What's a neural network?

We put it to Beth Singler, AI expert from the University of Cambridge...

Beth - A neural network is basically a layering of algorithms, that are all in connection with each other, so that the output of one algorithm then becomes the input on another layer. It's a connectionist approach. And the aim of the neural network is to sort of replicate the architecture of the human brain. So the neurons and the synapses that we're familiar with. So the thinking is that if everything is computable, a big if, but some think proven by Alan Turing after Alonzo Church, then the substrate, what the brain is actually built of, whether it's fleshy meat stuff, or Silicon might not actually matter. But what might matter is the architecture of the human brain, the connections between the synapses and the neurons and how they interact and flow to each other. And this is what neural networks are trying to replicate. Now, there are many uses of neural networks, including, basically assessing complex data relationships and doing pattern recognition. As we see in computer vision, in the control of systems like automated cars and in artificial assistants that can learn how to respond to us and learn our preferences. A current interest in neural networks is in using it to recognise and diagnose COVID-19 in patients' lungs' MRIs. However, neural networks can sometimes learn to look for the wrong things in data, or they can express biases that all were already present in the dataset. And this can lead to very negative or unexpected outcomes.