I don't think AI bias will mitigate it's evolving. Bias is mainly because as humans we (un)consciously pass our biases into the training data. It will take a long deliberate effort to improve on this.

For me, anomaly detection is among the best use cases for AI. Training data is the lifeblood of AI. Any improvements we want to see must start here. We have actually come a long way. I remember when face recognition models couldn't recognise my black face. I have faith in the evolution of this technology