Khan Academy founder Sal Khan recently acknowledged that the initial launch of the platform's artificial intelligence tutor, Khanmigo, did not improve student learning as much as hoped. This admission marks a shift in educational technology from hype to realistic design. As schools integrate automated tools, new research highlights why early educational chatbots struggle and how developers are restructuring them to keep from overwhelming students.
What Happened
In a recent public update, Sal Khan addressed rumors that AI tutoring had failed, clarifying that while the first iteration of Khanmigo three years ago underperformed expectations, the platform is actively evolving. The core issue was that the AI functioned as a standalone assistant next to the math problems, requiring students to seek out help and formulate their own questions. To address this, Khan Academy is redesigning Khanmigo to prompt students directly, forcing them to explain their mathematical reasoning instead of simply giving them answers.
Despite these initial setbacks with AI, Khan Academy's core, human-designed curriculum continues to show positive results. For example, a three-year study of 6,700 students in the Newark Public Schools showed that students using the platform gained an average of six additional points on state assessments, which is nearly triple the state average. This study, which featured a high percentage of students qualifying for free- or reduced-price lunch, showed the largest academic gains among the lowest-performing students. Newark was also recently honored for its effective implementation of the district's newer AI tools, demonstrating a commitment to scaling these programs.
The Bigger Picture
The struggle to make AI tutors effective is not unique to Khan Academy. A systematic review published in BMC Medical Education revealed that students using early generative AI tutors experienced a significantly higher "extraneous cognitive load." This means the mental effort required to navigate the AI interface actually distracted them from learning the actual material. The study also found that the academic benefits of these standalone AI tools were often marginal compared to traditional, expert human instruction.
When digital tools are highly structured, however, the results are much stronger. A study of tenth-grade students published in the Psychology and Education Journal found that students using Khan Academy in a blended learning environment experienced substantial improvements in mathematical problem-solving compared to those receiving traditional instruction. Similarly, a study on seventh-grade math performance in the Randwick International Journal demonstrated significant academic gains during a structured, self-paced five-week program. As we previously reported, seventh-grade math marks a difficult transition to abstract algebra, making structured support critical during independent study.
What This Means for Families
For parents and educators, these developments show that simply giving a child access to an AI chat box is unlikely to boost their grades. Unstructured AI often leads to "cognitive offloading," where students ask the chatbot for the answer rather than doing the hard work of learning.
To be effective, AI must provide "cognitive onloading" by prompting students to explain their steps and pushing back on logical fallacies. Furthermore, digital tools cannot replace teachers. Instead, they should act as an on-the-ground coach to catch minor mistakes, allowing teachers to focus on targeted, small-group instruction.
What You Can Do
- Look for integrated guidance. Avoid standalone homework-help chatbots that simply output answers. Opt for platforms where the AI is built directly into the exercises and prompts students to explain their thinking.
- Watch for cognitive fatigue. If your child seems frustrated or spent after using an AI learning tool, they may be experiencing extraneous cognitive load. Keep digital sessions short and highly focused.
- Combine AI with human check-ins. Use automated tools for immediate feedback during practice, but rely on teachers or parent check-ins to handle deeper conceptual misunderstandings and emotional motivation.