Considerations of an AI Chat Bot
Last week, Sal Khan came to speak at my school. As someone who learned how to factor from his videos, I was very excited for it. In his talk, he introduced Khanmigo, Khan Academy’s AI-Powered teaching assistant. He explained its many benefits, including helping students complete assignments and assisting teachers create lesson plans through personal, albeit experimental, help. His perspective of using AI as an assistant emphasized increased throughput while still preserving effort, allowing teachers and students to be more effective with their time.
Hearing about AI being incorporated into a platform that I grew up using was exciting to me. In the past year, we have witnessed a crazy progression of AI as it gets incorporated into existing products – Nearly every search engine includes AI generated answers, Adobe products have started to include generative fill, and social media has started using AI to maximize engagement. Like many, I am suprised by how quickly AI Chat Bots have improved.
I remember last winter break showing my mom ChatGPT. I just completed a class on Zora Neale Hurston, with my final paper analyzing one of her books. In an effort to show my mom how quickly ChatGPT can generate content, I asked it to create a final paper for the same prompt, using my original bare-bone outline. We were both impressed as ChatGPT delivered a polished paper, including quotations and references to her other works. However, upon closer inspection, I realized that I did not recall any of the ChatGPT proclaimed quotes from the book. Instead, these ‘quotes’ were made up to support ChatGPT’s arguments, using the similair langauge style that Hurston had used. This caused me to reconsider the strengths and weaknesses of AI Chat Bots. They can be useful in idea generation and pointing someone in the general direction, however lose accuracy in details.
With this personal experience under my belt, I was hesitant at Sal Khan’s optimism about Khanmigo. What would happen in Khanmigo was wrong about something as a student learns the fundamentals? This could negatively impact their education.
With this uncertainty in mind, it was interesting to learn the ways Khan Academy has minimized this risk. The plentiful testing and carefully choosen features lend itself to AI.
Moreover, it was refreshing to see AI implemented cautiously, and for the overall goal of imoproving education access. Many implementations of AI seem to be a rushed reaction to the AI movement, and used to sell a product. Overall, his talk made me hopeful for the future of AI.