The purpose of an item bank is to manage the large number of items needed for administration
With the use of metadata
➔Test content can be searched at the item and model level
➔Replacement for items can be easily specified for replacement
➔Relationships between the metadata, as a parallel layer of information inside this bank, can be used enhance how we can manage generated items, items to be generated, and item models

In contrast to traditional item writing, item models provide a source of explicit validation for the testing content, and an ability to scale item development on an ongoing basis.

In contrast to fully automated approaches to item generation, our cognitive modelling approach provides points of human validation, expert input, and a deterministic generation process.

Item models provide the flexibility to change and place requests on-demand
Rather than a traditional item development approach:
○     Item models are a collection of seeds for the possible items
○     Item generation grows the permutation of items from each model by needs
○     Items that meet the specification are harvested for development, with unused items remaining in the model

AIG amplifies subject matter expertise it does not replace it.
If subject matter expertise is not driving item development how is learner knowledge being validated? One item at a time after items are generated or one item model at a time prior to items being generated? How is the issue of scalability solved if items continue to be reviewed one item at a time? So many questions to ponder in this era of AI.