Unifying reasoning and search into something that scales up to frillions of triples

Recently, we became aware of a Telcom project which required reasoning about 10 billion Triple in less than 100ms. The use case is defined around the generation of new revenue streams through new context-sensitive and personalized mobile services. Currently, we know about approaches that can handle RDFS-query answering for around 100 million triples in 100ms (assuming triples are not deleted too often). But in this case, the number of triples is two orders of magnitude bigger and sophisticated reasoning is required. Certainly the requirements will grow. Over time, requirements of scale may grow much faster than any progress in reasoning algorithms, clever coding and improved hardware can compensate. Since we had to say no to potential customers, we started to wonder why this problem exists at all. Usually, problems become intractable through improper conceptualization. Intelligence is requested to introduce assumptions that on the one hand make the problem solvable, without on the other hand restricting them in a way that they lack any usefulness. So the question is: Why is reasoning not scaling for the web and how can this be fixed? In the remainder of the article, we will quickly sketch the state of the art and its deficits. Then we will relate this to a comparable problem (and paradigm shift) in economics, namely the assumption of complete information underlying complete rationality and the insight that reasoning with limited rationality is a much more rationale behavior. Finally, we sketch a new research direction that tries to integrate reasoning and search into one new paradigm that we call reasearch for the moment. It is an attempt to truly integrate the web and reasoning and therefore come up with something that works at web scale.