Toward a Comprehension Challenge, Using Crowdsourcing as a Tool

yet in many ways remains superficial. While the AI scientific community had hoped that by 2015 machines would be able to read and comprehend language, current models are typically superficial, capable of understanding sentences in limited domains (such as extracting movie times and restaurant locations from text) but without the sort of widecoverage comprehension that we expect of any teenager. Comprehension itself extends beyond the written word; most adults and children can comprehend a variety of narratives, both fiction and nonfiction, presented in a wide variety of formats, such as movies, television and radio programs, written stories, YouTube videos, still images, and cartoons. They can readily answer questions about characters, setting, motivation, and so on. No current test directly investigates such a variety of questions or media. The closest thing that one might find are tests like the comprehension questions in a verbal SAT, which only assess reading (video and other formats are excluded) and tend to emphasize tricky questions designed to discriminate between strong and weak human readers. Basic questions that would be obvious to most humans — but perhaps not to a machine — are excluded. Yet is is hard to imagine an adequate general AI that could not comprehend with at least the same sophistication and breadth as an average human being, and easy to imagine that progress in building machines with deeper comprehension could radically alter the state of the art. Machines that could comprehend with the sophistication and breadth of humans could, for instance, learn vastly more than current systems from unstructured texts such as Wikipedia and the daily news. How might one begin to test broad-coverage comprehension in a machine? Articles