Google neural network art (Photo : Twitter)
Google and Facebook are both creating their own artificial intelligence (AI) neural networks that can imitate the human brain in many incredible ways. They have already passed the image recognition of real objects in digital photos, such as faces, cars, and buildings. The neural nets can now do tasks such as creating photo-perfect drawings or abstract art.
For example, today's AI software can recognize spoken words, translate between two languages, and even teach a robot to do human-like tasks, according to Wired. It can also do more than recognize pictures.
Facebook released updated information on Friday about its neural nets. They can now make small images of objects such as animals and airplanes that look like a real photograph in about 40 percent of cases.
Rob Fergus, a Facebook researcher, said that the AI robot can tell the difference between an artificial image such as abstract art, and a camera photo. It processes the image structures.
Google's approach to neural networks is taking an entirely different approach. It is using AI robots to morph photos of real objects into surreal images, according to Apex Tribute.
The tech giant explained in a blog post that the process involves learning and then improving known patterns. For example, if the robot recognizes that a cloud looks somewhat like a bird, it will enhance the bird's features.
The AI uses a feedback loop. After each pass of the image it will add details until a real-looking bird appears.
The ultimate goal of Facebook's and Google's neural nets is to improve technology so the AI resembles human intelligence more.
Google's AI has created images that are sometimes a little disturbing. They often appear surreal and the result of a mind trip, merging the images of pigs and snails, for example.
Still, researchers are thrilled about the potential of the neural networks shown by the experiments. They are nearing "unsupervised" machine learning that requires no human instruction.
