From: redpointai

Khan Academy, an influential educational platform serving over 150 million learners across 190 countries, has made significant strides in developing and deploying AI powered tutoring tools [00:00:26]. Salman Khan, the founder and CEO of Khan Academy, highlights that schools may be among the first places to see mainstream adoption of AI for productivity [00:00:05].

Khan Migo: An AI Assistant

Khan Academy’s primary AI powered tutoring assistant is called Khan Migo [00:00:38]. It has already been deployed to over 1.4 million students and teachers [00:00:40]. The core mission of Khan Academy is to provide free world-class education for anyone, anywhere, and Khan Migo aims to approximate the personalization a good human tutor would offer [00:05:08].

Khan Migo is designed to function as both a tutor and a teaching assistant [00:06:30]. Key features and considerations include:

  • Guardrails Teachers can monitor student activity [00:06:34].
  • Anti-Cheating Measures It’s designed to prevent cheating [00:06:38].
  • Safety and Privacy Prioritizes user safety and privacy [00:06:40].
  • Socratic Method Emphasizes good pedagogy by being more Socratic [00:06:44].

While initially expecting around 100,000 users by 2025, Khan Migo quickly surpassed 1.3 to 1.4 million users, with districts paying around $15 per year per student to cover compute costs, support, and training [00:06:55].

Impact on Students

While highly motivated students can learn a lot from tools like Chat GPT through self-prompting [00:09:21], Khan believes that for most students, the AI is not yet ready to drive learning entirely on its own [00:09:12]. Khan Migo acts as a support, engaging students and reinforcing traditional practice on Khan Academy [00:09:57].

Engagement and Accuracy

A primary challenge is engagement, as efficacy often follows engagement [00:10:17]. Khan Migo aims to improve student engagement.

Regarding accuracy, Khan Migo, when anchored to Khan Academy content, boasts an error rate of about 2%, split between math errors and evaluation errors (e.g., incorrect feedback on partially correct answers) [00:12:57]. This is considered better than many human tutors [00:13:45].

Surprising Student Uses

Students have found unexpected ways to use the AI:

  • Literary Character Simulations Students engage in lengthy conversations with AI simulations of literary characters (e.g., Jay Gatsby) [00:11:52]. These simulations are designed to drive conversation, asking students about their thoughts and personal connections [00:12:05].
  • Concept Exploration Students, even in advanced math, use the AI to explore ideas [00:12:35]. They recognize the AI isn’t always right, similar to human tutors, but it’s a valuable tool for learning [00:12:38].

Impact on Teachers

Teachers are leveraging AI powered tutoring tools to streamline various aspects of their work, including planning, delivery, and gaining insights into student progress [00:14:52].

  • Lesson Planning AI assists teachers in creating more entertaining and appropriately sized lesson plans [00:14:18].
  • Question Generation Through a partnership with Blooket, Khan Migo can generate game-based questions in about two minutes, a task that previously took teachers 30-60 minutes [00:14:34].
  • Student Insights Teachers can use the AI to get insights into where students are at any given time [00:03:33].
  • Writing Coach Khan Academy developed “Writing Coach,” an AI tool that helps students with writing assignments. It acts as an ethical coach, provides brainstorming and outlining tools, and shows the teacher the student’s entire process, not just the final output. This helps combat cheating, including from other AI tools [00:15:41].

Teachers have been observed using AI simulations of historical figures (e.g., Harriet Tubman, George Washington) to engage students in class [00:15:25]. School districts are reporting that AI saves teachers at least 5 hours per week [00:21:03] and are using it as a recruiting and retention tool [00:21:06].

Future of AI in Education

The future of education will see classrooms resembling current “great classrooms” where students are actively engaged, problem-solving, and interacting [00:02:12]. AI is expected to provide teachers with more time by assisting with lesson planning, grading, and progress reports [00:02:50].

Proactive AI and Immersive Learning

The next phase of Khan Migo involves making the AI much more proactive [00:07:34]. Instead of just waiting for questions, the AI will greet students, remind them of tasks, and offer assistance [00:08:00]. This addresses the “blank screen problem” where users don’t know what to prompt [00:08:19].

Khan Academy plans to pilot “Khan Academy Classroom,” featuring a proactive AI that not only engages students directly but also informs teachers when a student needs help (e.g., with the distributive property), allowing the teacher to assign an AI tutoring session as an assignment [00:34:05]. This teacher-in-the-loop approach is considered essential for engagement [00:34:45].

Looking further ahead, within 20 years, a significant aspect of learning will involve virtual reality (VR) and augmented reality (AR) becoming mainstream [00:04:15]. Combined with generative AI, students will be able to immersively engage in simulations, virtual worlds, and historical periods, akin to a “magic school bus ride” [00:04:26].

Broader Adoption

Khan believes that schools will be among the first places to adopt AI mainstream for productivity and learning [00:19:12]. The cost-effectiveness of AI tutoring (15 a year) is dramatically cheaper than traditional live tutoring (50 an hour) [00:20:18], making it highly attractive for districts seeking to improve learning outcomes.

Challenges and Development

Building effective AI powered tutoring tools requires significant effort beyond just integrating with core AI models [00:32:32]. Khan Academy has focused on:

  • Safety and Moderation Initially conservative, they’ve refined moderation processes to reduce false positives [00:21:48].
  • Math Accuracy Extensive work has gone into reducing math and evaluation errors to near zero, including addressing subtle issues like correctly evaluating numerical precision (e.g., 1/3 vs. 0.33) [00:22:07].
  • User Interface Making the user interface intuitive and natural, moving beyond a simple chatbot [00:22:35].
  • Proactive AI Redesign Re-engineering the front-end of Khan Academy for an AI-first approach [00:22:41].
  • Context and Memory Integrating AI into every aspect of the learning experience, such as the Writing Coach which maintains context throughout brainstorming, outlining, and drafting [00:22:54]. Enhancing memory capabilities for models is crucial for providing more contextual support [00:23:58].

Future capabilities include advanced voice integration [00:24:26], and the ability for the AI to “see” and provide feedback on a student’s written work on a tablet [00:25:33]. While AI video creation tools and image generation are exciting, good pedagogical use cases are still being explored [00:25:08].

Broader Market and Global Impact

The AI in education market currently has “a lot of noise,” with many startups offering “thin prompting layers” over base AI models [00:38:25]. Khan Academy, as a nonprofit, can take a longer-term view on development and focus on pedagogy and efficacy, building trust within the education community [00:39:06].

Globally, while some highly motivated students can thrive with direct access to AI tools, structured content and guidance are generally necessary [00:36:07]. Khan Academy aims to offer high school and eventually college credits and diplomas, making AI-supported education compelling for students in underserved regions [00:37:39].

The skills needed for the future workforce will continue to emphasize strong critical thinking, reading, writing, and math abilities [00:42:57]. However, the ability to creatively combine existing resources and leverage AI as an “accelerant” – akin to entrepreneurship – will become increasingly essential in almost any career [00:43:28]. This includes tasks like using AI to quickly generate first drafts of speeches or other written content [00:46:15]. Khan also notes a desire to explore AI for code generation for app development [00:46:33].