[rcavezza]

This is 100% true. So, so, so true. In actuality, A.I. will not ever 100% solve all problems and some sort of human intervention is typically needed to serve as a QA purpose.

AI can get you 80%, 90%, 95%, and 99% of the way to solve a problem. When you first start out, a human element will be doing most of the heavy lifting and creating training data for you. As you iterate on the model and improve it, the human intervention element will be less and less time consuming over time. As you near 100% accuracy, you will probably still want a human intervention element acting as QA because all models have some rate of error.

[alexcnwy]

You can use the model's output probabilities to optimize QA routing - any cases with less than 90% confidence to humans but save humans doing the cases that can easily be done by models :)