If you went through public school in the last fifteen years, odds are that your math tests included a space for an answer and a follow-up question that asked “Does your answer make sense?” While plenty of kids might have been puzzled by that, being able to recognize a dumb mistake is one of the major things that separates human intelligence from artificial intelligence. With more and more technology becoming dependent on reliable AI, DARPA wants to start closing that gap before stupid mistakes start causing problems, like self-driving cars being trapped in salt circles.
Because our heads are full of contextual clues to how the world works, we can usually spot when something doesn’t make sense. Artificial intelligence, on the other hand, is more like an idiot-savant: it knows how to very specific things really well, but everything else is a mystery to it. According to Dave Gunning, a program manager at DARPA: “The absence of common sense prevents an intelligent system from understanding its world, communicating naturally with people, behaving reasonably in unforeseen situations, and learning from new experiences. This absence is perhaps the most significant barrier between the narrowly focused AI applications we have today and the more general AI applications we would like to create in the future.”
So how do you teach artificial intelligence common sense? Well, you could assemble a database of basic information (“Grass is green! Fire is hot!”) and feed it into an AI, but that would take a tremendous amount of time. Instead, DARPA is starting with teaching AI the basics of understanding language, which is a deep, dark forest in itself. Here’s one of the sample questions:
A student puts two identical plants in the same type and amount of soil. She gives them the same amount of water. She puts one of these plants near a window and the other in a dark room. The plant near the window will produce more (A) oxygen (B) carbon dioxide (C) water.
Notice how the question isn’t asking which plant will die first, or which plant will produce oxygen, which is a more binary response. Instead, the question relies on two pieces of knowledge: that photosynthesis creates oxygen, and photosynthesis needs light to work properly. In time, deftly designed questions like this will hopefully teach AI some basic knowledge of the world and how to solve simple problems in it.
Oren Etzioni, the CEO of the Allen Institute for AI, sums the problem up nicely: “Common sense is the dark matter of artificial intelligence. It’s a little bit ineffable, but you see its effects on everything.”