The Next Big Barrier Facing Artificial Intelligence: Common Sense
When armed with the right model, a machine-learning platform can rapidly improve its accuracy and success rate. Just look at Google’s A.I. tool that can detect the most common forms of lung cancer with 97 percent accuracy, or self-driving cars that travel for thousands of miles without so much as a fender-bender. With the incredible amount of resources and brainpower devoted to machine learning and A.I., it’s inevitable that these platforms will only get “smarter” in the years to come. However, there’s just one little problem: A.I. lacks what we call “common sense.” DARPA—that’s the agency of the Department of Defense (DoD) that researches and prototypes all kinds of crazy inventions—wants to build common sense into A.I., presumably so that future military robots don’t accidentally tumble off cliffs or into walls. DARPA’s Machine Common Sense (MCS) program will host a competition to come up with solutions. “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,” Dave Gunning, a program manager in DARPA’s Information Innovation Office (I2O), wrote in the agency’s blog posting on the program. “This absence is perhaps the most significant barrier between the narrowly focused A.I. applications we have today and the more general A.I. applications we would like to create in the future.” The MCS program will initially take two approaches. The first approach will attempt to create models that learn in a more human-like way. “Developmental psychologists have found ways to map these cognitive capabilities across the developmental stages of a human’s early life, providing researchers with a set of targets and a strategy to mimic for developing a new foundation for machine common sense,” Gunning added. The second approach will utilize machine learning, crowdsourcing, and information extraction to create a “common sense repository,” which will answer queries about common sense in general. “The resulting capability will be measured against the Allen Institute for Artificial Intelligence (AI2) Common Sense benchmark tests, which are constructed through an extensive crowdsourcing process to represent and measure the broad commonsense knowledge of an average adult,” DARPA suggested in its posting. You can imagine A.I. researchers using such a database to inform their platforms. In theory, the introduction of common sense will go a long way toward helping software and hardware to learn faster. For instance, if a machine knows that certain outcomes are inadvisable or impossible, it will disregard those in favor of workable possibilities. Your vacuuming robot won’t attempt to drive off a new, unknown cliff in your home. Research like this tends to end up in production environments; if DARPA succeeds, expect to see other A.I. builders rapidly incorporate “common sense” algorithms into their work.