An AI chatbot tricked me into teaching it algebra this weekend.
My son introduced me to chat.openai.com. I’ve been pretty impressed with the natural-language and literature-based responses it’s been able to give. Many of them are quite funny. I wanted to see how it would do with an algebra problem.
It got surprisingly close to a good explanation of how to solve the problem. Just not close enough to get the right answer.
But after I gave it corrective feedback nine times about different mistakes it made, the AI was able to answer simple algebra problems. (At least for me. I understand that nothing it learns from me will be reflected in anyone else’s experience.)
What struck me most, though, was that the way it responded to my corrections encouraged me to keep trying.
This was an AI bot. I didn’t use the sandwich method to buffer my feedback. I straight-up told it where it was wrong.
When I did, it simply agreed and apologized:
“You are correct that the pairs (4,3) and (3,4) do not satisfy the condition x*y=10. I apologize for the mistake in my previous response.”
Then it tried again with a new explanation, this time incorporating the feedback I had given. Eventually, it was able to answer new algebra problems and find the correct answer.
Here’s where it tricked me into persevering.
Even though my investment of time in teaching the AI bot algebra wouldn’t last beyond my session, I wanted to keep teaching it. Why? Because it accepted my feedback and showed me immediately that I had made a difference. It didn’t just appreciate my investment of time and energy, it appeared to value that investment.
The AI never took anything personally. There was no emotion. There was simply an eagerness to learn, and proof that it did learn.
But that’s rarely how I take corrective feedback.
My instinct is often to explain why I reached the (wrong or ineffective) conclusion I did. Or maybe I’ll bristle at the timing or tone of the delivery.
Imagine how much more we can learn, though, if we can take correction more like an AI bot does.