Texas is using AI to grade student essays. Arizona is taking a different approach

Published: Thursday, April 25, 2024 - 9:49am
Audio icon Download mp3 (988.91 KB)

The Lone Star State piloted a program that uses AI technology like ChatGPT to grade the essay responses for students' state exams, and quietly announced in early April that it would be continuing the program in order to cut the cost of using human graders. 

Joseph Guzman is the associate superintendent for Assessments, Standards, Accountability and Research for the Arizona Department of Education.

Guzman said that Arizona has been using some form of computer-assisted grading for state exams since around 2012, and though the technology has evolved over the years, the Department of Education has not yet begun using generative AI to grade students' exams.

"There's a pool of human scores that's generated, and it's generated item by item," Guzman said. "Then they use the computer to gather similarities with those scores so that they can then use the computer to actually score further surveys."

In order to validate the computers' scores, Guzman said there's around a 10% rate at which essays are hand-graded by humans to make sure everything is running smoothly.

He also mentioned that the exams being graded by computers are more "summative" and not "formative."

"Summative assessments give summary assessments of a group," he said, adding that while these can break down by individual students, they're meant more as an assessment of a class's overall performance. Formative assessments, he said, are "really designed to tell you a lot more about the [individual] student."

One of the biggest criticisms of Texas using generative AI to grade essay responses is that while people may think machines are unbiased, AI systems tend to reflect the internal bias of the people who developed them.

Guzman said this is one reason that Arizona is cautiously evaluating AI as a grading tool.

"It's a technology that's here, and if we don't understand how to use it, we'll only end up with non-optimal uses," he said. "We're very concerned with generating guardrails for the use of AI. We are interested also in the potential that it offers to the future of education."

According to Guzman, the education landscape is changing rapidly, and Arizona needs to be prepared for new forms of technology like AI rather than refusing to use it or understand it.

"We are very interested in finding constructive, useful, boundaries for AI so that we can realize the potential that it offers to our students," he said. "We don't want to misstep and set ourselves back. We want to step carefully forward."

Education