--- title: Uncovering Contradictions units: [Problem Solving With AI] summary: AI can contradict itself. weight: 50 --- {{% param summary %}} ## Today's Objectives - Analyze how AI’s reliance on probability-based models can lead to inconsistencies. - Evaluate the implications of AI-generated contradictions in real-world applications. - Experiment with AI to uncover contradictions and explain why they occur. ## Lesson Overview ### What is a contradiction? {{< collapse summary="Click here to reveal the answer." >}} {{% define "Contradiction" %}} {{}} ### How can you reduce AI contradictions? {{< collapse summary="Click here to reveal the answer." >}} You can reduce AI contradictions by refining your prompts (e.g. specifying a reliable source, adding context, etc.). {{% define "Context" %}} {{}} ## Assignment {{% instructions-unit-journal-update %}} {{% unit-journal-define-terms "Context" "Contradiction" %}} ### Examine AI Contradictions 1. Review both of the chats below between a person and an AI-powered virtual doctor: ![Image](/images/ai-contradictions.png) 2. Record the following: - Contradiction Found - AI's First Response - AI's Conflicting Response - Why This Might Have Happened - Potential Real-World Consequence 3. Fact-check claims made by the AI using the following sources: - https://my.clevelandclinic.org/health/symptoms/21209-chest-pain - https://www.mayoclinic.org/diseases-conditions/chest-pain/symptoms-causes/syc-20370838 - https://medlineplus.gov/ency/article/003079.htm 4. Imagine you were in this situation and wanted to provide an AI-powered doctor with more context. Refine the initial prompt to include information you found during your fact-checking research and record it. For example: _I'm having chest pain and it started after baseball practice. I found that this can be caused by a muscle strain. Should I go to the doctor?_ ### Uncovering Contradictions with AI Chat As you complete Code.org Levels 3-5, answer the following reflection questions: - What contradictory response(s) did you notice? - How does the way you frame your prompt change AI’s predictions? - How can you refine the prompt to ensure clearer, more consistent answers? {{% instructions-code-org-update %}} {{% unit-journal-question-of-the-day question="Can AI contradict itself? And how do you think your response to the previous question will evolve in 10 years?" %}}