Overlay
 
Digital World

The best connections

28.05.2019
 

What primordial animals of the universe do we see in this picture? A little hint: what the picture shows is actually all too human. It's surely one of the best networks in existence. But be careful! It has nothing to do with social media.

Networked information is as old as humans themselves. After all, our brain consists of up to 100 billion nerve cells, the neurons, which convert signals from our sensory inputs – such as sound waves – into impulses and ultimately into knowledge. All neurons are connected to each other via synapses and are constantly firing information back and forth in our heads using chemical messengers, known as neurotransmitters. This is how we humans learn; and we do this in every situation, at every age and in every circumstance. All neurons have the same structure and an input and an output for information. This means each neuron can act as transmitter or receiver. Over the course of our lives, a lot of knowledge comes together in the folds and wrinkles of our grey matter. Neurons shape our memories, our beliefs, our emotions and our cognition, ultimately our entire identity.

It is precisely these processes in the human brain that artificial neural networks seek to reproduce. They are based on computer programs that draw logical conclusions from a wealth of data. Linking these data into meaningful units allows them to process information. It's called deep learning and is also based on neurons – in the form of programmed functions. An artificial neuron receives input from several other neurons in the form of numbers and weights them according to their relevance. These assessments can be changed by training the neurons. The artificial neural network, in turn, consists of numerous such functions layered on top of one another. Artificial intelligence only emerges when an extremely large number of neurons is connected, because the individual units can do very little. Would you like some examples? In the Chinese board game Go, black and white pieces are placed alternately on the playing field. It is considered much more complex even than chess. Back in 2015, the AlphaGo AI program beat several world-class top players with an artificial neural network and a lot of practice in the form of deep learning. A short while later, the Alpha Go Zero AI program defeated its predecessor. This program did not train against humans, it only calculated moves against itself in advance. Apple's voice assistant Siri also works on the basis of such an AI, and self-driving cars learn to steer autonomously using artificial neural networks. There are hardly any limits to possible areas of application in the future: from image, text, face and error recognition to medical diagnostics, early warning systems and machine translations.  

Newsmail

Subscribe to our email newsletter and never miss anything again!

 

Share article via email

* = Required field