IAPP Artificial Intelligence Governance Professional (AIGP) — Question 136

Business A sells software that provides users with writing and grammar assistance.
Business B is a cloud services provider that trains its own AI models.

• Business A has decided to add generative AI features to their software.
• Rather than create their own generative AI model, Business A has chosen to license a model from Business B.
• Business A will then integrate the model into their writing assistance software to provide generative AI capabilities.
• Business A is most concerned that its writing assistance software could recommend toxic or obscene text to its users.

Which of the following governance processes should Business A take to best protect its users against potentially inappropriate text?

Answer options

• A. Business A should fine-tune the AI model on user-generated text that has been verified to be appropriate.
• B. Business A should test that the AI model performs as expected and meets their minimum requirements for filtering toxic or obscene text.
• C. Business A should establish a user reporting feature that allows users to flag toxic or obscene text, and report any incidents to Business B.
• D. Business A should ask Business B for detailed documentation on the generative AI model’s training data and whether it contained toxic or obscene sources.

Correct answer: B

Explanation

The correct answer, B, is vital because it ensures that the AI model is evaluated for its effectiveness in filtering harmful content before being integrated. While options A, C, and D are helpful measures, they do not directly address the immediate need to ensure the model's performance in mitigating toxic or obscene text, which is Business A's primary concern.