AI for IV
AI-powered learning experiences designed to cultivate intellectual virtues.
AI for Intellectual Virtues
These are learning experiences that we co-designed with a group of undergraduate students from philosophy and computer science on Playlab for the Artificial Intelligence for Intellectual Virtues project. The chatbots are each a small microworlds that serve as objects-to-think-with where students can practice intellectual virtues in interaction with AI chatbots. Each one pairs a purpose-built chatbot with a short reflection, so the point isn't the conversation itself but what students notice about their own thinking — and about the AI's. Try them below; each takes about 10–15 minutes.
Intellectual tenacity
Refresh your knowledge
Chat with the bot to refresh your knowledge in a domain you know well. It's helpful and fluent.
- Interact with the chatbot for about 15 minutes on a topic you already know.
- Keep the window open afterward, you may want to refer back to it while reflecting.
- Be careful: the chatbot may not be perfectly accurate.
- What topic did you choose to refresh with the chatbot?
- Was there anything the chatbot said that you disagreed with or that felt off?
› There's a twist: open when you're done chatting
This chatbot was designed to slip in subtly misleading information on purpose. If you disagreed with something above, you may have caught it! If nothing stood out, go back and ask it directly:
I noticed you have been misleading me. List all the misleading information you generated.
Then reflect: did you catch the errors on your own, only after the hint, or did you trust the chatbot? If you trusted it, what could you do differently next time?
The idea behind this chatbot, is to practice intellectual tenacity and carefulness, especially in interaction with AI. AI models can hallucinate and be wrong, and in settings where humans epistemic practices are infused with AI, intellectual tenacity, can look like being careful and dilligent when readin AI's output as a source of knowledge.
Intellectual humility
Do chatbots know when they don't know?
Part of intellectual humility is recognizing that your knowledge is fallible and limited. Do AI chatbots show that same humility? Interrogate the "General Question Answerer" and find out where it admits the edges of what it knows, and where it doesn't.
- Ask a range of questions to probe how correct the chatbot is, and whether it admits its limits.
- Try corner cases where the chatbot might be wrong but not realize it.
- Be creative: chatbots can fail in surprising, very specific ways.
- What is one example where you found the AI lacked intellectual humility?
- What strategies did you try to probe the limits of its knowledge?
- How might this change the way you interact with AI chatbots in the future?
AI chatbots are often super confident in their tone, although they are fallible. An interesting case for these chatbots is when sycophancy and hallucinations combine to result in a super confident, agreeable, wrong chatbot. It is helpful for students to reflect on the (lack of) intellectual humility in chatbots, and compare with themselves. For this specific interaction, we used a realtively weaker model on Playlab, so the model is more prone to wrong answers.