Neuroscientists at DeepMind, a Google-owned AI lab in London, have teamed up with academics at Oxford University and UCL to try and determine how the human brain navigates an underground train network.
The group — whose work was published in the journal Neuron this week — asked humans to plan a journey in a virtual subway network. Participants were tasked with getting from A to B while MRI scans of their brain were taken.
These scans showed which parts of the brain are involved in planning and making decisions.
The group, which included Google DeepMind CEO Demis Hassabis, concluded that the brain splits the task of completing a journey into different jobs, with different parts of the brain handling different elements of the task. Parts of the cortex, for example, showed more activity when the participant had to change lines, while other parts of the brain appeared “more excited” as the participant got closer to the final destination.
Studying the human brain like this can help companies like Google DeepMind to make advances in the field of artificial intelligence.
Developers can create AIs that can calculate all the possible consequences of a single action but humans have to create and execute a plan, the paper states. We mentally invent different “layers” to organise our actions and then think about the bigger picture rather than individual steps, according to the study.
“We’re interested in trying to find machine-learning solutions to difficult tasks and real-life problems,” said Google DeepMind scientist Jan Balaguer, who is also completing a PhD at Oxford University, in a statement. “Quite often it can be useful to draw inspiration from neuroscience.”
Baluger compared the group’s work to previous studies, claiming it shows in a more straightforward and direct manner that there are hierarchical representations reflected in the brain.
“We want to see how the human brain implements things like hierarchical structures in order to design more clever algorithms,” he continued. “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.”
As one might expect, humans took longer to plan a journey with several line changes than they did to plan journeys that involved a high number of station stops on the same line.
DeepMind, which employs around 250 people in King’s Cross, is famous for creating an AI agent called AlphaGo that successfully took on the best player in the world at ancient Chinese board game Go last month.
But the company, which Google acquired for a reported £400 million in January 2014, has a wide range of other projects on the go that it doesn’t reveal to the public, including a little-known fashion website called KITSEE that was killed off around the time of the Google acquisition.
DeepMind is also looking at how its technology can be applied in healthcare, announcing a partnership with the NHS in February that has proved to be controversial due to an extensive data-sharing agreement.