Google's Inceptionism Lets Us Look at an Artificial Intelligence Hallucination

Artificial Intelligence  

Did you ever look up at the sky and try to imagine the clouds as shapes or animals? A team at Google has programmed their neural networks to do something similar while recognizing images, with some intriguing results.

Google’s image recognition software can detect, analyze, and even auto-caption images by using artificial neural networks to simulate the human brain. Now, in a process they’re calling “inceptionism,” Google engineers tried to see if they could find out what these artificial intelligences “dream” of.

Google trained the neural network by feeding it millions of images, eventually teaching it to recognize specific objects within a picture. When the network is presented an image, algorithms process the image, trying to emphasize the object in the image that it recognizes.

Related articles Researchers Develop Neuromorphic Chip With Cognitive Abilities  Stanford Researchers and Google Create World's Largest Artificial Neural Network  Jason Silva Looks At Artificial Intelligence

The process uses many layers—or deep neural networks, to generate greater and greater precision. The final output layer, the network makes a final interpretation of the image.

The network typically consisted of 10-30 stacked layers of artificial neurons. Each image is fed into the input layer, which then talks to the next layer, until eventually the “output” layer is reached. The network’s “answer” comes from this final output layer.

It is almost like the system is playing  visual game of twenty questions, driving at a solution.



"We know that after training, each layer progressively extracts higher and higher-level features of the image, until the final layer essentially makes a decision on what the image shows," write the researchers.

"Neural networks could become a tool for artists—a new way to remix visual concepts—or perhaps even shed a little light on the roots of the creative process in general."

"For example, the first layer maybe looks for edges or corners. Intermediate layers interpret the basic features to look for overall shapes or components, like a door or a leaf. The final few layers assemble those into complete interpretations—these neurons activate in response to very complex things such as entire buildings or trees."

For this research, the team asked the network: “Whatever you see there, I want more of it!” This created a feedback loop: if a cloud looks a little bit like a dog or bird, the network would make it look more like a dog or a bird.

"We call this technique “Inceptionism” in reference to the neural net architecture used," write the researchers.

This in turn made the network recognize the bird even more strongly on the next pass and so forth, until a highly detailed bird appears, seemingly out of nowhere.

The results are very interesting.  Google has shown that even a relatively simple neural network can over-interpret an image, just like as children we enjoyed watching clouds and interpreting the random shapes. 

This network was trained mostly on images of animals, so naturally it tends to interpret shapes as animals. But because the data is stored at such a high abstraction, the results are an interesting remix of these learned features.

The researchers say the work has helped them understand and visualize how neural networks are able to carry out difficult classification tasks, improve network architecture, and check what the network has learned during training. "It also makes us wonder whether neural networks could become a tool for artists—a new way to remix visual concepts—or perhaps even shed a little light on the roots of the creative process in general."

What do you think?  Are these the dreams of artificial intelligence, a window onto the creative process, or just another filter for Photoshop? For more examples, check out the Inceptionism gallery for more examples.


SOURCE  Google Research

By 33rd Square

Embed