Consumer products like Alexa and Google Home can tell us how to prepare a meal and answer dinner-party trivia questions about the Treaty of Paris. So how long before tools powered by artificial intelligence start assuming more of the classroom work that professors handle today?
The answer is simple: They already have. At the University of Michigan at Ann Arbor, students in a statistics class last fall had their writing assignments evaluated by an automated text-analysis tool. And in 2016, some of the students in a computer-science class at the Georgia Institute of Technology were surprised to learn that one of the TAs answering their questions remotely all semester was actually a chatbot powered by artificial intelligence.
The bigger question, however, is this: How does teaching change as these AI tools start assuming more classroom work?
That’s harder to answer. A new grading tool called Gradescope — powered by AI and now being used for more than 175,000 students at 550 colleges — offers one window on the possibilities.
The tool helps automate the grading process without requiring instructors to rely on multiple-choice tests. Several professors who use it say the automation has actually made it easier for them to personalize their teaching — although some of their students have needed some convincing.
Instructors also say the technology has made their grading fairer and faster — the latter an especially useful factor for courses like chemistry, in which students’ mastery of new material depends on their understanding what came before. “Getting feedback from a test you took two weeks ago is not going to help,” says Alegra Eroy-Reveles, an assistant professor of chemistry and biochemistry at San Francisco State University, now in her third year of using the tool. “It’s important to get it in a timely manner.”
Because Gradescope has made grading tests easier, Eroy-Reveles is giving more of them. Instead of three major exams a semester, she now gives eight smaller tests. “It will take me about three hours to grade about 100 of them,” she says. The tool can read handwritten answers and can group together all the correct formulas in one batch and each variation of the wrong answer in separate batches. Thanks to those AI features, she’s able to quickly see what her students are missing.
The timesaving tool was developed by Arjun Singh and Sergey Karayev as Ph.D. students at the University of California at Berkeley. Working as TAs in a computer-science class on artificial intelligence in 2012, they felt buried by the tedium of having to grade 100 tests by hand and, as Singh recalls, “writing the same thing 50 times as your students make the same mistakes 50 times.”
As student of AI themselves, they figured there had to be a way to deploy machine learning to make their own jobs easier. By the summer of 2014 they had written the code that could read students’ names off a form and recognize patterns in handwritten answers, such as chemical formulas for acids, as long as they were entered in a designated area on standardized-test forms. Their software could also identify and cluster those answers into groups. “We built this because we wanted a tool for ourselves,” Singh says.
With another college friend and Singh’s faculty adviser, they established Gradescope as a company. They hoped that the tool’s main AI feature — its ability to read patterns not programmed in ahead of time — would set it apart from the few other automated grading tools in use.
The tool doesn’t do all the work. Instructors, using a standardized form, must still scan in all the tests and determine how much weight to assign to each answer. That’s fundamental to the process, says Singh. For many professors, “the exam is the last line of defense” in a teaching environment increasingly influenced by standardized-course materials provided by publishers and other courseware developers. “They want to be involved in it.”
Seth Anthony, an associate professor of chemistry at the Oregon Institute of Technology, tried Gradescope for a final exam last fall and is now using it for his general chemistry course, which has 40 engineering students. He says the tool has made him more attuned to the patterns of students’ mistakes, which he can more easily spot in weekly quizzes. And because the tool requires him to establish the grading rubric upfront, he says, “I can tell that my grading is fairer as a result.”
The AI features in Gradescope “are not profoundly deep at this point,” he notes. The tool recognizes handwriting and can cluster answers, but it does not actually learn from previous patterns.
Still, he appreciates its benefits, even if some of his students have been a bit uneasy about receiving their grades from an automated system. For some of them, he’s had to provide reassurance that he, as the instructor, was really determining their grades. Artificial-intelligence, Anthony says, “enables me to be more personal, but I can see how students would perceive it as more impersonal.”
“They want the personal touch.”
Goldie Blumenstyk writes about the intersection of business and higher education. Check out www.goldieblumenstyk.com for information on her book about the higher-education crisis; follow her on Twitter @GoldieStandard; or email her at goldie@chronicle.com.
