Grid cells are among the coolest things at the intersection of neuroscience and AI.
Short story: neuroscientists have been wondering about how the brain represents location. In the 1970’s, neuroscientists found so-called place cells – much like with Bill Clinton or Jennifer Anniston (whoever she is), there are neurons dedicated to particular places and fire only when you get to those places.
In 2005, the Moser lab in Norway found another type of neurons responsible for representing position. But unlike place cells, they fire in grid-like patterns (were they responsible for Mohenjo Daro?).
At first, I was actually quite confused about the level at which the “gridiness” arises. I originally parsed “rectangular firing patterns” as meaning that the neurons that fire in a given location are form a rectangular pattern through their connections.
I don’t know why they’re not called grid neurons. Anyway, they are neurons in a region adjacent to the hippocampus called the entorhinal cortex.
It’s quite beautiful.
But the part that I find the most interesting is that a good number of neuroscientists (including Jeff Hawkins and Tim Behrens) think that our navigation system is much more important than I thought.
The hypothesis (that I need to read up on more) is that a lot of our higher-level cognition (conceptual thinking in particular) is done by repurposed navigation and movement.
That makes a lot of introspective sense to me.
What’s even cooler is that Demis Hassabis mentioned grid cells as an example of taking inspiration from systems neuroscience back in 2010. From his talk:
Now, much more recently, what was found by the Mosers in Norway was a bunch of cells called grid cells. These were found in the entorhinal cortex, which is next to the hippocampus and provide input into the hippocampus. These are neurons with the spectacular regular periodicity, and they form hexagonal firing patterns. I just showed you place cells that had a single firing pattern, but here you’ve got hexagonal firing patterns. Let’s just look at this final column on the right here. It’s the only one that’s relevant for this, which is these are three different cells. And in this case, the rat is in a circular environment rather than a square one. You can see that this particular single cell is firing at these regular hexagonal places.
It’s incredible. When this was first found in 2005, no one believed it, because it looks non-biological. It looks like someone’s coded this, and this is what the brain uses to mark out space. It seems like this is what the brain is using to define an intrinsic measure for space. What you could think of it as is like the graph paper of the mind. It’s something that’s automatically tessellating space. We’re doing this right now. We’re looking at this room. Several new navigational systems, including several current DARPA projects, are using these two types of cells to build new kinds of navigation techniques.
Okay, so it provides direction, and we’ve seen that inspires new algorithms. So what about validation testing? Well, what do I mean by that? All this is is really, we’d like to answer the questions of this form: does an algorithm that we may have invented through machine learning or some other way, or through maths, constitute a viable component of an AGI system? Does this algorithm that I have sitting here on my machine, that does some interesting things, is that a useful component of an AGI system?