Thursday 19 May 2016 9:59 am

Here’s how your brain works on the Tube (and what it can teach AI)

Our brains work harder when we have to switch lines on the Underground, even if there are fewer stops than a longer journey which requires no changes, new research reveals.

Scientists have identified how our brains work when we're navigating the London Underground which could help artificial intelligence learn things more efficiently.

A London Underground-style map was used by researchers at Google's DeepMind and Oxford University, to analyse how we make plans.

Participants were asked to get from A to B while MRI scans of their brain were taken. These revealed in more detail than ever before which parts of the brain are involved in planning and making decisions.

Participants used the map in training, but were asked to plan the journey from memory during the MRI (Source: Balaguer et al./Neuron 2016)

The way we navigate the Tube demonstrates how the human brain uses a hierarchical method. This means we group tasks together, creating "layers" of actions that will help us reach our goal. In this case, we group together the lines rather than taking in every station individually.

The MRIs showed the areas in the brain linked to this type of decision making were the dorsal portion of the medial prefrontal cortex, which is known to support higher cognitive functions such as planning, and the premotor cortex, which is more involved in the execution of real or imaginary movements.

The parts of the brain working when we plan a Tube journey – top, premotor cortex, below medial prefrontal cortex (Source: Balaguer et al./Neuron 2016)

In contrast, artificial intelligence has access to huge amounts of computational power, and can work out all the possible actions that could help reach the goal.

However, the research could be used to train AI in a more efficient way of learning.

"The idea is basically to understand how humans or animals make long-term decisions," said Jan Balaguer, a PhD student at University of Oxford and member of Google DeepMind who was involved in the study, published in Neuron. "We're interested in trying to find machine-learning solutions to difficult tasks and real-life problems. Quite often it can be useful to draw inspiration from neuroscience.

"We want to see how the human brain implements things like hierarchical structures in order to design more clever algorithms. In machine learning, having a hierarchical representation for decision making might be helpful or harmful depending on whether you choose the right one to implement in the first place."

Hear the scientists talk about the research in more detail below.