How AI Writing Tools Are Helping Students Fake Their Homework

Getting good grades in school may soon be about artificial intelligence (AI) as much as hard work. 

Online software tools that help students write essays using AI have become so effective that some teachers worry the new technology is replacing creativity during homework assignments. Students are increasingly turning to these programs that can write entire paragraphs or essays with just a few prompts, often leaving teachers none the wiser. 

"As far as I can tell, it is currently not that easy to detect AI writing," Vincent Conitzer, a professor of computer science at Carnegie Mellon University, told Lifewire in an email interview. "These systems do not straightforwardly plagiarize existing text that can then be found. I am also not aware of any features of the writing that obviously signal that it came from AI."

Homework Helpers

The use of AI writing tools by students is on the rise, anecdotes suggest. Conitzer said he’s heard one philosophy professor say he would shift away from the use of essays in his classes due to concern over AI-generated reports.

Tools based on Large Language Models (LLMs), such as GPT-3/X, have seen tremendous improvement over the last few years, Robert Weißgraeber, the managing director of AX Semantics, an AI-powered, natural language generation (NLG) software company, said in an email interview. Users enter a short phrase or paragraph, and the tool extends that phrase or section into a lengthy article.

Don't expect LLMs to replace real authors anytime soon, though, Weissgraeber said. GPT-3X tools are just "stochastic parrots" that produce perfect-sounding text, "however when looked at in detail, they produce defects called 'hallucinations,'—which means they are outputting things that cannot be deduced from the arguments built into the input, data, or the text itself. The perfect syntax and word choices can dazzle the reader, but when looked at closely, they actually produce semantic and pragmatic gibberish."

Catching AI Cheaters

AI-assisted writing programs are now so effective that it's hard to catch cheaters, experts say. Other than making students write in a supervised setting, perhaps the best way for teachers to avoid the use of AI writing is to come up with unusual topics that require common sense to write about, Conitzer said.

"For example, I just had GPT-3 write the beginning of two essays," he added. "The first was about whether free speech should sometimes be restricted to keep people safe, a generic essay topic about which you can find all kinds of writing online, and GPT-3 produced sensible text listing the pros and cons.

"The second was about what a teenager who was accidentally transported to the year 1000 but still has her phone in her pocket should do with her phone. GPT-3 recommended using it to call her friends and family and do research about the year 1000."

Erin Beers, a middle school language arts teacher in the Cincinnati area, told Lifewire in an email interview that content from programs that rewrite or paraphrase content sticks out like a "sore thumb" at the middle school level. 

"I can usually spot fraudulent activity due to a student's use of complex sentence structure and an abundance of adjectives," Beers said. "Most 7th-grade writers simply don't write at that level."

Beers said she's against students using most AI writing programs, saying, "Anything that attempts to replicate creativity is likely limiting a writer's growth."

Weißgraeber recommends teachers not be fooled by smooth-looking prose that may have been generated by AI. "Look at the argumentation chains," he added. "Are all statements grounded in correlating facts and data that are also listed?"

However, despairing teachers take note. There's at least one upside to students using AI tools, Conitzer contends. 

"In principle, students could learn quite a few things from AI writing," he said. "It often produces clear and well-structured prose that could serve as a good example, though the style is usually generic. Students could also learn more about AI from it, including how it sometimes still fails miserably at commonsense reasoning and how it reflects the human writing it was trained on."