AI in education

From: redpointai

AI is rapidly transforming education, with schools potentially being one of the first sectors to see mainstream adoption for productivity and learning [00:00:05]. Khan Academy, an influential education platform serving over 150 million learners in 190 countries, is at the forefront of this shift, particularly with their AI-powered tutoring assistant, Khanmigo [00:00:26]. Khanmigo has already been deployed to over 1.4 million students and teachers [00:00:40].

The Future Classroom with AI

In 20 years, classrooms are envisioned to resemble the best classrooms of today, where students are actively engaged, interacting, and participating in problem-solving and group work, rather than passively listening to lectures [00:02:12]. Teachers will continue to walk around, observe, and guide students [00:02:33].

AI is expected to play a crucial role by:

• Giving Teachers Time Back AI can streamline tasks like lesson planning, grading, and writing progress reports [00:03:18].
• Providing Insights Offering teachers better understanding of student progress and needs [00:03:33].
• Enhancing Interactivity Facilitating more dynamic interactions within the classroom [00:03:37].
• Creating Immersive Experiences With virtual and augmented reality becoming mainstream alongside generative AI, students could immersively enter simulations, virtual worlds, or even historical settings, similar to a “magic school bus ride” [00:04:10]. This includes simulations with AI characters like historical figures for engaging discussions [00:15:25].

Khan Academy’s Approach to AI

Khan Academy’s core mission has always been to approximate the personalized learning experience of a good tutor [00:05:36]. With the advent of advanced models like GPT-4, they saw the potential to significantly enhance tutoring and teaching assistance [00:06:08].

Khanmigo: An AI Assistant

Khanmigo functions as both an AI tutor and a teaching assistant [00:06:30]. Key features include:

• Guardrails Ensures teachers can monitor student activity, prevents cheating, and prioritizes safety and privacy [00:06:34].
• Socratic Method Focuses on good pedagogy, encouraging students to explain their reasoning [00:06:44].
• High Adoption Surpassed pilot expectations, reaching 1.3-1.4 million users across districts paying $15 per year per student [00:06:55].
• Proactive AI The next phase, “Khan Academy Classroom,” will feature a more proactive AI that offers guidance and suggestions to students and teachers upon login, acting as a concierge [00:07:33]. This addresses the “blank screen problem” where users might not know how to prompt the AI [00:08:18].

Student and Teacher Use Cases

• Student Engagement While self-motivated students can learn extensively with AI tools like ChatGPT, most students require more structure [00:09:16]. Khanmigo’s role is to support and drive engagement with traditional Khan Academy practice [00:09:57]. Unexpectedly, students have been observed explaining their reasoning to the AI, leading to mutual learning and correction [00:11:12].
• Teacher Productivity Teachers use AI to tweak lesson plans, make them more entertaining, and right-size them for the classroom [00:14:16]. A partnership with Blokit allows Khanmigo to generate game-based questions in about two minutes, a task that previously took teachers half an hour to an hour [00:14:34].
• Writing Coach Khan Academy developed a “writing coach” tool to address cheating fears [00:15:39]. Teachers create assignments with the AI, and students complete them with the AI acting as an ethical coach. The teacher receives not just the final output but the entire process, allowing them to see student engagement and deter copying from other sources [00:15:46]. This approach aims to teach students to use AI responsibly, developing their own thinking skills [00:16:26].

Challenges and Solutions in AI Development

Developing AI for education involves significant challenges beyond just integrating models:

• Safety and Moderation Essential for a student audience, requiring careful implementation and tuning to avoid false positives [00:21:48].
• Math Accuracy Achieving low error rates in math is particularly challenging [00:22:07]. Khanmigo, when anchored on Khan Academy content, has an error rate of about 2%, split between straight math errors and evaluation errors (e.g., accepting 0.33 instead of 1/3) [00:13:02]. This is already considered better than many human tutors [00:13:45].
• User Interface and Experience Making the AI feel natural and integrating it seamlessly into the learning experience, rather than just being a chatbot, requires significant re-engineering [00:22:35].
• Memory and Context Future advancements in AI memory will allow models to retain context from previous interactions, providing more personalized and effective support [00:23:58].
• Multimodal Capabilities The ability for AI to “see” and understand student work (e.g., written notes on a tablet) and provide real-time feedback is a significant upcoming capability [00:25:33].
• Robust Evaluation Khan Academy uses a framework including hundreds of “tough test cases” where models historically struggled, machine labeling, and human labeling of conversations to assess model performance and identify areas for improvement [00:29:12].

Policy and District-Level Adoption

School districts are showing strong momentum in adopting AI, often at a faster pace than other sectors, driven by the clear benefits for teachers and students [00:19:12]. The cost-effectiveness of AI compared to traditional tutoring ($10-15ayearvs.$25-50 an hour) makes it highly compelling for remediation efforts [00:20:18]. Districts report AI saving teachers at least 5 hours per week, also serving as a recruiting and retention tool [00:21:03].

The most impactful AI use cases involve the teacher in the loop. For example, if AI identifies a student struggling with a concept, it can alert the teacher, who can then assign an AI tutoring session as a mandatory assignment, ensuring student engagement [00:34:03].

Broader Impact and Future of Work

While the “blank slate” problem means simply giving students AI access might not be universally effective, structured platforms like Khan Academy, integrated with AI, can enable significant self-directed learning [00:36:09]. The goal is to eventually offer high school and college credits/diplomas via these platforms, making high-quality education internationally accessible [00:37:39].

Regarding future skills, while foundational skills like writing, reading, and math remain crucial, the ability to act as an “entrepreneur” – combining existing resources and AI tools in novel ways to create value – will become increasingly important across nearly all careers [00:43:26]. This involves using AI as a “massive accelerant” to get 80% of a task done, with human skills refining the output [00:18:28].

Observations on the EdTech Market

The AI in EdTech market currently features much “noise,” with many startups offering thin prompting layers over base models [00:38:23]. This environment, while allowing faster app creation, also means low barriers to entry and high competition [00:38:46]. Nonprofits like Khan Academy have an advantage due to the ability to take a longer-term view and built trust with the education community [00:39:24].

Areas ripe for AI innovation include:

• Interviewing and Assessment Streamlining the resource-intensive and often inefficient hiring process, potentially with AI certifications reducing redundant evaluations across companies [00:40:33].
• Corporate Training Creating more engaging and interactive simulations for compliance and development training [00:42:11].

Personal AI Usage and Future Hopes

The speaker uses AI for preparing videos by asking “dumb questions” that might be overlooked, using it to rapidly clarify concepts [00:45:24]. AI also helps generate fun images for videos and drafts formal speeches from verbal recordings [00:46:00]. A personal wish is to try “vibe coding” – prompting AI to create entire apps [00:46:38].

The biggest surprise in building AI features has been the sheer amount of work required to make solutions truly effective and robust at scale, contrary to initial impressions of models being “magical” out-of-the-box [00:47:28]. However, the internal team’s rapid pivot to an “AI-first organization” mindset and the education community’s faster-than-expected adoption have been positive surprises [00:48:20].

Learn More: To experience Khan Academy’s AI work, users can visit Khan Academy to try Khanmigo and the writing coach [00:50:00]. Additionally, Schoolhouse.world, a sister nonprofit, offers free tutoring through volunteers and plans to launch “dialogues initiatives” for conversations on tough topics [00:50:07].