What’s Old is New Again: The Evolutionary Context of the Neocortex
Our primary research interest at Numenta is the neocortex in mammals. We are on a mission to understand how the neocortex works, in order to understand intelligence in the brain. While we often talk about the essential principles of the neocortex, we don’t talk much about how it came to be.
The neocortex evolved and rapidly expanded in a very short amount of time (on evolutionary scales, at least). Yet it has an incredibly complex structure. So much so, that an understanding of the functionality of the full circuit eludes researchers to this day. How did this massive part of the brain, which accounts for 70% of the brain’s volume and is responsible for intelligence, develop so quickly?
To answer this question, we have to go back to a time before the neocortex. Long before the cortex existed, animals were able to navigate complex environments in the world. They could recognize environments from sensory cues, even with noisy and changing environments. They created robust and sophisticated mapping and navigation through space, a crucial capability required for all animals to find food, locate mates, and return to their nests and burrows. This problem of being able to track location and navigate environments is one that animals solved and perfected over many millions of years.
One of the big ideas that has come out of our research is that the mechanism that first evolved to allow animals to track their location exists in the neocortex as well. But in the neocortex, the mechanism is used for something slightly different. We hypothesize that the neocortex co-opted a location-based framework that originally modeled complex environments and repurposed it for learning complex tools and objects. When you look at the complexity of the neocortex, combined with the similarity of substructures to previous parts of the brain, it seems more likely that the neocortex borrowed and built upon an existing framework in the brain rather than build it from scratch.
Of particular interest to our current work at Numenta is learning through movement. We’ve been exploring how location mechanisms that are known to encode places in environments could be used to encode locations on objects. In both cases, sensory cues and self motion must be integrated into a model of spatial relationships. In the paper we are currently writing, we explore location mechanisms in more detail. Our theory proposes that everything we learn relies on location mechanisms, whether it’s an object and its behavior or a high-level concept. Our Co-founder Jeff Hawkins discussed some of these ideas in a recent Numenta On Intelligence podcast episode.
We’re excited about the ideas in the location-based framework, and believe they will have a significant impact on the field of neuroscience as well as AI. Whether our ideas are correct will play out over time. But perhaps the evolutionary history of the neocortex is an indicator that we’re on the right track.