Play

How do we know where we are, and where we are going? While we use our eyes to take in the information, it is our brains that are doing the heavy computing as we navigate the world. Now, new research using machine learning models is perhaps giving us a glimpse into what is going on inside our minds, and how the cells needed to navigate the world might form. 

On this episode of Approximately Correct, we are joined by two Amii Fellows and Canada CIFAR AI Chairs — Marlos C. Machado and Quinn Lee — who combined their expertise in machine learning and neuroscience to better understand the mysteries of navigation. 

“I am very interested in trying to use [AI] to understand intelligence," Machado tells host Alona Fyshe. 

"And of course,  biological beings are the ultimate proof of intelligence."

Loading...

To navigate our world, our brains develop specialized cells: place cells (which react to specific locations or landmarks) and grid cells (which tell us how far we have moved through a space). Machado says he was fascinated by a 2017 paper that used reinforcement learning to explore how these specialized cells develop and interact with one another. He was driven to know more. So Machado’s lab began their own experiment, building a model focused on navigation. He didn’t build a specific structure of cells — instead, he directed the model to learn how to navigate by feeding it visual scenes, similar to the information eyes pass to the brain. And he was excited by what he found.

“It was uncanny,” he says.The model developed place cells before grid cells. And not only that, but the model also began to develop neurons that mimicked other cells in the brain that represent other navigational features (e.g. agent head direction) , without being explicitly programmed to. That’s when Machado decided to team up with Lee, to see if the predictions made by the model could be compared against data from a biological brain. Comparing the model against data collected from mice, they found that the model had figured out how to map space in a way very similar to a biological brain. Their work suggests that temporal proximity — representing things that happen close in time in similar ways — could be an important general principle of how the brain develops, giving us more insight into our own minds.

Listen to the full episode to hear more from Lee and Machado about interdisciplinary AI, the bridge between neuroscience and machine learning, and the next steps for this exciting research. 

Approximately Correct: An AI Podcast from Amii is hosted by Alona Fyshe and Scott Lilwall. It is produced by Lynda Vang, with video production by Chris Onciul. Subscribe to the podcast on Apple Podcasts or Spotify.