{"id":3774,"date":"2026-11-02T15:02:57","date_gmt":"2026-11-02T15:02:57","guid":{"rendered":"https:\/\/tcjsj.com\/?p=3774"},"modified":"2026-11-02T15:02:57","modified_gmt":"2026-11-02T15:02:57","slug":"beyond-the-click-what-a-major-study-on-khanmigo-reveals-about-motivation-engagement-and-the-limits-of-ai-tutors-in-schools","status":"publish","type":"post","link":"https:\/\/tcjsj.com\/?p=3774","title":{"rendered":"Beyond the Click: What a Major Study on Khanmigo Reveals About Motivation, Engagement, and the Limits of AI Tutors in Schools"},"content":{"rendered":"<p>For two years, thousands of students across the United States had access to a sophisticated, patient, and highly publicized artificial intelligence tutor living just one click away on their screens. Out of the students who participated in a comprehensive randomized study across 18 Tennessee middle schools, an overwhelming 96% tried the AI tutor, known as Khanmigo, at least once. <\/p>\n<p>Yet, beneath that initial surge of curiosity lay a sobering reality: the median student ultimately used the tool on only a third of the days they actually had access to it. <\/p>\n<p>That finding serves as the core insight of &quot;One Click Away,&quot; a National Bureau of Economic Research working paper authored by economists Philip Oreopoulos and Nina Low. The study closely tracked the implementation of Khan Academy\u2019s Khanmigo during daily math intervention sessions. In these settings, the AI was specifically configured to act as an interactive coach rather than a shortcut machine, designed to guide students through problems by asking questions instead of simply handing over the final answers.<\/p>\n<p>When researchers compared the outcomes of these students against peers receiving traditional remedial instruction, the AI-assisted group made only modest gains in mathematics. These improvements were roughly on par with what educators would expect from standard Khan Academy practice sessions conducted without any artificial intelligence whatsoever. Most students engaged with the AI tutor infrequently, and when they did converse with it, their messages were rarely about math. In essence, while the technology was readily available, students made minimal effort to leverage it for academic benefit.<\/p>\n<p>In a subsequent blog post reviewing the study, Khan Academy CEO Sal Khan noted that this low utilization rate directly mirrored what his organization had observed and discussed publicly. In a later interview with education publication Chalkbeat, Khan remarked that the findings powerfully demonstrated &quot;how much motivation matters.&quot;<\/p>\n<p>Classroom educators are hardly surprised by this conclusion. However, the broader educational technology sector has spent substantial sums of money operating on the opposite assumption\u2014that building a capable, empathetic AI tutor and placing it directly in front of students is enough to guarantee heavy usage and improved outcomes. Major technology firms have poured resources into the space, such as Microsoft donating significant Azure infrastructure to make Khanmigo\u2019s teacher tools completely free for every educator across the United States. Concurrently, various state departments of education purchased bulk licenses for public schools. Maryland, for instance, acquired over 4,300 licenses for a two-year pilot program, while Indiana launched a similar initiative spanning more than 100 schools.<\/p>\n<h2>How Khan Academy Adjusted<\/h2>\n<p>Faced with these engagement hurdles, Khan Academy has already begun altering its strategic approach. In a July blog post, Sal Khan reflected that the AI tutor could no longer simply sit adjacent to the core educational content; instead, it needed to be deeply integrated into the fabric of the assignments. The organization realized it had to make &quot;productive struggle harder to sidestep.&quot; Earlier this year, Khan Academy shifted its software architecture to embed Khanmigo directly into practice problems, removing the barrier where students had to actively initiate a conversation.<\/p>\n<p>While this technical fix suits Khan Academy&#8217;s platform, there is a risk of misinterpreting the broader lesson as a mandate that every single student requires an AI tutor. They do not. Students have historically learned calculus, crafted powerful essays, and developed critical thinking skills entirely without artificial intelligence, and many teachers will reasonably choose to keep these tools out of their classrooms.<\/p>\n<p>If an educational institution genuinely wants artificial intelligence to support student cognition, relying on voluntary adoption clearly falls short. When AI assistance remains optional, students frequently bypass the sophisticated Socratic tutor that asks probing questions in favor of alternative chatbots in neighboring browser tabs that instantly yield direct answers. <\/p>\n<p>This behavior is not isolated. In higher education and high school classrooms where instructors introduce AI study tools on a purely optional basis, students consistently defer engagement. However, when instructors make the use of AI mandatory\u2014while actively monitoring the resulting transcripts and providing constructive feedback\u2014the landscape shifts, showing the initial signs of meaningful student interaction.<\/p>\n<p>Even so, mandating the use of a tool only solves the problem of participation, leaving the deeper challenge of effort unresolved. A student can be forced to open a chat window and converse with an AI tutor while still contributing virtually nothing of intellectual value. Kristen DiCerbo, chief learning officer at Khan Academy, noted during an EdTech Insiders panel that her team observed &quot;more passive kinds of interaction than we would like.&quot; A built-in tutor can successfully initiate conversations, but it cannot fundamentally force a student to care about their academic progress.<\/p>\n<h2>Requirement versus Assessment<\/h2>\n<p>To bridge this gap, some educators argue that the educational field is overlooking a vital step: treating the digital conversation itself as the primary form of assessment.<\/p>\n<p>This philosophy underpinned the research design of &quot;From Tutor to Test,&quot; a book chapter co-authored by Aim\u00e9e Skidmore and educational researcher and entrepreneur Ethan Danahy. The study followed 21 twelfth-grade students taking a media and communications course at the Coll\u00e8ge du L\u00e9man International School in Switzerland. Over a four-week period, the students examined transcripts of both strong and weak interactions with artificial intelligence, working collaboratively with their teacher to establish a clear rubric for effective AI engagement. <\/p>\n<p>Armed with these criteria, the students used ChatGPT to brainstorm ideas for an ongoing media analysis project, ultimately submitting their complete chat transcripts for evaluation against the rubric. Grading was based on the precision of their questioning, how critically they responded to the AI&#8217;s suggestions, and whether they successfully maintained intellectual ownership of their thinking process.<\/p>\n<p>Following the assignment, the vast majority of the students reported that their fundamental approach to artificial intelligence had shifted. Ten of the participants stated that they paid significantly closer attention to their own input and how to build upon it later in a conversation. Furthermore, the number of students who felt entirely confident in obtaining productive responses from the AI rose from just two to eleven.<\/p>\n<p>While rising student confidence can sometimes serve as an early warning sign of over-reliance, individual tracking revealed nuanced psychological shifts. Five students experienced major gains in confidence while simultaneously developing a heightened awareness of AI&#8217;s inherent risks. Four students became less confident in their ability to judge the accuracy of the AI output, which researchers interpreted as a healthy recalibration of skepticism. <\/p>\n<p>Only a small subset exhibited a concerning pattern: substantial confidence boosts paired with a dismissive attitude toward the underlying mechanics of critical thought. One student, who had never interacted with AI prior to the project, observed that students &quot;need to know how to approach AI to not lose their capacity to think.&quot;<\/p>\n<p>Although this particular pilot was small, lacked a traditional control group, and relied heavily on self-reported data, the underlying pedagogical logic remains straightforward. When a live conversation constitutes the formal test, students clearly understand what is expected of them and are internally motivated to exert genuine effort because they know an authority figure whose judgment they respect will evaluate the final output.<\/p>\n<h2>Firsthand Experience and Classroom Applications<\/h2>\n<p>Practical experimentation by educators exploring these concepts has yielded further insights. When using AI as a conventional Socratic tutor for personal graduate coursework, researchers often found the experience tedious and uninspiring, immediately validating why students tend to avoid standard AI tutors. Given the choice, many would prefer working independently or utilizing compliant shortcuts. <\/p>\n<p>However, when the AI is configured to play a specific, challenging character\u2014such as a difficult person who must be persuaded, taught, or negotiated with\u2014the dynamic changes. Knowing that an instructor will closely review and evaluate the performance in the chat raises the stakes of the interaction, compelling the user to think far more critically than they otherwise might. The underlying artificial intelligence technology remains identical; the variable that shifts engagement and effort is the awareness of external evaluation.<\/p>\n<p>Building on these insights, classroom testing has expanded into diverse subjects. In a calculus course taught by Bill Fallon at Cheshire Academy in Connecticut, students participated in a role-play exercise where they assumed the role of the teacher. The artificial intelligence portrayed &quot;Devon,&quot; a fictional classmate who genuinely struggled to understand tangent lines and refused to pretend otherwise. If students relied heavily on jargon like &quot;instantaneous rate of change,&quot; Devon grew even more confused. To succeed, the students had to comprehend the underlying mathematical concept well enough to explain it in clear, plain language. <\/p>\n<p>The resulting transcripts were evaluated first by an AI engine and subsequently by the teacher, measuring student persistence, clarity, and adherence to the teaching persona. Several students opted to retake the simulation voluntarily, and every single one of them scored higher on their second attempt, improving by an average of roughly 34 points. The primary driver for their return was clear: the transcript and AI feedback explicitly highlighted where their explanations had broken down.<\/p>\n<p>Similar results emerged in science classrooms at San Domenico School in California, where an AI configured to act as a seasoned ecologist persistently questioned students with follow-up inquiries. One student noted that the experience forced them beyond rote memorization of terminology into a genuine understanding of the underlying scientific mechanisms.<\/p>\n<p>Naturally, not every student embraced these exercises with enthusiasm. Some reported feeling interrogated, while others likened the relentless questioning to provocative internet behavior or found the process repetitive. Educators point out that such friction points often highlight necessary pedagogical tuning, better scaffolding, and the need for teachers to monitor students who feel overwhelmed rather than constructively challenged. <\/p>\n<p>As Fallon observes, life and human interactions are inherently frustrating, and learning to navigate that friction productively represents an advanced level of thought and skill development. Students rarely express deep affection for traditional examinations, yet educators do not take student complaints as proof that tests measure nothing.<\/p>\n<h2>Where This Leaves Us<\/h2>\n<p>The accumulating evidence suggests that while Sal Khan is entirely correct in asserting that student motivation is paramount, the structural design of learning environments matters immensely. Schools retain the absolute right to determine that artificial intelligence has no place within their classrooms. However, in environments where educators actively choose to use AI to support critical thinking, keeping adoption strictly optional fails because students naturally gravitate toward tools that perform the thinking on their behalf. <\/p>\n<p>At the same time, merely mandating the use of AI is insufficient. Students exert real effort when they understand what they are expected to demonstrate and know that a respected human evaluator will examine their work\u2014a principle that governed effective assessment long before the advent of generative artificial intelligence.<\/p>\n<p>Much of the early vision surrounding AI tutoring relied heavily on autonomous, self-directed learning: a student, internally motivated, sitting alone with a patient machine. The emerging alternative places the human teacher back at the center of the educational ecosystem. Automated systems generate detailed chat transcripts that capture real-time reasoning, tracking where students stumble and how they recover. While AI engines can draft preliminary evaluations against predefined criteria, the human teacher retains final authority, adding nuanced notes, correcting automated drafts, and determining the appropriate feedback loop for the student.<\/p>\n<p>Rather than trying to build cheat-proof systems that block students from avoiding mental labor, educators are finding greater success in designing assignments where student reasoning is fully visible. When an assessment takes the form of a live, adaptive conversation judged on the quality of a student&#8217;s argumentation, there is far less intellectual work to outsource, and the transcript provides a clear record of whose mind was at work. <\/p>\n<p>Ultimately, artificial intelligence makes it possible to construct assessments that mirror genuine real-world scenarios\u2014whether dealing with a confused peer, a skeptical reviewer, or a reluctant colleague. By requiring students to navigate these conversations using core academic skills while providing teachers with transparent records of the process, classrooms can utilize technology far more effectively than relying on a passive tutor waiting quietly at the edge of a screen.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>For two years, thousands of students across the United States had access to a sophisticated, patient, and highly publicized artificial intelligence tutor living just one click away on their screens. Out of the students who participated in a comprehensive randomized study across 18 Tennessee middle schools, an overwhelming 96% tried the AI tutor, known as [&hellip;]<\/p>\n","protected":false},"author":23,"featured_media":3773,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[1465,1545,5517,1463,1525,5518,1464,5520,464,5519,753,184,892,5521],"class_list":["post-3774","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized","tag-academics","tag-beyond","tag-click","tag-education","tag-engagement","tag-khanmigo","tag-learning","tag-limits","tag-major","tag-motivation","tag-reveals","tag-schools","tag-study","tag-tutors"],"_links":{"self":[{"href":"https:\/\/tcjsj.com\/index.php?rest_route=\/wp\/v2\/posts\/3774","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/tcjsj.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/tcjsj.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/tcjsj.com\/index.php?rest_route=\/wp\/v2\/users\/23"}],"replies":[{"embeddable":true,"href":"https:\/\/tcjsj.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3774"}],"version-history":[{"count":0,"href":"https:\/\/tcjsj.com\/index.php?rest_route=\/wp\/v2\/posts\/3774\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/tcjsj.com\/index.php?rest_route=\/wp\/v2\/media\/3773"}],"wp:attachment":[{"href":"https:\/\/tcjsj.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3774"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/tcjsj.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3774"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/tcjsj.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3774"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}