One commonly used form of AI is supervised learning, in which a large number of samples with tags (such as names of items in a photo) are used to train the AI. That sample set should be divided randomly into 3 sets, training, testing, and validation. Great care needs to be put into managing these sets and making sure they stay separate and are used appropriately.

Reason I brought this up --> If a company fails to manage these 3 sets properly, they will end up with an AI that is 100% accurate in demos, and fails miserably in the real world.