Several learning models have been proposed to explain how relatively simple tasks are learned in the laboratory. Some of these models are fairly robust in that they can even be used to explain the learning of tasks as different as arithmetic and visual pattern recognition. Can these models also be used to explain human learning of complex tasks that are encountered in the real world? The goal of a learning model could be to provide a deterministic explanation how someone learns to master a complex task, for example how a novice driver learns to drive a car through a busy city center. Such a model would need so many parameters and processes to represent initial conditions and subsequent states of physical reality that it would be almost impossible to test it. One may therefore consider that an attempt of a novice driver to drive a car through a busy city center has some randomness in it. A model that does take into account the randomness can omit large parts of the physical, but irrelevant, reality, at the expense of exactness. However, such a model might still be able to explain how long it would take a novice driver on subsequent attempts to drive unsupervised from one location in the city to another location in the city, without violating the traffic rules or causing damage. Such a model seems feasible. The colloquium aims at explaining a stochastic model of human learning. Analysis of data from training experiments and test of the model's predictions will be discussed.