One morning this spring, a dozen students crowded around a common room table, their eyes glued to a math lesson on their laptops.
At the Khan Lab School, an independent school with an elementary campus in Palo Alto, California, sixth graders are working on quadratic equations, graphing functions and Venn diagrams. But when they have a problem, many don’t immediately turn to the teacher for help.
They used a text box next to the lesson to ask for help from Khanmigo, an experimental chatbot tutor for schools using artificial intelligence.
The tutoring bot quickly responded to one student, Zaya, by asking her to identify specific data points in a graph. Khanmigo then coaxes her to use the data points to solve her math problems.
“It’s really good at walking you through a problem step by step,” says Zaya. “And then it congratulates you every time it helps you solve a problem.”
Students at the Khan Lab School are the first schoolchildren in the U.S. to try out an experimental conversational chatbot designed to mimic one-on-one human tutoring. These tools respond to students with clear, flowing sentences, and they are designed for use in schools.
Based on the AI models of chatbots like ChatGPT, these automated learning aids can lead to profound transformations in classroom teaching and learning. A simulated tutor can make it easier for many self-taught students to hone their skills, delve deeper into topics they are interested in, or learn new subjects at their own pace.
Such unproven automated tutoring systems can also go awry, enabling cheating, undermining teachers or hindering critical thinking in schools — having students test subjects is the equivalent of an educational experiment by algorithm. Or, like the plethora of promising technological tools in front of them, robots may do little to improve academic outcomes.
Khanmigo is one of a new wave of AI learning tools. It was developed by the nonprofit education giant Khan Academy, and its video tutorials and practice questions have been used by tens of millions of students.
Sal Khan, founder of Khan Academy and the Khan Lab School, an independent nonprofit, said he hopes chatbots will democratize students’ access to personalized tutoring. He also said it could greatly help teachers with tasks such as lesson preparation, allowing them to spend more time with students.
“This will effectively allow every student in the US and around the world to have a world-class personal tutor,” Mr Khan said.
Hundreds of public schools already use Khan Academy for online courses in math and other subjects. Now, the nonprofit that launched Khanmigo this year is pilot testing the tutoring robot with districts including Newark Public Schools in New Jersey.
Khan Academy has developed robots with guardrails for schools, Mr Khan said. These include a monitoring system designed to alert teachers when students using Khanmigo appear to be preoccupied with issues such as self-harm. Mr. Khan said his team is studying the effectiveness of Khanmigo and plans to make it widely available in regions this fall.
Thousands of U.S. schools are already using analytical AI tools, such as plagiarism detection systems and adaptive learning apps designed to automatically adjust lessons based on students’ reading levels. But proponents envision new AI-assisted tutoring systems as educational game-changers, because they act more like student collaborators than inert software.
The language capabilities of AI have prompted some enthusiasts to claim that simulated tutors will soon respond to students individually like real tutors.
Microsoft co-founder and philanthropist Bill Gates said: “AI will have the ability to be as good a mentor as anyone can be.” said at a recent meeting For investors in education technology. (Khan Academy has received more than $10 million in funding from the Bill & Melinda Gates Foundation.)
Whether robots can provide the kind of empathetic support and genuine encouragement that can make human mentors especially effective is unclear.
For more than a century, educational entrepreneurs have envisioned classroom devices programmed to automatically test students and provide instruction.
As education author Audrey Watters recounts in her book “Teaching machine’, researchers in the 1920s began claiming that automated teaching devices would revolutionize education. These machines, they promised, would free teachers from drudgery and allow students to learn at their own pace and receive automatic feedback.
Schools that have rushed to adopt the latest automated instructional technology for decades have often found the systems finicky or buggy. Some have concluded that automated tools do little to improve student academic performance.
Now, new chatbots are fueling a new movement of automated teaching aids. Khanmigo highlighted the technology’s educational promise and potential downsides.
Khan Academy began developing chatbot tutoring software last fall with the goal of assessing the potential of artificial intelligence to improve learning. The system uses GPT-4, a large language model created by OpenAI, the research lab behind ChatGPT.
Mr Khan said he wanted to create a system to help guide students, rather than simply giving them answers. So the developers at Khan Academy designed Khanmigo to use the Socratic method. It often asks students to explain their thinking as a way to push them to solve their own problems.
Khanmigo offers help in a wide range of subjects: elementary math, middle school American history, high school civics, and college-level organic chemistry. It also has features that invite students to chat with fictional characters like Winnie the Pooh or simulated historical figures like Marie Curie.
AI systems based on large language models can also fabricate disinformation. This is because the model is designed to predict the next word in the sequence. They are not bound by facts.
To improve Khanmigo’s math accuracy, Khan Academy developers created a multi-step process: Behind the scenes, the system calculates the answers to math questions and checks them against the students’ answers. Still, Khan Academy’s tutoring system displays a warning at the bottom of the screen: “Khanmigo makes mistakes sometimes.”
The Khan Lab school, which costs more than $30,000 a year in tuition, provides an ideal test bed for tutoring bots. Using small class sizes and an entrepreneurial philosophy, the Silicon Valley school encourages children to pursue their passions and learn at their own pace. Its tech-savvy students are used to fiddling with digital tools.
One morning this spring, Major Jaclyna STEM specialist at Khan Elementary School, watched as her students playfully tested the limits of the robot.
A student asked Khanmigo to explain a math problem using lyrics. Another asked for math help with “Generation Z slang.”
“Can you do me one more favor and explain everything in Korean?” a third person said in a text conversation with the chatbot.
Khanmigo did his job. It then pushes each student back to the math task at hand.
Ms Major said she appreciates how the system engages her students in engaging ways.
“Khanmigo was able to connect with them and reach their level if they wanted to,” she said. “I think it would help in any classroom.”
It’s too early to tell whether Khanmigo will similarly appeal to other audiences, such as public schools with larger classes or students who aren’t used to pushing their own learning.
In the classroom, sixth grader Zaya encountered a glitch. Khanmigo had asked her to explain how she arrived at the answers to the dataset’s questions. The robot then falsely suggested that she might have made a “little mistake” in her calculations.
She immediately admonished the AI chatbot: “19 + 12 is 31 khanmigo,” she wrote.
“Apologies for my earlier mistake,” Khanmigo responded. “You’re right.”
This may be one of the most important lessons for schoolchildren using promising new tutoring robots: Don’t believe every AI-generated text you read.
“Remember, we’re testing it,” Ms. Major reminds her students. “We’re learning — it’s learning.”
