How about some kind of machine learning algorithm that is regularly trained on user feedback and ratings of search results? The system has one mode that starts

How about some kind of machine learning algorithm that is regularly trained on user feedback and ratings of search results? The system has one mode that starts out feeding you approximate matches for your search criteria, ordered pseudo-randomly, and has a secondary mode of viewing results based solely on user feedback to order search results? Yeah, it probably wouldn't work but it would be fun to see what it produced.