Do androids, as science fiction writer Philip Kindred Dick have suggested, really think of an electric sheep in their dreams? Scientists have been debating the meaning and purpose of dreams for many years. Today, another step has been taken towards understanding what humans dream about - and how robots can mimic dreams.
In 2013, the neuroscientist Yukiyasu Kamitani conducted an experiment: people in an MRI were repeatedly woken up as soon as they started to fall asleep and asked to describe their dreams. Kamitani had previously identified several unique patterns of brain activity, depending on which objects were shown to the "test subject" when they were awake. While people slept, their brains were scanned to detect these patterns, and a computer program converted the dreams into short videos. The study showed that the content of dreams coincided by 70 percent with what people themselves remembered about them.
Two years later, engineers at Google received pictures of "computer dreams." They ran millions of images through a program that mimics the neural networks of the human brain to understand how artificial intelligence learns to recognize objects.
And then they expanded the capabilities of this intelligence using the DeepDream program, and the artificial neural network created its own dream pictures, finding and amplifying certain shapes in a random set of video signals. The result is psychedelic landscapes. Just like in human dreams, the images previously seen by the artificial web have taken on a new form.
True, it is unlikely that it will be possible to accurately reproduce our dreams until scientists figure out exactly how the brain creates them, or they do not reveal all the patterns of brain activity, says Jack Gallente, a psychologist at the University of California (Berkeley).