities. It got my childhood aspirations wrong as well as my undergraduate alma mater and major. And no, I definitely wasn’t quickly accepted by colleagues when I entered surveying. I’d give ChatGPT about 50% in factuality, and 100% in how believably and readably it was written.
Then I decided to see what fabulous tale ChatGPT would spin about my life as a martial artist. Amazingly, I found out that my parents, who in real life were a small step above being couch potatoes, valued physical fitness, and encouraged me and my (not yet born) siblings to explore various physical activities. This biography is 700-plus words of fantasy, with a few near-facts and still fewer full facts thrown in. For this attempt I’d give AI somewhere about 2% for truth and another 100% for its writing ability. Clean writing with an upbeat tone: it sure sounded good!
From my experiments, it seems ChatGPT scrapes what information it is fed or can find from the Internet and fabricates the rest if it finds little or nothing. While it will put words together in good syntactical and grammatical construction based on statistical validity of one word following another (relying on how it was initially programmed and what it learns through usage), the chatbot has no idea if what it is spitting out is true or not.
Can we use AI for anything? Absolutely. We just have to check what it churns out. I asked ChatGPT to write an outline with supporting citations for a class I’ve presented several times, just to see what it would suggest. Its outline was remarkably close to mine. So far, so good. The citations, though, were not so great, some being to non-existent material. That’s worrisome. There are lazy people who will ask AI to write articles or reports and not verify everything it prepares.
Different AI programs are meant for different applications. Some generate documents, some automate complicated formulae creation for spreadsheets from your input data, others create and edit images, videos, or music. There are hundreds of programs out there to help with a multitude of tasks.
While I have worked only with one AI tool, everything I have researched repeats one rule for best results across all: For the best results, be very specific in your prompts.
For text generators, the first result may be on the right track but not there yet. After an initial response, re-prompt to add more details or correct misstatements AI has made. For instance, I entered information about a fictional employee (name, responsibilities, quality of work) and asked for a job performance review. AI created a document in seconds. I added new prompts to note this employee wrote reports in jargon instead of keeping clients in mind and didn’t always help colleagues when asked. A modified document scrolled onto my screen. This process can be repeated many times. You can upload files and photos for the program to include. You can copy the generated text and paste it into whatever document you want. But always read it and check it for accuracy. See my examples above!
For help in writing style, either attach a file or type in a draft of what you already have written. You can ask AI to write your piece in a different tone, make it more readable and persuasive or more formal and authoritative. But a word of warning here, too: I uploaded this column ending at the prior paragraph and told ChatGPT “Make this article more authoritative but engaging.” The result was two-thirds of the original length, with most details removed and an inappropriate insertion of the word “engaging.” This version read like dull “fluff” to me. I used Microsoft Word’s “document stats” calculator in its Editor tool to compare the readability of the version you see above with the AI-generated one.
| Chat GPT: grade level 13, with readability ease at 30.2%. |
| Wendy: grade level 9, with readability ease at 58.5%. |
I told ChatGPT to make its version more readable, then ran its newly shrunken version (now half of the original, with even sparser details) through Microsoft Word’s “Document stats” again:
| ChatGPT: grade level 10, with readability ease at 49.0% |
Score one for the humans?
A note of caution
Because ChatGPT pulls from the web (when it isn’t making things up) and then shares what it generates with the web, it can spread its falsehoods and fantasy (along with your secrets) to the world. To minimize this, create only “temporary chats,” under the “File” menu of the app. Also turn OFF ChatGPT’s ability to share data under “Settings>Data Controls” and disable model training.