Machine-generated content: Creating compelling new content from existing online sources.

摘要

Human beings are prolific content producers and consumers. We are constantly talking, telling stories, reading books, drawing cartoons, listening to music, sending emails, watching YouTube, and relaxing with reruns of "The Office". It's not surprising, then, that since almost the dawn of modern computing, researchers have been trying to imbue automated systems with some of our ability to create interesting content. Unfortunately, creating content that is interesting--that others will find genuinely compelling--is extremely difficult to do, for people as well as machines. Furthermore, people have always possessed a distinct advantage over machines when creating content: humans are social creatures, and we are able to use popular culture, shared references and context, existing content, in-jokes, societal memes, and the national zeitgeist as an effective starting point for our own content creation. In contrast, computers have typically been effectively stranded on cultural islands, trying to create and compose in a vacuum. Humans' advantage here is rapidly dissolving, however; as more and more data, information, news, jokes, stories, pictures, and movies move online and become machine-readable, automated systems are able to use these available materials as "grist for the mill" to create their own content. Driven by programmable narrative arcs, such systems can autonomously discover, extract, and refine the vast amounts of information, content, and media already available on the internet; these systems can then intelligently combine and repurpose the material to generate a wholly new, rich, and compelling presentation. Below, I discuss in detail this method for creating "machine-generated content." I also discuss various systems developed in the lab that embody our approach, including one in particular: "News at Seven", an automatically generated news and entertainment show. These systems are rudimentary, but I believe the future is clear: as presentational elements improve, and as more and more data, information, and content move online and become machine-readable, the power, flexibility, and breadth of systems that are able to autonomously create new, personalizable content from these materials will improve likewise. In the not-distant future, I expect that a significant portion of the content we consume on a daily basis will be created by machines.