There are three main ways to mitigate bias in AI models: One way is to ensure biases are addressed during the model building by setting guidelines and principles. Another method is to correct biases after the model is built by instructing the model to adhere to specific rules like not being offensive or stereotyping. The third technique involves prompt transformation where the requests made to AI models are modified by prompt engineers to ensure desired results are achieved. This unseen editing of prompts before they reach the AI model improves the quality of responses and ensures better outcomes.