In this presentation, I will show through a number of examples how Linked Open Data, and especially DBpedia, have contributed to AI by making it possible to create intelligent, open domain applications, i.e. applications which do not have a fixed domain, or for which this domain is not known in advance. This was made evident through a number of high profile applications (e.g. IBM Watson), but is also now being used for a wide variety of cases, including when having to process streams of ill-described content not restricted to a specific application domain. I will therefore show the advantages of using Linked Data sources such as Dbpedia in those scenarios, as well as the challenges they raise, opening up a number of research directions following the idea of "using the Web as an open knowledge base".