Artificial intelligence is taking large steps toward "broader AI", i.e. the application of methods of intelligence to the study of problems which humans do not solve (or solve well or inefficiently). Since there is no human template for solving the problems, there is a requirement for methods of discovery and unsupervised analysis. These methods need to satisfy a number of requirements, including that they be genuinely unsupervised, without biases coming from the hypotheses of the investigator, and that they possess transparency, permitting the understanding of how the algorithm works, explanations of results in human terms, and the ability to diagnose problems in the algorithm. I will discuss such a methodology, with examples.
.jpg)
.jpg)
