For decades software has been available to catch exceptions in processes and learn from them. That is one major difference between machine learning and simple scripting. The goal of people developing automation is often something like automate the most common 90% of user interactions and leave the most difficult 10% for people, maybe catch them on the next iteration.