A while back, an intriguing photograph made an appearance on Reddit which showed a grotesque mutant: a shimmering, hydra-headed, slug-like being covered with liquefying animal faces.
Soon enough, the true origin of the image surfaced, in the form of a blog post made by a research team from Google. It turned out that the otherworldly picture was indeed, inhuman. It was the result of an artificial neural network (a computer brain) with an ability to recognise images.
Even the media picked up on the story, with reports in the Tech Times and PBS
Is there any connection between the psychedelic visuals of these pictures and Google’s neural network which created these images, and what do our brains do when faced with psychedelics?
Computers designed to mimic the human brain are known as Artificial Neural Networks (ANNs). They have been around since the early part of the 50s, but have made amazing progress in image recognition over the last few years.
The networks are composed of “neurons” that are software-based, which can communicate
They alter the strength of their connections in order to reflect the results of their calculations, similar to real neurons. What makes ANNs special is this adaptability. It provides them with an ability to learn.
The neural networks learn in a manner similar to children, by observing and imbibing the knowledge of the world around them. Usually, this data is put by people directly into the system.
For a neural network that is designed to recognise images, if it sees a hundred photos of cats, it shall begin to identify a cat on its own. The more photos of cats that it sees, the more efficient it will become. Should the network see a photograph of a cat-shaped thing, then a particular neuron in the network’s topmost layer will become distinctly activated, and the network will throw out the result: cat.
With such skills, ANNs have become vital for identifying faces and features in imagery, the sort of thing that the photo service from Google takes benefit of to create automatedfilms and albums.
The type of neural network used by Google is the convolutional neural network
This is what was used to create these odd images. It is composed of layers of neurons which send messages up through a command chain, understanding information with more features and abstraction as it goes upward, which means that each layer focuses only on one small task at a time.
As the network is self-teaching, what goes on exactly within each of those layers is still mostly a mystery. Google does not know which the exact pathways are that the information is moving through. It does not even fully know how the labour is divided and shared between the layers.
The experiment by Google was undertaken to split open these layers and observe what was happening inside
After examining the results, the researchers could quantify the accuracy of the machine’s knowledge.
The results were always not exactly on point. An example being, each image that was produced for “dumbbell” showed not just a metallic weight but also showed the muscular arm holding on to it. This gave an important insight: the computer had in most likelihood only ever seen a dumbbell which had an arm attached.
The complexity of the brain is such that artificial neural networks cannot be easily compared to it. People, for thousands of years, have turned their minds inside out with hallucinogenic substances, trying to get a clearer look at what we have learnt and what we still have to learn. The artificial brains from Google are a reminder that there is still plenty of research left to be done.
Keywords – artificial neural network, automated films and albums, psychedelic visuals, Google’s neural network
References:
https://www.reddit.com/r/WTF/comments/3a2bl6/supposedly_this_image_was_created_by_ai/
http://www.theverge.com/2015/6/4/8729943/laughing-and-crying-my-way-through-the-new-google-photos
https://research.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html