This interactive learning resource has presented the issue of bias in ChatGPT in several modules. This module will allow you to imagine that you are in the position of choosing what training data to use. It will allow you to think critically about how biases can be addressed and reduced in future iterations of AI language models, like ChatGPT.
Scenario: You are in the position to decide what future training data will be used with ChatGPT.
What considerations will you make to ensure that ChatGPT is trained on data that represents the diversity of society? What populations/groups will you prioritize? Why?
How will you obtain this information? What sources will you use and how will you ensure the credibility of these sources?
Will you publish the training data that you use to improve the transparency of ChatGPT? What implications might this have for individual privacy?
Submit your answers to these questions in the Google form below.