During a spring morning, a group of twelve students gathered around communal classroom tables, their attention focused on math lessons displayed on their laptops.
These sixth graders, attending the elementary campus of Khan Lab School in Palo Alto, California, were engrossed in activities involving quadratic equations, graphing functions, and Venn diagrams. However, instead of immediately seeking assistance from their teacher when faced with questions, many turned to Khanmigo, an experimental chatbot tutor for schools utilizing artificial intelligence.
With the aid of a text box integrated into their lessons, the students requested help from Khanmigo, which promptly responded. For instance, when Zaya sought assistance, the tutoring bot directed her to identify specific data points in a chart and then guided her in employing those points to solve her math problem.
Zaya praised Khanmigo, stating, "It's highly effective in providing a step-by-step walkthrough of the problem. Furthermore, it offers words of encouragement each time it assists in problem-solving."
As some of the first students in the United States to experience these experimental conversational chatbots, Khan Lab School students are at the forefront of a potential revolution in classroom teaching and learning. These simulated tutors, powered by AI models like ChatGPT, have the capability to transform the educational landscape. They can facilitate the skills development of self-directed learners, enable deeper exploration of subjects that capture students' interest, and allow for personalized learning at one's own pace.
While these innovative automated tutoring systems hold great promise, there are concerns about potential pitfalls. They might introduce errors, foster cheating, diminish the role of teachers, or impede critical thinking in schools. In essence, students could become subjects in an educational experiment driven by algorithms. Alternatively, like numerous previous technology tools, these bots may fail to significantly enhance academic outcomes.
Khanmigo represents one of the latest AI-powered learning tools, developed by Khan Academy, a nonprofit education organization renowned for its video tutorials and practice problems utilized by millions of students. Sal Khan, the founder of Khan Academy and Khan Lab School, envisions this chatbot as a means to democratize individualized tutoring and alleviate teachers' burden by aiding with tasks such as lesson planning, thereby allowing them to allocate more time to students.
While Khan Academy's online lessons are already utilized by hundreds of public schools across the United States, the nonprofit organization is now pilot-testing Khanmigo with districts like Newark Public Schools in New Jersey. The development of the bot incorporates safeguards for schools, including a monitoring system that alerts teachers if students using Khanmigo exhibit signs of self-harm. Khan Academy is actively studying the effectiveness of Khanmigo and plans to make it widely available to districts in the upcoming fall.
Analytical AI tools such as plagiarism-detection systems and adaptive learning apps, already employed in thousands of U.S. schools, have laid the foundation for the potential game-changing nature of AI-assisted tutoring systems. Unlike inert software, these tutoring bots function as student collaborators, responding to inquiries with clear and articulate sentences. Enthusiasts even speculate that, with their language proficiency, simulated tutors may soon match the responsiveness of human tutors on an individual level.
Bill Gates, the Microsoft co-founder and philanthropist, believes AI will reach a level where it can become as proficient as any human tutor. Speaking at a recent conference for educational technology investors, he expressed his confidence in AI's ability to achieve that milestone. (Khan Academy has received substantial grants from the Bill & Melinda Gates Foundation, totaling over $10 million.)
Nevertheless, it remains uncertain whether these bots can replicate the empathetic support and genuine encouragement that make human tutors exceptionally effective.
The idea of using automated teaching devices in classrooms is not new. As education writer Audrey Watters recounts in her book "Teaching Machines," researchers in the Sure, continuing from where we left off:
1920s, there were already claims that automated teaching devices would revolutionize education by freeing teachers from monotonous tasks and enabling students to work at their own pace with automated feedback. However, over the years, schools that eagerly adopted these automated teaching technologies often found them to be unreliable or ineffective, leading to little improvement in student outcomes.
Now, a new generation of chatbots is reigniting the push for automated teaching aids. Khanmigo exemplifies both the educational potential and possible drawbacks of this technology. Khan Academy began developing chatbot tutoring software last fall, aiming to assess the potential of AI to enhance learning. The system employs GPT-4, a powerful language model created by OpenAI, the research lab behind ChatGPT.
Rather than providing answers outright, Khan Academy's goal was to create a system that guides students by using the Socratic method. Khanmigo frequently prompts students to explain their reasoning, encouraging them to find solutions to their own questions.
Khanmigo covers a wide range of subjects, including elementary school math, middle school American history, high school civics, and college-level organic chemistry. It also offers interactive features that allow students to engage in conversations with fictional characters like Winnie-the-Pooh or simulated historical figures such as Marie Curie.
While AI-based systems like large language models have the capacity to generate false information, Khan Academy took measures to improve Khanmigo's accuracy in math. They implemented a multi-step process where the system independently solves math problems behind the scenes and compares its answer with the student's response. However, even with these precautions, the tutoring system includes a warning at the bottom of the screen: "Khanmigo makes mistakes sometimes."
Khan Lab School, with its annual tuition exceeding $30,000, provides an ideal testing ground for tutoring bots. The school's entrepreneurial philosophy and small class sizes encourage students to pursue their passions and take charge of their own learning. Tech-savvy students at Khan Lab School are accustomed to experimenting with digital tools.
One morning, Jaclyn Major, a STEM specialist at Khan's elementary school, observed as her students playfully pushed the boundaries of the chatbot's capabilities. Some students requested math explanations using song lyrics, while others sought math help in "Gen Z slang." One student even asked the bot to explain everything in Korean, to which Khanmigo dutifully complied before nudging the students back to their math tasks.
Ms. Major expressed appreciation for how Khanmigo engaged with her students in a captivating manner, stating, "Khanmigo connects with them on their level, should they desire it. I believe it could be beneficial in any classroom setting."
However, it is still too early to determine whether Khanmigo will have the same level of engagement in other contexts, such as public schools with larger class sizes or students who are not accustomed to self-directed learning.
Inside the classroom, Zaya, the sixth-grade student, encountered a glitch. Khanmigo had asked her to explain her reasoning for solving a data set problem, but then incorrectly suggested she had made a "small mistake" in her calculations. Zaya promptly corrected the chatbot, stating, "19 + 12 is 31, Khanmigo."
Apologizing for the earlier mistake, Khanmigo responded, "You are indeed correct."
This episode may serve as a valuable lesson for students using these promising new tutor bots: not to blindly trust every text generated by AI.
Ms. Major reminded her students that they were in the testing phase and that both they and Khanmigo were learning together.