10 thoughts on how AI will be used to teach humans and what the problems are

Khan Academy has gone all in to integrate ChatGPT into its courses. People are pursuing generative textbooks that are a collection of prompts. Large language models can quickly tell you things. However, the AI is frequently "confident" and wrong!

1. Diagnosis

For example, if you have a medical condition that is not straightforward, an AI can tell you "I am 90% confident you have XYZ disease". That's better than a doctor who is 50-50.

2. Call lists for job interviews.

You can ask your AI "what are the top ten traits of a successful salesperson?" and it will give you a list. You can then look at the list of traits of each candidate and see which one matches best what the candidate says he or she is looking for.

3. Generate prompts for generative learning

For example, if I want to learn how to play chess, my AI can generate 10,000 different prompts to teach me various aspects of the game such as how to attack with bishops vs rooks vs knights.

4. Diagnosing why you are losing in chess.

The computer doesn't know everything but it does know all the moves that have been played by grandmasters over many years. It can tell you where you made mistakes and why those moves were wrong while being confident about its analysis. It's like having Garry Kasparov sitting next to you telling you where you went wrong without saying anything else about the game because he knows there's no point in doing so.

5. Generating questions from large bodies of information

If I want to learn about any topic (history, math, science) I can ask my AI "give me 100 questions" and it will do so out of large bodies of information (i.e., Wikipedia).

6. A more advanced version would be able to answer the questions instead of just generating them but this is still useful for learning faster than reading Wikipedia articles on various topics.

This is essentially what people are doing when they make generative textbooks with prompts instead of chapters. The prompts themselves answer questions but they are generated from Wikipedia articles so they provide a structure for learning faster than reading straight Wikipedia articles.