There’s an idea that simply won’t leave me alone – coupling largely deterministic algorithms with machine learning capabilities in order to guide behavior (but not completely determine it, as with genetic algorithms). For example, if we want to search for objects with a color attribute of red and we know that red clusters with blue through ML techniques, we can do a general binary search on a blue (or red) object and focus there. In this way, we can give an ordering to data that may not necessarily have as natural an ordering as an ordinal number. If the ML doesn’t dominate the algorithm, this could potentially speed up the creation of indices and search operations, among other things.

Just a thought.