Last week, Business Insider published an interesting article about a leaked memo that Walmart sent its regional and store managers allowing them to either fully or partially shut down its automatic inventory management system (called ISA): “‘We have heard your concerns around the ISA tiering system,’ and regional general managers ‘will now be able to submit changes,’ the memo said. Depending on how overloaded their stores are, these managers can decide to keep ISA on, turn ISA off for general merchandise but keep it on for food and consumables, or turn off ISA completely. The memo also outlined various inventory processes stores should complete before opting to turn off ISA.”
Totally agree with the collaboration of algorithm and human can bring the most benefits. However, do you think that once the algorithm is strong enough, like it has the database of extreme situations like festivals and will adjust automatically to it. Do we still need people to intervene or supervise algorithm？
Providing model explainability/interpretability should also help reduce model aversion since the humans can see the top drivers for a given recommendation. Such increased transparency can also allow humans to provide targeted feedback on model performance.