The crystal ball in the head
Many neuroscientists believe that our brains are constantly predicting the future. Does this theory explain the basic workings of the mind?
In the summer of 2018, the Google company DeepMind presented new software. Using a few photos of objects, she creates three-dimensional representations of these objects. The program is based on a technique called "generative query network" (GQN), which not only inspires IT specialists. Neuroscientists have also kept an eye on it. They are particularly interested in the algorithm through which the program learns independently.
Roughly speaking, it works like this: From an image, GQN derives predictions about how the respective scene would look from other positions. Where are the objects located, how do their shadows fall, what is visible from different perspectives? The system uses differences between the predictions and the real input to continuously improve the accuracy of its predictions. The algorithm therefore changes the parameters of its prediction model in such a way that it deviates less from the real situation from time to time.
Many researchers suspect that our brain works in a very similar way. As such, it constantly creates models that describe what's going on in the outside world. From this it derives predictions about what will happen next, which in turn are compared with the sensory data. This is how ever better models of reality are created.
The number of neuroscientists who are convinced by this approach known as "predictive coding" is constantly growing. Some already declare it to be the comprehensive theory of how our gray matter works. "The brain adjusts its internal models in such a way that the prediction error shrinks," says Karl Friston. The brain researcher from University College London is considered a pioneer of "predictive coding". Most scientists share the basic thesis, according to which the brain constantly draws conclusions from sensory data and checks the predictions generated from them against reality. However, the empirical evidence for this has so far been scarce …