Central banks using machine learning face the problem of hidden biases in models and the tendency for artificial intelligence (AI) to “hallucinate”, says the Bank for International Settlements (BIS). 

In a bulletin published on January 23, BIS researchers outlined both the promise and potential pitfalls of AI. The non-linear interaction between many variables can make it hard to understand how a particular machine learning model arrived at a conclusion. This is known as the “black box” effect