Backpropagation, short for "backward propagation of errors," is a cornerstone in the training of artificial neural networks. It operates on the principle of gradient descent, where the algorithm calculates the gradient (or derivative) of the loss function (a measure of the difference between the network's prediction and the actual data) with respect to each weight in the network by propagating the error backward through the layers. This process allows the algorithm to adjust the weights in a way that minimizes the overall error. Backpropagation's efficiency in training deep neural networks has been pivotal for advancements in fields like computer vision, natural language processing, and beyond, making it a critical technique in modern AI systems.

The concept of backpropagation has been around since the 1970s but gained significant popularity in the 1980s, particularly with the publication of the work by Rumelhart, Hinton, and Williams in 1986, which presented it in the context of neural networks and learning processes.

The development and popularization of backpropagation are credited to David E. Rumelhart, Geoffrey E. Hinton, and Ronald J. Williams, whose collaborative work in the mid-1980s laid the groundwork for the algorithm's application in neural network training.