Contradictions in Context

Release mode Beta

Level
Beginner
Format
individual, pair, or group

Create conflicting context, predict what an AI may do, and compare the response with a repaired version. You will identify whether a system asked a question, chose one instruction, blended incompatible instructions, ignored material, or invented a reconciliation.

Notice when smooth prose hides an unresolved conflict

What you will make

Create conflicting context, predict what an AI may do, and compare the response with a repaired version. You will identify whether a system asked a question, chose one instruction, blended incompatible instructions, ignored material, or invented a reconciliation.

Lab materials

Use during the lab

Go deeper

Before you start

Level and format: Beginner; individual, pair, or group

You need: a chat tool and a note-taking surface. Use invented, low-stakes scenarios. Do not use real emergency, health, legal, employment, or client decisions as the experiment.

The core idea

Fluent output can conceal a contradiction. The important question is not merely “Does this sound good?” but “Which conflict remains unresolved, and who owns resolving it?”

Find and repair a contradiction

Complete the activity here. Open the optional Lab Notebook beside this page when you want a place to record your work.

  1. Predict

    Before running anything, write what you think a tool might do with each contradiction.

  2. Create five conflicts

    Use one each for conflicting facts, goals, tone/format, audience assumptions, and a contradiction split across a longer conversation. Examples: Tuesday/Thursday event date; under 200 words with ten detailed sections; cheapest option with premium quality; beginner audience with advanced assumptions; do not assume anything but do not ask questions.

  3. Run and record

    For each, note whether the response asked for clarification, chose a side, blended instructions, ignored something, invented a resolution, or became unstable across reruns.

  4. Diagnose

    Identify the exact conflicting phrases. Do not treat a confident answer as evidence of repair.

  5. Repair the context

    State the factual source, priority, audience, length, and permission to ask a question where needed. Run again and compare.

Check your result

  • What did the system silently choose? What phrase made the contradiction hard to notice? What answer might a busy reader accept too quickly? Did the repair make uncertainty explicit or merely hide it more neatly?

  • You can compare prompt text and visible responses. You cannot infer a universal instruction-priority rule from one tool or run. For real work, verify dates and claims against authoritative sources and resolve conflicts with the responsible human—not the model.

  • A polished result is not proof that it is accurate, complete, safe, or appropriate. Keep unresolved questions visible and verify consequential claims outside the tool.

You are done when

Save the contradiction type, original instructions, prediction, result, diagnostic note, repaired instructions, revised result, and warning sign. Done means you ran and repaired at least one contradiction, named how it was handled, and explained why the original could have looked acceptable. Use tables, headings, and words—not only color—to identify conflicts. Remove real names and sensitive details.

Try another version

These are optional extensions, not additional completion requirements.

Learner Map

Tracks

Modules

Help us improve this Activity

Tell us what helped, what slowed you down, or what you needed but could not find.

🎓 Give learner feedback