AI, Creativity, and Ethics

Establishing a Working Model

The morning opens with a minimal conceptual model of AI:

• AI systems generate text by predicting the next word based on patterns in large datasets
• They do not verify truth, consult sources, or possess understanding
• Their outputs are shaped by statistical likelihood, not intention

Students are asked to hold this model provisionally and test it against what they observe.

Key question:

• If this system does not understand, why does it so often sound like it does?

Demonstration: Competence Without Understanding

Students examine an AI-generated response to a structured prompt (for example, a literary or historical explanation). The output is typically:

• Coherent and well-organized
• Written in recognizable academic prose
• Supported with plausible examples

Working in small groups, students evaluate the response as if it were submitted by a peer.

Discussion focuses on:

• What signals “competence” in writing
• Which elements create trust (structure, tone, specificity)
• Whether those signals actually indicate understanding

Key question:

• What would you infer about the author if you did not know this was AI?

Demonstration: Failure Under Constraint

Students are given a second AI-generated response to a more demanding prompt (one that requires integrating multiple kinds of knowledge; for example, combining textual analysis with technical detail, or connecting evidence to a specific formal constraint).

The output typically:

• Maintains fluency and confidence
• Produces plausible but loosely connected elements
• Fails to achieve genuine integration

Students are asked to identify where the response breaks down.

Discussion focuses on:

• “Constraint collapse” (the system cannot reliably satisfy multiple precise demands at once)
• The difference between surface coherence and internal coherence
• The role of disciplinary expertise in detecting error

Key questions:

• Where does the argument stop making sense?
• What knowledge would you need to detect the problem quickly?

Structured Analysis: Reliability and Risk

The group shifts from examples to generalization. Students synthesize what they have observed:

• AI outputs can be correct, incorrect, or indeterminate—and these cases are often indistinguishable at a glance
• Confidence and correctness are not correlated
• The burden of evaluation shifts to the reader

Students consider practical implications:

• When is AI output usable without verification?
• When does it require expert review?
• What kinds of tasks are most vulnerable to subtle error?

Closing

The session ends by returning to the central premise:

• Students are not learning to use AI as a tool. They are learning to interpret its outputs, to recognize its limits, and to make informed judgments about when and how it can be trusted.

Question for the afternoon/next day:

• What does it mean to “know” something in a context where fluent answers are always available?