Further complexities arise from the mutability of the system.

“Typically, recommender models get re-trained pretty often and will keep changing over time,” Bindu Reddy, CEO and co-founder of Abacus.AI , an artificial intelligence startup, told TNW.

“While it is possible to release all trained models as well on an ongoing basis, it won’t be very useful either unless you understand exactly what inputs and outputs go into the model for predictions.”