Computers can now “dream” – or so say software engineers from Google. Apparently, they dream of strange creatures: dog-fish, camel-bird, pig-snail, you name it. Now, doesn’t this sound totally crazy?!
AI dreams: A strange carnival with automobile-animal hybrids. Photo credits: Michael Tyka/ Google.
When the ability of the servers of Google to identify and create images of objects like bananas and measuring cups was tested by the researchers, it was found that the artificial network created its own images from keywords.
Google’s AI coming up with its own images using keywords. Photo credits: Michael Tyka/Google.
Google’s artificial neural network is compared to a computer brain – similar to the central nervous system of animals. When the network comes across images, its first “neurones” view them. These then relay the information to the next layer of neurones which then process the images. This is repeated 10 to 30 times, as each layer spots main features and isolates them to “understand” the images. Thereafter, the neural network “responds” to what it has been fed with and gives out a representation of what it has identified the images to be.
The researchers then aimed is to find out how effective are computers at learning. They reversed the process: they provided the AI an object and asked it to produce an image thereof. The computer did so by using specific features it spotted in the image.
The computer does not always succeed at making out the images though. Furthermore, it seems to have an imagination of its own. It “sees” objects in clouds. For instance, when fed the image of a cloudy sky, it interpreted it as highly-defined objects:-
The processed image of a cloudy sky. Photo credits: Michael Tyka/Google.
Close-up of the “creatures” the AI found in clouds. Photo credits: Michael Tyka/Google.