While today's machine learning systems are more advanced than ever, says the agency, a critical component of artificial intelligence (AI) remains just out of reach – machine common sense. The agency defines "common sense" as essential background knowledge that offers "the basic ability to perceive, understand, and judge things that are shared by nearly all people and can be reasonably expected of nearly all people without need for debate."
Such an ability, says the agency, could significantly advance the symbiotic partnership between humans and machines, but encoding it is "no easy feat."
"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," says Dave Gunning, a program manager in DARPA's Information Innovation Office (I2O). "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."
While efforts to instill common sense in machines are not new, none have yet succeeded in achieving a widely applicable common sense capability. However, says the agency, significant progress in AI along a number of dimensions has made it possible to address this difficult challenge today.
The newly created Machine Common Sense (MCS) program is designed to develop just such new capabilities by soliciting research proposals in the area of machine common sense to enable AI applications "to understand new situations, monitor the reasonableness of their actions, communicate more effectively with people, and transfer learning to new domains." The program will explore recent advances in cognitive understanding, natural language processing, deep learning, and other areas of AI research to find answers to the common sense problem.
The program will pursue two approaches for developing and evaluating different machine common sense services:
- Create computational models that learn from experience and mimic the core domains of cognition as defined by