Four monkeys are placed in a cage. Every day, a plate of fresh fruit is placed into the cage. As the monkeys reach for the food the keepers come in and beat them. Over the course of a few weeks the monkeys learn not to touch the fruit.
Now one of the monkeys is replaced with a new monkey. This new monkey is happy to reach for the food. Fearing retnbution, the other monkeys attack him. Soon, this monkey also learns not to touch the fruit.
At this point another monkey is replaced, and the process repeats.
By the time the nineth monkey is introduced he is being attacked by four monkeys that have never been beaten by the keeper.
How is this relevant to machine learning?
Like monkeys, machines leam based on prior experience, in this case represented by data.
This makes machines bad at suggesting new, better ways of doing things. Inherent bias in the data, missing or inaccurate data, and other data quality issues will lead machines to learn bad habits, or to make poor recommendations.