While artificial intelligence is being painted by companies and government as the catch-all answer to many of today’s inefficiencies and problems, it currently has one glaring shortcoming: It can’t answer Common sense questions.
In an effort to address this current shortcoming of AI, The U.S. Department of Defense (DoD) is committing $2 billion dollars over the next five years to its Machine Common Sense (MCS) Program. The program aims to enable computers to communicate naturally, behave reasonably in new situations, and learn from new experiences.
Thanks in part to Iron Man (and Elon Musk) fame, the Defense Advanced Research Projects Agency, aka “DARPA”, an agency within the DoD, may be one of the few alphabet soup government agencies with a future-tech-savvy reputation. That reputation is well deserved, too, if history has anything to say about it. As the agency that gave us the Internet through an extension of a defense communication project, just having a discussion online about DARPA itself is testament to the tech potential it represents. The challenge of creating true, thinking computers is perfectly aligned with what DARPA has done well with overall.
As the advancement of computer technology increases at a near exponential rate, so too has the potential relationship between them and humans. However, the possibility of a troubling disconnect is also a growing reality. In other words, humans and computers currently operate very differently from one another, and that could spell bad things for the weaker logician of the two. Yeah, that means us.
Elon Musk has famously harped about this predicted disconnect on numerous occasions, and one of the companies he’s invested in, Neuralink, is working on preemptive solutions for its coming problems. While Neuralink generally aims to help human brains work more like computers, DARPA is taking the approach of having computers work more like humans.
The term “common sense” can often be tossed around in conversations to imply a variety of shared knowledge bases, but as a federal government agency, DARPA has its own specific definition for this context: “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.” By mimicking the cognitive processes we go through when we are young, the agency hopes computers will develop the “fundamental building blocks of intelligence and common sense” just like a human.
With advanced neural networks making amazing (and humorous) headlines regularly, what would a “common sense” machine bring to the table in terms of advancement? One primary answer is the requirement for less initial information. To quote Dr. Brian Pierce, director of DARPA’s Innovation Office, at a recent summit, “We’d like to get away from having an enormous amount of data to train neural networks.” If a machine could use its environment to deduct answers when compared to its existing knowledge base, as humans do, it wouldn’t need to be taught to interpret data solely based on an enormous amount of data previously provided. Essentially, it could think for itself using common sense.
DARPA has now completed a “Proposers Day” wherein potential contractors were presented with the agency’s specifics for its MCS program. The next step is a “Broad Agency Announcement”, i.e., a formal invitation for proposals to work on the project with the hope of obtaining a federal contract to fulfill its aim.
If the contract winner is successful, will common sense lead to computer behavior we’d welcome rather than fear? Hopefully that will be figured out sooner rather than later.
The post US Department of Defense commits $2B to training AI to have “common sense” appeared first on TESLARATI.com.