AI models and educational engagement

From: redpointai

AI is expected to be one of the first areas to see mainstream adoption for productivity and learning, transforming education for students, teachers, and school systems alike [00:00:05], [00:00:46]. A primary focus of AI and education technology development, particularly for organizations like Khan Academy, is to enhance student and teacher engagement [00:10:00].

Enhancing Classroom Engagement with AI

Ideally, a great classroom is characterized by active student engagement, interaction, and hands-on activities, rather than passive listening [00:02:18]. Teachers in such environments walk around, working with students to facilitate interesting activities [00:02:33]. AI can empower more teachers by:

• Freeing up time from administrative tasks like lesson planning and grading [00:03:18].
• Providing insights into student progress and classroom management [00:03:31].
• Facilitating better interactivity with students [00:03:37].

In the future, classrooms could incorporate ambient AI that observes and provides insights without requiring constant screen interaction [00:03:50]. Additionally, advancements in virtual reality (VR) and augmented reality (AR), combined with generative AI, could enable immersive simulations, allowing students to experience virtual worlds, such as “literally like a magic school bus ride” back to ancient Rome [00:00:20], [00:04:15].

Khan Academy’s Approach to Engagement

Khan Academy’s core mission is to personalize education, replicating the effectiveness of a good tutor [00:05:10]. This long-standing goal has been significantly advanced by generative AI [00:06:06].

Khanmigo: AI-Powered Tutoring

KIGO (Khanmigo), Khan Academy’s AI-powered tutoring assistant, serves both students and teachers [00:00:38], [00:06:30]. It has been deployed to over 1.4 million students and teachers [00:00:40], with districts paying around $15 per year for its use [00:07:17].

Key features and observations:

• Proactive AI: The next phase of Khanmigo will be a more proactive AI, welcoming students and suggesting activities, acting as a “concierge” for both students and teachers [00:07:33], [00:08:00], [00:09:01].
• Addressing the “Blank Screen Problem”: Initial challenges with users not knowing what to prompt the AI are being addressed through proactive suggestions and dynamic action bubbles [00:08:19], [00:08:43].
• Integration with existing content: Khanmigo is “anchored on Khan Academy content” to improve accuracy, achieving a 2% error rate, split between math errors and evaluation errors [00:12:59]. This is considered better than many human tutors [00:13:45].
• Socratic Method: Khanmigo emphasizes a Socratic approach to learning, focusing on good pedagogy and safety [00:06:40], [00:06:44].

Student Engagement

While highly motivated students can learn a lot from tools like ChatGPT by spending time with them [00:09:21], most students require more structured engagement [00:09:30]. Key insights on student usage include:

• Efficacy vs. Engagement: The primary challenge is not efficacy (as most reasonably healthy interventions are efficacious), but engagement [00:10:09], [00:15:18].
• Student-driven corrections: Students have been observed explaining their reasoning to AI models, leading to the models iterating and correcting themselves [00:11:12].
• Literary Simulations: Students can have lengthy conversations with AI simulations of literary characters, which are designed to drive the conversation and deepen understanding [00:11:52].
• Exploring advanced concepts: Highly curious students use AI to explore complex ideas, even if the AI is not 100% accurate, accepting that even human tutors can make errors [00:12:35].

Teacher Empowerment

AI offers teachers “superpowers” by augmenting their abilities [00:04:47].

• Streamlined planning: Teachers use AI to tweak lesson plans, making them more entertaining and appropriately sized [00:14:18].
• Automated content creation: Partnerships, such as with Blkit, allow Khanmigo to generate game-based questions in minutes, a task that previously took half an hour or more [00:14:36].
• AI Simulations for class activities: Teachers leverage AI to create engaging activities, such as having students talk to AI simulations of historical figures like Harriet Tubman or George Washington [00:15:27].
• Writing Coach: The “writing coach” tool addresses cheating fears by enabling teachers to assign and manage writing assignments through AI. The AI acts as an ethical coach, and teachers can review the student’s process, not just the final output. This tool is designed to undermine all forms of cheating, not just AI-assisted cheating [00:15:41].

Broader Implications for Education and the Workforce

District-Level Adoption

Schools are observed to be early adopters of AI for productivity and learning [00:19:12].

• Cost-effectiveness: AI offers a dramatically cheaper alternative to traditional tutoring, costing $10-$15 per year compared to $25-$50 per hour for live tutoring [00:20:18].
• Teacher Retention: Districts are finding that AI saves teachers at least 5 hours a week, making it a powerful recruiting and retention tool [00:21:03].
• Accountability: AI can help “human systems” hold students accountable, for instance, by teachers assigning AI-led interventions for specific learning gaps [00:10:44], [00:34:05].

Future Skills and Workforce Adaptation

The fundamental skills of critical thinking, writing, reading, and mathematics remain essential [00:43:00]. However, a key skill for the future workforce will be entrepreneurship – the ability to combine existing resources in new ways to create value [00:43:26]. This involves:

• Proactively seeking out new tools and combining them [00:44:02].
• Leveraging AI to get 80% of the way to a solution, then using human skills to refine it [00:44:13].

Organizations that cultivate this entrepreneurial mindset within their teams will be better positioned to innovate and maintain lower cost structures [00:44:34].

Challenges and Development Considerations

Building Beyond Thin Prompting Layers

Developing effective AI tools like Khanmigo requires significant effort beyond simple prompting of foundation AI models [00:21:41], [00:23:19]. This includes:

• Safety and Moderation: Implementing robust safety and moderation features, initially being overly conservative, but later refining them to avoid false positives [00:21:48].
• Math Accuracy: Extensive work is needed to achieve low error rates, especially for evaluation errors where student answers are close but not exact [00:22:07].
• User Interface (UI): Creating natural and intuitive user interfaces that integrate AI throughout the experience, rather than just a chatbot [00:22:35].
• Context and Memory: Future advancements require better AI model memory to provide more context and allow for “prompt chaining” and dynamic prompt swapping [00:23:37], [00:23:58].
• Multimodal Capabilities: Integrating advanced voice capabilities and enabling AI to “see” and provide feedback on student’s written work (e.g., on a tablet) are crucial future developments [00:24:26], [00:25:33].

Model Evaluation

Khan Academy uses a rigorous AI model evaluation and benchmarking framework:

• Tough Test Cases: A collection of several hundred “tough test cases” (e.g., specific math problems, common student misconceptions) that models have historically struggled with [00:29:16].
• Machine and Human Labeling: AI interactions are labeled by machines for potential errors, which is then verified and refined by human labeling to get accurate error rates [00:30:10]. This also involves labeling the “productivity of the conversation” to assess student engagement [00:30:50].

Strategic Considerations

• Trust and Long-Term View: Khan Academy’s non-profit status allows it to take a longer-term view on AI development, focusing on pedagogical soundness, efficacy studies, and building trust within the education community, which some startups lack due to market pressures [00:39:06], [00:40:08].
• Global Impact: While some students are self-driven, structured approaches are needed for widespread global impact [00:35:31]. Providing access to basic devices with AI in underserved areas, combined with accredited content and certifications, could transform education globally [00:36:55], [00:37:39].

Despite initial concerns about adoption and potential backlash, the education community has moved faster than expected in embracing AI [00:48:36]. Organizations like Khan Academy now view themselves as AI-first, constantly adapting and innovating with these new capabilities [00:48:22].