Or could
transfer in these experiments be explained by internal production
alone, regardless of whether prediction is
involved? One argument for the role of prediction in transfer
is the phenomenon of error-based learning, i.e. the
learning that arises when a learner’s prediction is wrong.
Error-based learning is a cornerstone of psychological theory and is formalized in the delta rule, a learning algorithm used to train many connectionist
networks. The network’s weights are updated as a function
of how the network’s actual output deviates from the
target output.