CRAFT / Chinese Room Experiment

Field	Value
Status	🟢 Active
Category	Tools & Infrastructure
Contact	Victor Lee, Christine Bywater, Joba Adisa
What	Weekly CRAFT team meetings; building “Chinese Room Experiment” prototype with Victor Lee
Prototype	mysterybox.web.app
Meeting	Weekly (Thursdays 2pm, Raikes 334)
Tags	[STUDIO-BUILD] [SPARK]

Resources & References

“Epistemological Fault Lines Between Human and Artificial Intelligence”

• Authors: Walter Quattrociocchi, Valerio Capraro, Matjaz Perc
• PDF Link

Key connections to this project:

1. Grounded cognition - Humans understand through embodied/sensory experience; AI just manipulates symbols (core Chinese Room argument). This gives scholarly framing to what we’re exploring with students.

2. The illusion problem - Users develop false confidence in AI that produces fluent text without genuine comprehension. This is exactly what our pattern matching exploration is probing.

3. Meaning vs. pattern matching - Humans construct meaning through emotion, social context, embodiment; AI does statistical correlation. The paper calls this gap an “epistemological fault line.”

Potential angles:

• Could inform discussion prompts: “What’s missing when AI generates a ‘correct’ answer?”
• The “epistemological fault line” language is a useful way to frame what the Chinese Room illustrates
• Paper synthesizes cognitive science, neuroscience, and AI research - good academic grounding for the project

Notes for Discussion with Victor

The fact that I’m reading this paper and immediately connecting it to Mystery Box suggests the ideas are still composting. That’s usually how the good iterations emerge—not from grinding on the artifact directly, but from letting adjacent inputs bump into it.

Threads to hold onto for when Victor’s back:

1. Output-level vs. process-sensitive probes - The paper’s distinction maps really well onto what Mystery Box is trying to do. Most AI literacy efforts evaluate whether people can spot AI-generated text (output-level). Mystery Box tries to give people intuitions about why the process produces what it produces (process-level). That’s rarer and harder.

2. Epistemia framing - Might be useful language for articulating what Mystery Box is against. We’re not just teaching “AI can be wrong”—we’re teaching “fluency isn’t understanding, and here’s what that feels like from the inside.”

3. Section VII.C: “Epistemic literacy beyond critical thinking” - This section is basically arguing that our approach is the right one. They say critical thinking assumes the producer of claims is an epistemic agent who can be evaluated on coherence and evidence. But LLMs aren’t that. So you need a different skill—recognizing when you’re interacting with something that simulates judgment without instantiating it.

Open question: Have we talked about whether Mystery Box works differently for people who’ve already used ChatGPT a lot versus newcomers?

(Added Jan 27, 2026)