Descriptive languages and problem solving
Marvin Minsky · 1961
Advances in machine problem solving may depend on use of internal languages for description and abstraction of the outcomes of experiments. As more complex problems are attempted there will have to be less trial and error and more systematic analysis of the results of each trial. Learning on the basis of experience will require a phase of refinement in which the machine will attempt, by analysis and inductive inference, to get as much as possible from each experiment.