Machine learning may never outperform the human brain, but it can still shape the world around it based on what we want it to see.

The brain isn’t uniform. It’s complicated, and is part of a complex system of functions that help us walk to the shop to buy bread. It plays its significant part in helping us model the world around us so that we recognise what a shop is, how we get there, and what kind of bread we like the taste of. From these sensory motor models come reference frames built from our interactions with the world. For instance, close your eyes, point your finger and slowly move it forward until it touches something. With your eyes still closed, run your finger around the object until you’ve identified it.

The processing power of the neocortex is incredible. In those few seconds of blind exploration, you drew on the capacity of the brain to reference models we have shaped, and have been shaping, since birth (and before). Prior to identifying the object, you instructed parts of your body to cooperate, such as your eyes, your shoulder and your finger. Complex algorithms determined the texture of the object, its shape, and accessed reference models to arrive at a conclusion as to what the object was.

Before your finger arrived at the object, your brain was predicting what it was going to feel. It’s already predicting an outcome based on reference frames of where things are in the world relative to other things. Now imagine being able to create reference frames for machines, which then go on to draw on these points of reference and deduce a new path to take based on fresh evidence.

The power of circuitry

It’s so easy to be sucked into using words like deduce, when we know machines can’t actually deduce anything. But they can be trained to recognise patterns based on previous input, and predict many outcomes. They can also work beyond the confines of rules-based patterns by using complex algorithms that recognise where things are missing in data. Effective algorithms can “deduce” that what it’s looking at is a purchase invoice with some missing information. It references previous examples of purchase invoices to determine what the missing items might be. If it’s unable to fill in the gaps, it can send the invoice to the right department (among many) based on other information in that document.

We can emulate the principles of how the human brain works directly in machines, but machines don’t have motivation or desire, so an architecture of knowledge has to exist. Machine learning is undoubtedly what it says: a learning machine. It will take our input, our data, our knowledge, and work out patterns based on reference frames of what it has seen before. Through the power of circuitry, it will model its world, even if that world is based on boring purchase invoices!

The benefit to us as people who work within machine learning is so rewarding. We get to see businesses increase productivity, make more money, and make people happy. Our complicated brains, packed full of neurons, get to spend time on more productive uses of those things that make us unique, such as creativity, desire, and the taste of bread.