February 24, 2024

Technology Websites

I Need Technology Websites Right

ChatGPT safe from cheating, contend ed tech firms; teachers have doubts

Two leading education technology companies have released products claiming to make the artificial intelligence tool ChatGPT safe from academic dishonesty, but teachers are not so sure.

The homework learning app Brainly last month launched the beta version of Ginny, a ChatGPT-powered guide that lets users simplify or expand answers to complex math and science problems.

That came right after Turnitin.com — a leading academic misconduct website that teachers use to catch plagiarism in written essays — released an AI detection tool that claims to be 97% effective at flagging computer-generated writing.

“Ultimately, it is the student using the AI that will be doing the learning, and it is up to that student to use the AI in the right way,” Bill Salak, Brainly’s chief technology officer, told The Washington Times. “The student comes to Brainly to seek understanding, and Brainly’s technology generates explanations and solutions in response to that.”

“I don’t think anyone can tell you the future of AI or how fast it will develop, but we’re willing to make the investment to keep updating,” added Annie Chechitelli, chief product officer for Turnitin. “We’re confident that we’ll be able to keep up in detecting patterns of misconduct.” 

ChatGPT, Microsoft-controlled AI that grows smarter at mimicking human behavior as it assimilates more knowledge in a massive database, presents the illusion of talking with a friend who wants to do your work for you. The chatbot can compose college essays based on assignment prompts, solve math or physics equations and pass the exam required to become a doctor.

Public K-12 school districts from New York City to Los Angeles have banned the next-generation technology since it became available in late November — as have several countries, including Italy, North Korea and China.

Teachers say they have no way of telling when students are using AI to cheat on writing assignments, noting that even the new Turnitin detector sometimes flags the wrong essays. 

“The rapid growth of AI and other technologies will always supersede the plagiarism detection tools,” said Trey Vasquez, a professor of special education at the University of Central Florida, a pilot school for Turnitin’s AI detection tool. He has tested ChatGPT with students for months.

ChatGPT has pledged to produce a watermark to help teachers identify its output. Users have noted that the software produces writing quirks that let companies such as Turnitin find gaffes in computer-generated essays.

But no tool exists yet to keep students from cheating with it on math and science equations by using the explanations supplied by GPT-powered apps like Brainly, to prevent them from copying AI essays into another file, or to catch plagiarism in foreign-language assignments like a Spanish essay on “Don Quixote.” 

“I don’t think [the anti-cheating websites] are keeping up,” said Thomas Plante, a member of the American Psychological Association who teaches at Santa Clara University in California. “The genie has left the bottle and we are all scrambling to either put the genie back or, more likely, learn to live with and adapt to it being out.”  

According to Brainly’s Mr. Salak, the app guide Ginny uses an alpha version of GPT-4 to relieve frustrated students and parents who have the answer to a homework problem but cannot explain it.

“Many students already have the answer, but it is too complex, or they are missing a step. They are looking to Brainly to show them the steps to get there,” Mr. Salak said, noting that Brainly began working with ChatGPT’s creators last year to develop the feature.

The ability of ChatGPT to synthesize complex information in seconds and to grow smarter as it gobbles up more information means teachers need to move beyond “rote memorization of factoids,” said Connor Boyack, president of the free-market Libertas Institute, a Utah-based think tank.

“Learning should always be about more than regurgitating information, and perhaps ChatGPT will pressure educators into focusing on higher-order opportunities to inspire and elevate young people,” said Mr. Boyack, author of the “Tuttle Twins” children’s book series and a recent American history textbook.

As anecdotal reports emerge of thousands of students using ChatGPT to cheat, Turnitin’s Ms. Chechitelli said the company’s new AI detection feature looks for the sort of too-perfect writing that a word processor’s autocomplete feature would insert in a student essay.

She said signs of AI writing include made-up facts and footnotes, repetition of the same points, an inability to answer questions about current events and a lack of deeper analysis.

“It’s basically a massive, sophisticated autocomplete,” Ms. Chechitelli said of the new Turnitin feature. “Humans write with a lot more idiosyncratic behavior in word choice and style than an AI.”

Educational institutions next year will have to start paying for the Turnitin AI tool, which is currently free, as part of a premium package.

In the meantime, some college professors have returned to making students take in-class essay exams in blue books — no computers allowed — after two years of COVID-induced hybrid learning. 

“We know that allowing students to wrestle with ethical dilemmas like when and how to use AI promotes their growth in decision-making,” said Sara Rimm-Kaufman, a professor of education at the University of Virginia. “Yet the new AI is really putting that to the test, especially in high-stakes academic situations.”

Teaching programs have started preparing future educators for the AI generation of students, said Jodi Feikema, chief academic officer and provost at the American College of Education in Indianapolis. 

“The opportunity is for educators to change their approach,” Ms. Feikema said. “Rather than focusing on banning the use of ChatGPT, educators should be focused on embracing the technology and encouraging students to use critical thinking in their evaluation of ChatGPT outputs.”