Computer Program Able To Teach Itself Common Sense
The fears of those who suffer from Frankenstein complex are beginning to be met with a new computer program aimed at teaching itself common sense.
The Frankenstein complex is a phobia that one day robots, or mechanical men, will take over and control the world. This phobia has been unfounded because, so far, robots are only able to know and learn what it has been taught, but researchers at Carnegie Mellon University are creating a computer program that can understand things on its own.
The Never Ending Image Learner (NEIL) is a computer program that runs 24 hours a day, searching through the Web and analyzing images to teach itself about common sense. Researchers say data being generated will further enhance the ability of computers to understand the visual world.
NEIL is able to identify and label objects, as well as recognize attributes like colors, lighting and materials. The program also makes associations between these things to obtain common sense information that people seem to know without ever saying, such as that cars are often found on roads and buildings tend to be vertical. The team pointed out that based on text references, the color associated with sheep could be black, but NEIL is able to scour the Internet for images and understand that sheep are typically white.
â€śImages are the best way to learn visual properties,â€ť Abhinav Gupta, assistant research professor in Carnegie Mellonâ€™s Robotics Institute, said in a press release. â€śImages also include a lot of common sense information about the world. People learn this by themselves and, with NEIL, we hope that computers will do so as well.â€ť
NEIL has been running since late July and has already analyzed three million images, identifying 1,500 types of objects in half a million images and 1,200 types of scenes in hundreds of thousands of images. The team says that NEIL has been able to connect the dots to learn 2,500 associations from thousands of instances.
Other projects, such as ImageNet and Visipedia, have tried to compile this structured data with human assistance. However, the scale of the Internet is vast, so the only way to help create the worldâ€™s largest visual structured knowledge base is to teach computers to do it by themselves.
While NEIL may be a stepping stone to a future with robots in control of the human race, NEIL still needs its human interaction from time to time to sort a few things out, particularly with pop culture. Abhinav Shrivastava, a Ph.D. student in robotics, says NEIL can make erroneous assumptions that compound mistakes, such as NEIL thinking that â€śpinkâ€ť is just the name of a singer rather than a color.
â€śPeople donâ€™t always know how or what to teach computers,â€ť he said in a press release. â€śBut humans are good at telling computers when they are wrong.â€ť
While researchers have directed NEIL as to what categories of objects and scenes to search and analyze, the diabolical computer program has surprised even the researchers. NEIL was able to determine by itself that apple is both a fruit and a computer company, and F-18 was a fighter jet, as well as a catamarans class.
As searches broaden, NEIL is able to create subcategories of objects and is also able to notice associations, such as the fact that zebras tend to be found in savannahs. The program, funded by the Office of Naval Research and Google, runs on two clusters of computers that include 200 processing cores.
Image Credit: Thinkstock