How do machines actually learn?

Chris put this question to Peter Clarke from Resurgo Genetics...

Peter - There are similarities. The way you can imagine your brain working is there’s a bunch of neurons, maybe about 100 billion neurons in your brain, and they’re all wired up. Each one of them is taking in inputs from another neuron and deciding how to behave and firing out its message, which is then being input to all these other neurons. Really this is a fairly simple core underlying structure behind all of the brains in nature but it’s really how all of these neurons work together. You have this property in nature which is called “emergence” where a lot of simpler things interacting with each other lead to high level, more complex behaviour and you see this over and over again.

One of the best places to see that is in a bees nest or something like that where each individual bee is behaving in a relatively simple way, but together they manage to create these huge hives which can think in far more sophisticated ways than any one of the individual bees can. The same process is happening inside your brain and across you body and all sorts of other ways. These are called “connectionist models” so  models of how simpler components can come together and, through their interactions, lead to a higher level, more complex behaviour. Our brain works like that, and these artificial neural networks work in exactly the same way, by the same process.

There are a bunch of aspects in which they’re not the same. We really don’t understand all sorts of aspects of how human brains work, and we also don’t understand how cells work. So, we can’t really say, in a very precise way, they work in the same way as the way humans learn but they do use very, very similar processes.