Slashdot Mirror
The US Military Wants To Teach AI Some Basic Common Sense (technologyreview.com)
DARPA, the research arm of the U.S. military, has a new Machine Common Sense (MCS) program that will run a competition that asks AI algorithms to make sense of questions with common sense answers. For example, here's one of the 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." MIT Technology Review reports: A computer program needs some understanding of the way photosynthesis works in order to tackle the question. Simply feeding a machine lots of previous questions won't solve the problem reliably. These benchmarks will focus on language because it can so easily trip machines up, and because it makes testing relatively straightforward. Etzioni says the questions offer a way to measure progress toward common-sense understanding, which will be crucial. [...] Previous attempts to help machines understand the world have focused on building large knowledge databases by hand. This is an unwieldy and essentially never-ending task. The most famous such effort is Cyc, a project that has been in the works for decades. "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,"https://www.darpa.mil/ Dave Gunning, a program manager at DARPA, said in a statement issued this morning. "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."
I always thought that a reasonable definition of common sense was the set of rules you fall back on when you don't have sufficient specific knowledge to address the issue at hand. For example, you may not specifically know how to repair your car, so common sense tells you to seek help from someone more qualified.
As you say, there is no common sense that can be applied to plants and sunlight, you either know about the process or you don't. A system applying common sense would defer to a botanist or refer to some reference material to improve its skillset or some other such thing
Now that's not to say that you can't infer from other data that perhaps it takes more energy to produce O2 than C02 and guess that the light might be such an energy source but at this point you're falling back on specialist knowledge that it either has or it lacks
Nullius in verba
Yeah, even in the attempted example question, the "common sense" answer is A, but the "actual fact" answer is A and B. A during the day, B at night. C is most likely also true in real scenarios.
My goal in writing automated systems is to make less of the mistakes known by the moniker "common sense," not to make more of them.
If you lack information and are forced into action, "common sense" might be a decent least-bad semi-random choice, but it should never be expected to be correct or optimal.