Program with brains
Artificial neural networks work with a mechanism that is not available to the human brain. In order to unravel the mysteries of learning, researchers are looking for biologically plausible alternatives.
Deep learning is now an integral part of artificial intelligence (AI) applications: we owe it to him that Siri understands spoken commands or that computers create works of art on their own. But that was not always so. When some computer scientists proposed a workshop on deep neural networks at a renowned conference on artificial intelligence in 2007, the organizers promptly rejected it. They did not want to give too much space to the marginal area, which was ridiculed at the time.
The interested parties then organized an unofficial meeting, which was attended, among others, by the cognitive psychologist and computer scientist Geoffrey Hinton from the University of Toronto. He has contributed to some of the greatest breakthroughs in deep neural networks. He began his talk by joking: "About a year ago I came home for dinner and said, 'I think I finally figured out how the brain works' - to which my 15-year-old daughter sighed, 'Oh, Dad, not again.‹” The audience laughed. Behind the joke, however, was a serious goal: He wants to use AI to understand the brain.
Today, deep neural networks play a central role in AI research. In large part, they owe this to a particular algorithm called "backpropagation" (which roughly translates to "error feedback"). It allows programs to learn from sample data. The computational process allows computers to classify images, understand and translate speech, recognize the environment for self-driving cars, and perform numerous other tasks. However, experts consider it highly unlikely that our brain processes information according to a pattern similar to that of the algorithm…
