We've all heard of ChatGPT by now. It's completely sweeping the business and education world & shows absolutely no sign of slowing down any time soon.
It's raised a ton of concerns from angry teachers, recruiters, and other professionals across the globe. I'm in no way against AI writing, I use it daily to supplement my articles & optimize my life. I mean I started this blog to help others optimize theirs!
Where I get upset is not when somebody uses AI to write a quick article, finish a report, or generate ideas, it's only when it's becomes an unethical way of getting something done.
Again, I use AI to help write some of my articles as it can speed up some of the mundane writing that makes an article flow better.
But where I really ran into an issue was a few months ago trying to hire a writer. I remember saying explicitly in the job post that you'll actually be allowed to use AI to help write, but I wanted a sample done just to showcase general skills.
Well I got an applicant who seemed super promising. Smart, friendly, and honest (seemingly so).
I had them write an essay on regulating AI, funny enough.
I got an email back a few days later with a link to a Google doc and an article that looked perfect. Like, too perfect.
After reading 3 sentences in I went "yeah no way this was written by a human." It just couldn't be. It wasn't even like they tried to hide it, they just threw the article title into ChatGPT and said "write it"
Well the short answer is it isn't concretely possible. That means there's no foolproof way to determine if a piece of text has been generated by an AI or not.
At the end of the day, you're literally looking at words. AI prediction works by guessing patterns in text and trying to mimic what should come next. If the computer's guess matches with the entered writing, you're more than likely looking at AI.
But tools like ChatGPT are amazing, especially the new GPT-4.
It mimics human writing with such finesse, it often leaves even the most experienced readers baffled. From the flow of sentences to the richness in vocabulary, there's nothing that immediately stands out as non-human.
AI-generated content is sometimes indistinguishable from human-written content, and this is both fascinating and alarming.
Even though I ran the writing through several AI detection tools (Originality, Content at Scale, and Copyleaks) and it came back as 100% AI AND even though I was positive it was written with AI (because I've used ChatGPT enough to know what the generic syntax looks like) – I STILL couldn't be sure.
It's created a loophole in the content creation industry where people can pass off AI-generated content as their own work.
The fine line between ethical usage and fraudulent practice is blurring. A human and an AI could write on the same topic, and it's becoming increasingly difficult to tell who's who. Again, this isn't saying AI writing is bad, but if someone told you not to use AI and you claimed you didn't when you did, that's pretty unethical. And this is only the start of a massive headache that's about to rock the world.
I mean the writer said themselves it was all them, no AI:
So how can you actually be sure? I mean it's not like plagiarism where you can find a direct source to the content. AI is fairly unique, every time.
Visualize AI Writing with Google Docs and Originality
So how do you do this? First go ahead and run the writing samples through the detectors. See what you get. Here's what my "writer" had produced:
So I was onto something, right?
Now I went to the document after installing the Originality Chrome Extension and ran a report. Once you run it, you'll get a score (but you can ignore that because we just did that a few minutes ago).
Go ahead and click on "View Writing"
You'll actually be able to visualize how the writer constructed their article. It works by placing together the article after each revision (Google docs saves your edits after every sentence or two).
You can play the entire thing and watch the article get "built" right in front of you. I do need to mention that you NEED edit access to the article for it to work (and it obviously has to be written on Google docs).
Watch how it rebuilds the document (this is what the article rebuild actually looks like):
You could clearly see when this writer just pasted in content. And they didn't get it from another website since there was no plagiarism found. So what's the culprit? AI!
The code was cracked.
Reasonable suspicion + AI detection tools + Originality Visualizer = proof enough
This goes way behind what traditional AI detection tools do, it gives you actual measurable proof of copy paste content (which you can match line-by-line to what the detectors say was written with AI).
This gets rid of nasty verdicts, especially ones that are not true. I would hate to be one of the students that got in trouble because of TurnItIn's poor AI detection software.
This aren't reliable, conclusive, and who knows how accurate they are. TurnItIn boasts a 98% accuracy but what does that even mean? Based on what? You can't claim accuracy on something you can't truly measure.
Well... that's pretty much it!
This is the closest method I've found to determine if something was truly written with AI. And you have proof now.
Remember that you will need to have edit permission of the Google doc article, so make sure whoever is submitting writing lets you edit it. Drop a comment if you have any questions!