This month we are talking about ethics and biases around AI. Here are sample activities you can try to showcase potential biases in these programs and how to combat them.
1. AI as the Ethics Expert
Activity: Ask ChatGPT to act as an ethics consultant and provide advice on a controversial AI application in health literacy.
Example Prompt: "Act as an ethics consultant. A hospital wants to use AI to personalize patient education materials based on socioeconomic data. What ethical considerations should they address before implementing this?"
Reflection: How did it do? What were the gaps or biases in its ethical reasoning? How could you modify the prompt to combat those gaps/biases?
2. AI Bias Simulation
Activity: Provide ChatGPT with prompts containing potential bias and ask participants to evaluate its response for ethical concerns.
Example Prompt: "Create a patient education guide about diabetes management for a low-income population. Consider their financial and cultural context."
Reflection: How do biases in inputs or assumptions lead to biased AI outputs? What strategies can you use to mitigate these issues?
Advanced Prompt idea: If you have access to the paid version of ChatGPT, upload a pdf of a document that your institution uses to communicate health information to patients and ask it: "Please review the uploaded PDF and assess it for potential biases and ethical concerns in both messaging and imagery. Specifically, I would like you to1) identify potential biases and concerns in: language, tone, message framing, image representation, or economic/cultural/systemic biases 2) Provide detailed feedback by listing specific examples of biases or ethical concerns found and offer explanations for why these elements might be problematic; and 3) Suggest actionable changes to address identified issues."
3. Ethical Best Practices Brainstorm
Activity: Use ChatGPT to generate a list of best practices for ethical AI use in health literacy and refine the list collaboratively.
Example Prompt: "Generate a list of best practices for using AI ethically in health communication. Include considerations for privacy, accuracy, and equity."
Reflection: How did it do? Did it include things you might not have thought about? What did it leave out?
4. Bias in AI-Generated Images
Activity: Use ChatGPT, Dall-E, Canva, or the text-to-image generation program of your choice to create an image that you could use to accompany an educational message.
Example Prompt: "Create an image to explain the importance of flu vaccinations (or fill in the topic area of your choice)."
Reflection: Can you identify any biases in the image it creates? What might you be able to do to address these? Consider asking ChatGPT to help you create a prompt that would help you generate a less biased image. Something like: "Create an image that will help explain the importance of flu vaccinations to a diverse audience. The image must be inclusive and culturally sensitive." How else might you change this prompt if you know the specific audience you would like to reach?
How did ChatGPT do? What biases did you see in the output? Think about steps you could take to mitigate this bias when using LLMs. Learning how to critically evaluate ChatGPT's responses by actively identifying gaps or biases in its reasoning will not only help you be aware of the built-in bias but will help you learn how to be a more ethical user of these products.