The New AI Challenge Facing Modern Universities

The introduction of artificial intelligence has significantly influenced universities across the country, and the University of Illinois is no exception. Tools like ChatGPT, Atlas, Gemini, and TurboAI have allowed students to complete high-quality work on a mass scale, often without ever learning the core class material.

ChatGPT, a prominent large language model trained on vast amounts of text, often produces writing with recurring phrases. While professors initially turned to AI-detection tools to combat plagiarism, the rapid advancement of AI has rendered these tools less effective. Furthermore, studies have found these detectors can be negatively biased against non-native English speakers, increasing disparity in the classroom.

A New Era of AI Automation with Atlas

OpenAI’s newest product, Atlas, has sparked fresh concern among educational institutions. Marketed as a “webpage co-pilot,” the tool integrates an AI assistant directly into the user's browser to streamline daily tasks. A more advanced feature, “agent mode,” allows the AI to complete complex jobs without direct human interaction.

For example, a user could ask their agent to filter flights for a vacation while they simultaneously search for a hotel. Atlas can even purchase the prompted findings with user consent. Sam Altman, CEO of OpenAI, explained that this feature “allows AI to perform multi-step tasks autonomously like a virtual computer.”

Student Perspectives: A Double-Edged Sword

Sam Blaker, a senior in ACES, praised the new tool for its efficiency in multitasking. “I have already installed (Atlas), and it is really useful,” Blaker said. “It helps me a lot with investing because I can ask it to do things in the background while I am focusing on a task.”

However, this same capability allows students to consult with the tool to write essays, code projects, or fill in answer blanks on quiz sites like PrairieLearn. With the agent feature, an assignment could theoretically be completed 100% by AI without any student interaction. This development comes as AI usage among students is at an all-time high.

Ritisha Bansal, a freshman in Engineering, has witnessed this heightened AI dependency firsthand. “When midnight comes around and there’s only five minutes left to complete homework, they panic,” Bansal said. “So then you snip the question and get the answers from ChatGPT so you get it done. But this work isn’t your own, and when exams come around, you struggle because you have no idea what you actually learned.”

The Impact on Learning and Recall

Mike Szymanski, a clinical research professor in Business, noticed the uptick in AI usage and conducted a pseudo-study in his Business 301 course. He divided the class into two groups: one could use AI to complete assignments, and the other could not. He then tested both groups on their ability to recall information from the assignment, first after 10 minutes and again 48 hours later.

“The students that used AI had significantly lower recall and originality scores than those who worked with their brains,” Szymanski reported. “When you normalize using AI, you’re basically skipping the part where you build familiarity. And then you leave not knowing what ‘good work’ actually looks like.”

While Szymanski’s research is under internal review, his preliminary findings align with a broader Massachusetts Institute of Technology study that examined how AI writing assistants affect cognitive engagement and learning outcomes.

Redefining Academic Integrity

Using AI to complete assignments generally falls under plagiarism per the University's academic integrity handbook. As AI continues to advance, academic policies may need updating to address this new landscape.

A recent New York Times article highlighted a peculiar trend at the University: an overwhelming number of students were using the phrase “sincerely apologize” in emails to professors, a tell-tale sign of AI-generated text. Szymanski noticed a similar pattern in his own grading. “The pattern was unbelievably consistent,” he said. “I’d get five different students turning in papers that all opened with the exact same sentence... That’s how I knew they were just feeding the course description into ChatGPT and calling it ‘my personal growth.’”

The Real Crisis: Is It Apathy, Not Cheating?

Szymanski believes the problem runs deeper than simple dishonesty. “After watching this play out in my own classroom, I honestly don’t think the crisis is that students are cheating,” he concluded. “I think the crisis is that they don’t see it as cheating. What I saw with my students is that AI didn’t kill their ability to write. It killed their willingness to try.”