A computer is hoping to learn something that much of humankind arguably has trouble with - common sense.
The fantastically-named Never Ending Image Learner (NEIL) analyses images 24 hours a day in an attempt to learn what links different things.
NEIL - who uses the tagline "I crawl, I see, I learn" - has been running since 15 July and has analysed three million images and learned 3000 common sense relationships.
Ever so clever
Abhinav Gupta, assistant research professor in Carnegie Mellon's Robotics Institute in the US, said: "Images are the best way to learn visual properties.
"[They] 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."
Although computers can already identify objects, NEIL is adding an extra level of analysis by making common sense links.
So far NEIL has made a number of rather endearing discoveries.
It knows for example:
- A rifle can be a kind of / look similar to Ak_47
- An Airbus_330 can be a kind of / look similar to Airbus_340
- Leaning_tower can be found in Pisa
- Trading_floor can be / can have Crowded
- Zebra can be found in Savanna
You can even submit your own category submissions here.
It doesn't always get it right though. Searching for terms with two meanings can flummox it but a team is on hand to provide assistance.
Abhinav Shrivastava, a PhD student working on the project, said: "People don't always know how or what to teach computers.
"But humans are good at telling computers when they are wrong."
The project is being funded by the US Department of Defense's Office of Naval Research and Google.