Algorithmically-Guided User Interaction

There are many practical problems in GIS that currently cannot be solved automatically, not because our algorithms are too slow but because we have no satisfactory algorithm at all. This can occur when semantics are involved, such as when extracting information or designing visualizations. A computer currently cannot be expected to solve such problems in a completely unsupervised manner. To achieve high productivity anyway, we explicitly consider human effort as a resource. Clearly the algorithm should do as much of the work as is possible, at high quality -- but crucially the algorithm should also be smart enough to see where it needs help, what it should ask the user to do, and how it takes those answers into account. This concept relates to emerging fields such as human(-based) computation and active learning, but we put the focus on the proper design and analysis of algorithms, and on the resulting dialogue between algorithm and human, which we call algorithmically-guided user interaction. As a showcase, we argue that this approach should be applied to information extraction from historical maps.