The US Department of Defence (USDoD) has developed a computer program which it hopes will develop ‘common sense’ in the same way that humans do.
The target of the NEIL (Never Ending Image Learner) project is to see whether computers are able to comprehend links between images to develop a better understanding of the visual world.
The scheme, which has been running since July, has so far seen three million images analysed. Of these, 1,500 objects have been identified from half a million.
The program has also made almost 2,500 associations, which include:
“‘Airbus_330’ can be a kind of / look similar to ‘aeroplane'”
“‘Car’ can have a part, “‘wheel'”
“‘Zebra'” can be found in “‘Savanna'”
There are already computer programs that are able to identify objects using hardware and software; this is known as computer vision. However, the NEIL team are hoping the program will eventually learn the relationships without being taught this way.
Talking about the project, assistant research professor Abhinav Gupta said:
“Images are the best way to learn visual properties.
“People learn this by themselves and, with NEIL, we hope that computers will do so as well.”
If successful, the technology could be very wide-reaching, and could eventually be used to write articles and web pages which make sense.
NEIL is a joint project of the USDoD’s Office of Naval Research and internet giant Google.