The media update team explores how artificial neural networks work, and why they are important within the AI and machine learning spectrum.

Neural networks follow a very different process from algorithms. According to MIT News Office writer Larry Hardesty, neural networks learn to do tasks by studying training examples. They consist of thousands of processing nodes that are tightly interconnected.

Neural networks are organised into layers of these nodes, and that the data moves through them in one direction only. One node is connected to a number of other nodes above and below it. Data will come from one direction, and this single node will then send the data along in the same direction.

Each node gives a ‘weight’ to its incoming connections, deciding which information is most valuable. Every node also gets its own ‘weighting’ over each connection, and then multiplies this by the ‘weight’ from incoming connections. It then adds the results together from all of its incoming connections and produces a single number, Hardesty explains.

Neural Network Diagram

If that number is below a set threshold, the node sends no data to the next layer. However, if it is above that threshold, the node sends the number – the sum of the weighted inputs – to all its outgoing connections.

“During training, a neural net[work] continually adjusts its internal setting in ways that even its creators can’t interpret,” says Hardesty, explaining how they differ from algorithms.

In a piece for Gizmodo, Jamie Condiffe, a lecturer at Imperial College London at Oxford University, says that neural networks will give everyone access to information that currently lies in the hands of very few. That, Condiffe says, will lead to smarter systems, better services, and more time to solve the human problems that computers “will never be able to fix”.

Want to stay up to date with the latest media news? Subscribe to our newsletter.

The use of neural networks is making artificially intelligent technology smarter and more capable of performing tasks in a way a human might. To test the ‘humanity’ of AI systems, the Turing Test is used. Find out how it works in our article, What is the Turing Test?