The Register: Alcorn State professor uses white text hidden in assignment instructions to phish out AI-using cheaters, catching 32 out of 35 students
A long time ago, during one of the agonizingly arduous job searches in my life, I once filled out an application that took me the better part of an hour to complete, what with all the PII I had to disclose, as well basically manually typing in all of my resume’s information, while uploading the resume itself, as well as composing a thoughtful cover letter that basically said hi my name is danny look at my resume to see why i believe I can do your shitty job.
Submit, confirmation email five minutes later, check the link to see what application status page looks like.
Applicant rejected.
Fucking seriously? In five minutes?
It became apparent that I had been an immediate victim to bot screening, little possibility that any human had seen my application, and that my application had been phished out presumably because it did not contain any of the unknown criteria that the position had been looking for.
But really, I was incensed that nearly an hour of my already frustrating-on-account-of-being-unemployed life’s time had been wasted in less than five minutes because a fucking bot didn’t see enough of the words they needed to see in order to even consider my application for human eyes.
After the initial rage subsided, I decided to take a different approach to building my resume, taking into consideration that my resume was more than likely going to be passed through bot screeners that do little else than analyze the words they identify, with greenlit terms scoring higher and increasing my chances of getting seen by a human screener who then can possibly be convinced to reach out.
In both the Word doc and the InDesign-created PDF variants of my resume, I created floating text boxes on a separate layer, and wrote in every single term that I thought might increase my odds at ticking green flags from a bot screening; skills, software, awards, orientations, but mostly race-related.
adobe microsoft photoshop illustrator indesign acrobat flash html css java final cut video illustration mac mobile desktop print digital omni concepting art direction management
award winning accolade gallery featured magazine publication ad creative organization
gay lgbt lgtbq pride college university masters veteran military
asian black hispanic latino minority korean japanese african woman spanish bilingual language fluent
And then I’d set the entire block of text to white, and build out my actual resume on top of it. Invisible to the human eye, but in the view of zeroes and ones that the bot sees, every single one of these terms is as clear as day, and my hope was always that if any of them helped flag my resume as worth looking at, why the fuck didn’t I think of this sooner.
I’m not saying that this tactic necessarily worked like I had invented Google, but I can say that I was never screened out of a position in five minutes ever again. Maybe it’s on account of bots becoming as lazy as the humans who implemented them, maybe it’s because the bots’ processes aren’t disclosed anymore. Maybe it’s all of the above, or maybe, just maybe, my tactic might have actually proved to be useful.
Needless to say, this story about the Alcorn State professor using white text to phish out and bust his own students, tickled me in a manner that brought great joy and great satisfaction to my soul. The tactic is as old as when font color selection in word processor software was launched, and no manner of AI advancement has been able to get around the fact that at some point, humans are still involved in the process.
I already have a contentious opinion and relationship with AI, and of course everyone including this professor knew that people were lazy and were already utilizing it to do things that they should be doing and trying to pass the results off as their own. And frankly, the fact that he ended up busting that many students in his class isn’t at all surprising, and the real surprise is how three of them didn’t get caught.
Because it’s not like the three that didn’t get caught probably weren’t cheating with AI either, it’s just they were probably the students that either had the 1-2 IQ points higher to have the wherewithal to review and proofread their AI-generated papers to catch the dummy prompt’s trapdoor in their assignments, and/or maybe they had an appearance setting in their device’s OS running to where they had any sort of colored background other than default white, and were able to see the dummy trap prompt in the assignment before they ran it through an AI.
Frankly, I’m more annoyed with the verbiage that states that the students were only partially failed, because I’d have failed them all outright. AI-generated work is tantamount to plagiarism in my opinion, and in this day and age, a lot of time if people are not punished severely, they’re going to just look for ways to circumvent in the future, and not learn all the same. I’m not saying that making them repeat and pay for class all over again won’t result in the same outcome, but at least the school might make some money on it in the process, and smarter people might think twice about doing it and might actually learn something when they do things themselves.
All the same, the real winner in this entire story is the white text strategy, which has proven to be as unbeatable now as it was in the 1980s when WordPerfect and Microsoft Word first came into existence. As long as humans are involved in the process, failure will be too.
