Skip to main content
Tech

ChatGPT may harm learning outcomes: Study

Share

AI has become more common in education, and a new study asks the question, does it help students learn? The study’s authors conducted an experiment with high school students, and the findings suggest the answer is not simple.

Researchers at the University of Pennsylvania tested Turkish high school students using ChatGPT on math practice problems. Students who had access to the tool during practice solved 48% more math problems correctly, but scored 17% lower on the subsequent test compared to those without it.

QR code for SAN app download

Download the SAN app today to stay up-to-date with Unbiased. Straight Facts™.

Point phone camera here

A third group used a modified ChatGPT, which acts at a tutor, offering hints instead of direct answers. They did much better on practice problems and were able to solve 127% more of them. But these students still did not perform any better on tests than normal.

The study suggests students rely too much on AI as a crutch, asking for answers instead of solving problems themselves. ChatGPT’s inaccuracies didn’t help either, as the tool was wrong about 42% of the time in its problem-solving approach.

Despite this, students believe ChatGPT hasn’t hurt their learning. The researchers compared this to pilots overly relying on autopilot, warning that too much dependence on AI could undermine learning.

Tags: , , , ,

Lauren Taylor

AS AI BECOMES MORE COMMON IN EDUCATION, A NEW STUDY ASKS: DOES IT REALLY HELP STUDENTS LEARN? 

 

A RECENT EXPERIMENT WITH HIGH SCHOOL STUDENTS SUGGESTS THE ANSWER MAY NOT BE SIMPLE.

RESEARCHERS AT THE UNIVERSITY OF PENNSYLVANIA, TESTED TURKISH HIGH SCHOOL STUDENTS USING CHATGPT FOR MATH PRACTICE PROBLEMS. 

 

STUDENTS WHO HAD ACCESS TO CHAT GPT DURING PRACTICE SOLVED  48-PERCENT MORE MATH PROBLEMS CORRECTLY, BUT SCORED 17-PERCENT LOWER ON THE TEST COMPARED TO THOSE WITHOUT IT.

 

A THIRD GROUP USED A MODIFIED CHATGPT ACTING AS A TUTOR, OFFERING HINTS INSTEAD OF DIRECT ANSWERS. THEY DID MUCH BETTER ON PRACTICE PROBLEMS, SOLVING 127-PERCENT MORE, BUT DIDN’T PERFORM ANY BETTER ON THE TEST.

 

THE STUDY SUGGESTS STUDENTS RELY TOO MUCH ON AI AS A CRUTCH, ASKING FOR ANSWERS INSTEAD OF SOLVING PROBLEMS THEMSELVES. 

 

CHATGPT’S INACCURACIES DIDN’T HELP EITHER, AS IT WAS WRONG ABOUT 42-PERCENT OF THE TIME IN ITS PROBLEM-SOLVING APPROACH.

 

DESPITE THIS, STUDENTS BELIEVE CHATGPT HADN’T HURT THEIR LEARNING. THE RESEARCHERS COMPARED THIS TO PILOTS OVERLY RELYING ON AUTOPILOT, WARNING THAT TOO MUCH DEPENDENCE ON AI COULD UNDERMINE LEARNING.

 

I’M LAUREN TAYLOR…FOR MORE OF OUR STORIES — DOWNLOAD THE STRAIGHT ARROW NEWS APP AND VISIT SAN-DOT-COM.