Technology just got a little weirder.
Last month, Google debuted a new software called Deep Dream, which is intended to be used for image recognition purposes. A form of artificial intelligence (aka it’s capable of learning), the technology utilizes artificial neural networks and is based on the actual structure of biological brains. Essentially, to perfect the software’s image recognition, it’s fed millions of images and trained it to filter the images, layer by layer, in order to produce a final solution.
The AI’s artificial neurons have specific functions: The lowest levels interpret an image’s basic features, such as outlines and edges, while the upper levels look for more detailed features. The job of the last layer is to make the final call and decide what, exactly, the object in question actually is.
Through the ability to recognize images, this software can now also produce images based on keywords entered into the system—the results of which have ranged from funny to downright terrifying.
And this isn’t a technology Google is keeping to itself. The AI code was released on Github, and programmers across the web have been getting crazy and creative generating imagery with the code, including this fascinating (and totally terrifying) interpretation of a psychedelic-fueled scene from Fear and Loathing in Las Vegas.
The applications for image recognition reach far beyond the creation of creepy GIFs and puppy pizza, however (though that’s kind of a noble pursuit in itself). As it turns out, these wild images are actually incredibly useful and provide researchers with a baseline for what the software recognizes, allowing them to work with the artificial intelligence and help it pick up patterns and “learn” until they’re satisfied with the results.
“As computers begin to interpret what they see and to understand the context of what is actually in images, we can start to transition from complicated methods of data collection to simple snapshots taken with a mobile camera,” said Chaotic Moon CEO Ben Lamm.
“The potential applications of image recognition are even greater when we consider improvements to mobile imaging,” Lamm continued, “such as improved resolution of mobile cameras, IR blasters collecting point data, and a larger repository of online tools.”
Here at CM Studios, we’re busy implementing image recognition technology for some particularly interesting (and hush-hush) endeavors.
“We’re currently using images as predictive data in order to foresee potential issues in areas such as medicine, transportation and environmental studies,” Lamm said. “Improvements in image recognition technology present us with more opportunities than ever to present future-forward solutions that constantly keep our clients and partners ahead of the curve.”