Why Sal Khan’t: The Gap Between Learning by Doing and Teaching by Telling

Originally published on Punya Mishra’s Web and cross-posted to the Civics of Technology blog.

Recently, two distinct pieces of commentary caught my attention, both centering on Sal Khan and the revolutionary stalled promise of AI-driven tutoring.

1. The Chalkbeat Report: Matt Barnum detailed how Khan admitted that Khanmigo—the AI chatbot launched three years ago with global ambitions—ended up being a "non-event" for the majority of students. Even Kristen DiCerbo, the Chief Learning Officer, was blunt:

   “So far I am not seeing the revolution in education.”

2. The LinkedIn Obituary: Dan Meyer wrote a scathing piece titled "RIP Khanmigo," tracing the trajectory from optimistic TED talks and massive funding to a product that had to be forced upon students because they wouldn't use it voluntarily.

Meyer’s point was simple: If a tool with early OpenAI access, Microsoft's backing, and Sal Khan's immense network failed, what chance does the rest of the edtech industry have with similar chatbot models?

---

The Deeper Question: Why was it doomed?

The failure isn't just about software or adoption rates; it's about the fundamental nature of how humans learn. To understand this, we have to look at the difference between constructing knowledge and receiving it.

The Paradox of Sal Khan's Method

Years ago at MSU, I used to share a clip from a Charlie Rose interview where Sal Khan explains his own process for mastering a new topic (like the French Revolution or Napoleon).

Khan's personal learning workflow looks like this:

• Scaffolding: He starts with Wikipedia to build a mental map.
• Visualizing: He creates timelines and diagrams.
• Deep Inquiry: He hunts for the "missing" answers. For example, when reading about neurons, he noticed textbooks mentioned the myelin sheath makes signals faster, but didn't explain how.
• Analogical Thinking: He used concepts like fiber optics to hypothesize the answer.
• Social Validation: He called biologist and engineer friends to ask, "Does this make sense?"

When those experts admitted they didn't know the answer, Khan didn't give up; he was fascinated that the textbooks had omitted the uncertainty.

The "Product" vs. The "Process"

There is a massive cognitive gap between what Sal Khan does to learn and what he asks students to do.

We intuitively know that watching a video of someone explaining a concept is not the same as understanding it. Yet, we often fall for what I call the reductive seduction of other people’s problems—the belief that we can solve learning by simply making the delivery of information more efficient.

---

The Dewey Framework: Four Impulses of Learning

To analyze this further, we can apply the theories of John Dewey. He argued that learning is driven by four innate impulses:

When we look at Sal Khan's own habits through this lens, we see a perfect alignment:

• Inquire: His relentless "why" and refusal to accept surface-level textbook answers.
• Construct: The maps, drawings, and scaffolds he builds for himself.
• Express: The act of creating the video—giving a physical form to his understanding.
• Communicate: Testing his theories against other minds.

In Khan's own life, these are intermingled. In Khanmigo, however, the student is offered a chatbot to help them receive information more efficiently. This is why students responded with IDK IDK instead of curiosity.

---

Conclusion: The Residue of Learning

The tragedy of Khanmigo is that Sal Khan experienced the full, rich journey of learning and then designed a system that offered students only the residue of that journey.

We can express the failure of the AI tutor conceptually: $\text{Learning} \neq \text{Efficient Reception of Information}$ $\text{Learning} = \text{Inquiry} + \text{Construction} + \text{Expression} + \text{Communication}$

Because the motivation and purpose of learning live within the journey of making, students naturally declined a tool that only offered the destination of telling.

Summary Checklist for Future EdTech

• Provide tools for inquiry, not just answers.
• Encourage the construction of personal artifacts.
• Move beyond the "chatbot as tutor" paradigm.
• Prioritize the process over the product.