The general idea behind machine learning is that, instead of people writing programs to extract information from data, machines extract information by learning from examples. Neural networks are widely used in machine learning, and this paper tries to give an accessible introduction. Currently, machines don't learn like people. A central idea is that things can be associated with lists of numbers, and that similar lists of numbers are associated with similar things. A list of numbers can be considered to be points in a multidimensional data space. The points associated with similar things cluster in multidimension space. Neural networks are very good at separating out the clusters. A major application of machine learning is classification, and this has enormous commercial value, e.g., many problems require a classification between "yes" and "no"---take an action or don't take an action. Any intelligence associated with neural networks comes from (i) the humans who design the network, (ii) the humans who collect and preprocess the training data before it enters a network, and (iii) the humans who postprocess and interpret the numerical outputs of the network. An open problem is that many machine learning systems are black box classifiers and cannot explain their decisions or recommendations. As such, their use may be unethical or even illegal in some jurisdictions. Machine learning is a powerful and valuable technology that can be understood by everyone.
Johnson et al. (Sun,) studied this question.