Backpropagation is a training algorithm for neural networks. It adjusts the weights of the network by propagating the error backwards from the output layer to the input layer, improving the model's accuracy.

Systematic errors or distortions in the outputs of an AI system which reflect or reenforce discriminatory patterns or stereotypes. This is often due to underlying issues in the training data of an AI system (also see Prejudice).

An AI system whose internal workings are not easily interpretable or transparent, even to its developers.