Any experienced traveler knows that planning is key if you want to avoid missed flights, getting lost, or wasting time. We rely on our brains to help us navigate with maps and plan routes, efficiently deciding which transit stops will get us where we want to go. A computer could do this by calculating all possible consequences of a single move along a route, but unfortunately, our brains don’t have the same kind of raw computational power. Instead, we invent mental layers to plan, paying the most attention to the higher levels than individual steps, according to researchers from Google DeepMind and the University of Oxford.

“The idea is basically to understand how humans or animals make long-term decisions,” said Jan Balaguer, a Ph.D. student at Oxford and member of Google DeepMind, in a statement. “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.”

Balaguer and his colleagues created a map for a fictional subway to use as a proxy in decoding the brain’s sequential decision-making process. The researchers represented a single step in the process with an individual station, while different-colored subway lines represented a higher-level layer. Twenty-two study participants looked at the map during training, then received a destination station as a goal. While they figured out how to get there, researchers imaged their brains with fMRI.

The team looked specifically at how participants focused their attention during the task; whether on individual stations or subway lines as a whole. The participants’ brain activity and response time generally increased with the number of line changes between the person and the destination station, rather than with the number of individual stations between them. The researchers linked this type of decision-making to the dorsal part of the medial prefrontal cortex, known to help with higher cognitive functions including preparation and planning. They also noted activity in the premotor cortex, which is usually more involved in the execution of real or imaginary actions.

“We show, in a more straightforward and direct manner than previous studies, that there are hierarchical representations reflected in the brain,” Balaguer said.

In addition to the mental layers, the research shows there are some brain regions that are more active as participants come closer to achieving their goals. Studies have previously demonstrated the hippocampus as reactive to proximity to a goal, and Balaguer’s study supports this. In addition, the researchers identified the ventromedial prefrontal cortex as behaving similarly.

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

Source: Balaguer J, et al. Neural Mechanisms of Hierarchical Planning in a Virtual Subway Network. Neuron. 2016.

Published by