Measuring Machine Intelligence Through Visual Question Answering

As machines have become more intelligent, there has been a renewed interest in methods for measuring their intelligence. A common approach is to propose tasks for which a human excels, but one that machines find difficult. However, an ideal task should also be easy to evaluate and not be easily gameable. We begin with a case study exploring the recently popular task of image captioning and its limitations as a task for measuring machine intelligence. An alternative and more promising task is visual question answering, which tests a machine's ability to reason about language and vision. We describe a data set, unprecedented in size and created for the task, that contains more than 760,000 human‐generated questions about images. Using around 10 million human‐generated answers, researchers can easily evaluate the machines.