Can you teach common sense – even to a computer?

So it's interesting to see US scientists trying to teach a computer to apply common sense. They want to determine whether computers can learn what links images, in the same way a human would. The goal is to help computers to understand the visual world. The approach is called the Never Ending Image Learner, or NEIL. Those with children will be familiar with the process of sitting down with them and pointing to a picture of a cat, for example, and saying 'cat'. Computers have the ability to do this rather more quickly; in this instance, 200 processing cores are being applied full time to the project. Since July 2013, NEIL has scanned more than 5million images from the internet, from which it has labelled 500,000 of them and made 3000 common sense relationships. For example, NEIL has determined – all by itself, apparently – that cars are found on roads, ducks resemble geese and deers look like antelope. Needless to say, NEIL will not be able to learn independently. But it appears the researchers will only correct the program when it's wrong, rather than try to tell it what to learn. When it is wrong, however, it tends to be amusing. For example, it determined that 'jellyfish can be part of a desktop computer'. Where are the applications? Perhaps the presence of the US Office of Naval Research as a funding body gives us a clue. But you can see value in being able to tell a system to watch a video stream and 'tell me when you see this'.