Neural Networks: Mark Humphries Discusses the Latest Discoveries Regarding Memory, Consciousness and Cognition

Photo by Daniel Hjalmarsson on Unsplash

I've been fascinated with how the brain works for as long as I can remember, keeping a "dream diary" from junior high till college, studying consciousness from a philosophical viewpoint in college half a century ago. During the course of a lifetime we have been repeatedly reminded that neuroscience has become remarkably advanced. For this reason, I reached out to a fellow Medium writer who specializes in this field. Not only are the topics engaging that he writes about, Dr. Mark Humphries writes about them in an engaging manner. (e.g.: How To Find Out If Your Brain Is a Computer, and end dumb arguments about metaphors.) The work he and his team are doing seems absolutely fascinating. I'm certain you will come away with the same impressions I have, that we are living in a compelling period in history.

EN: How does memory work?

Dr. Mark Humphries: Memory comes in many forms, and those forms seem to map onto different brain regions. One broad division of memory is into three: stuff that happened to you (episodic memory); stuff that you know (declarative memory); and stuff you know how to do (procedural memory). From research in psychology, we know a lot about how these kinds of memories can be manipulated in humans, and the time-scales at which they work.

But our interest lies in how the brain actually creates its representations, and uses them. And here that means understanding how neurons create, store, and recall memories. Unsurprisingly, we know much less about the nitty-gritty, because we can only capture the activity of a tiny handful of neurons at a time, relative to size of the number of neurons involved. And because we can only do this work in animals - it is tough to test memories of facts in a rat.

That said, we do have a beautiful theory of how neurons store and recall episodic memory. I give a full account of recall here and of making a memory here, but the basic idea of recalling a memory is simple. We recall an episodic memory based on some prompt - like the smell of baking bread reminding you of playing in your grandmother’s kitchen. We think the prompts coming from the world around you - the smells, the objects, the sounds - are fed into a bit of hippocampus with rather special wiring between its neurons. A specific memory is stored in that bit of hippocampus as the set of neurons that are active together; say neurons 120, 3045, and 10567 firing together mean “playing in grandmother’s kitchen as a child”. Those neurons are connected together.

The key is that the incoming prompt need only activate some of the neurons representing a single episodic memory. In our example, the neurons that represent the smell of the bread are only a few of the neurons representing the many things in that memory. So the baking smell turns on just a few of them. But, then, because those neurons are connected to all the other neurons that make up the complete memory, these other neurons are also activated - and the complete memory is recalled!

EN: Where does the “I” in self consciousness reside? In what way is consciousness connected to the brain? (For example, I have an out of body experience and see myself lying in the road after an accident.)

MH: The scientific study of consciousness waxes and wanes. While much of it remains controversial, what isn’t controversial is that we have no insight into how self-consciousness, or other forms of consciousness, arises from the firing of individual neurons. This is an ethical problem: generally speaking, the more confident we are that a species has a self-consciousness approaching ours in complexity, the less we will tolerate the kinds of experiments that record neurons in their brains.

The only way we can get a peek at brain activity in a human is via neuroimaging - fMRI, EEG etc. These are fantastic tools, but get nowhere close to neurons. fMRI is best, yet it records blood flow in volumes of tissue around neurons; and each little coloured speck in a fMRI image is somewhere around 100,000 neurons typically.

So at the moment, neuroscience is silent on where the “I” in self-conscious resides. But that experience of the “single” consciousness is made up of many elements - like memory, awareness, perception and so on. What animal studies and fMRI can show us is that brain activity related to those elements are distributed all over the brain, including in regions below cortex. So it seems likely “I” am spread over billions of neurons.

Anil Seth has just published an accessible introduction to the neuroscience of consciousness here: https://journals.sagepub.com/doi/full/10.1177/2398212818816019

EN: In technology we hear a lot of discussion about A.I. and “the Singularity.” What is the next “big thing” in Neuroscience?

MH: Neuroscience’s “singularity” moment will come when all the current explosive advances in technology are brought together. Right now we can record from exponentially more neurons than we could 10 years ago, and do it in awake, moving animals that are solving problems; we can trace the connections between neurons by tagging them each with a unique barcode of synthetic RNA, and finding out where that tag ends up as its gets passed along each neuron’s wires; and we can turn neurons on and off using light. Each alone is a really powerful bit of kit for advancing science.

The next big thing is then when all three of these collide: when we can trace the connections between the neurons we are recording from right now, and so work out which we want to turn on and off. With that, we can test all sorts of hypotheses that are impossible right now. Crucially, we will be able to finally reach that gold standard of science: causality. Did neuron A make neuron B fire? If we can record it and trace it and trigger it, then: yes.

EN: What was the biggest Neuroscience story of 2018?

MH: One that was totally outlandish came from Jason Shepherd’s lab back in January. As a mark of how potentially huge it was: I’m a theorist of how neurons give rise to behaviour, this work was on sub-cellular genetics, and it still blew my mind.

I’ve yet to fully digest it myself, but the rough gist is this: there is a gene called Arc that we know is involved in some types of learning, like learning a new skill. We know it’s involved because people can knock-out this specific gene from mice, and the mice cannot learn new movements. And the Arc gene often turns up in those big fishing expeditions for genes that are different in various disorders of brain development. We also know that the protein this gene makes -- the cunningly named Arc protein -- is involved in changing things at the synapse between two neurons, to make that synapse weaker or stronger. In short, this Arc gene is one part of the mechanics of how neurons change during learning.

Jason Shepherd’s lab gave us a totally new view of how this gene (and its protein) may drive learning. They showed that Arc creates a bag around its own messenger RNA (mRNA), the bit of RNA that encodes the instructions of how to make the Arc protein. And that bag is released from the neuron, letting the mRNA for Arc cross from one neuron into another. What that means is crazy: the instructions to make the Arc protein inside one neuron - to change its synapses - can be transmitted into other nearby neurons, and make the same protein there. Even if those nearby neurons have not switched on their own Arc gene.

It all suggests that we have a new, exciting avenue to explore for how brains learn: that neurons tell each other when they are changing their synapses, by passing around mRNA.

EN: How did you come to take an interest in neuroscience?

Design by Tara Austin. 

MH: I came at it via coding. Big into computers as a kid, I was interested in robots and AI. When deciding what to do at university I came across a degree called “Cognitive Science”, which married coding, neuroscience, psychology, AI, and a bit of philosophy chucked in for good measure. Sounded fantastic for someone who couldn’t decide what to do - just do everything. And I was lucky that the psychology department teaching on the degree had a really strong, deep systems neuroscience team - so we were taught neuroscience from year 1, and the deep dives into various brain systems (basal ganglia, cerebellum) and animal behaviour came as we got further in. I’ve been discussing theories for how the brain works at the level of neurons since the second year of my degree - over 20 years now!

EN: Finally, what are you currently working on that gets you excited?

MH: Like most scientists -- and certainly most theorists -- I’m most excited by what I’m about to work on, and deeply fed up with what I’m currently working on -- because the future project will of course not be full of all the dead-ends, backtracking, and mistakes in the current ones!

Still, we’ve got some cool stuff in the pipeline. One thing we’re looking at is what the prefrontal cortex knows about. The enigmatic bit of cortex that sits behind your forehead, its thought involved in learning about the complex facets of the world - what action leads to what outcome, and what needs to be held in memory in order to make sense of the world. We’ve been asking what populations of neurons in prefrontal cortex remember about what just happened when someone -- a rat in this case -- is learning a task to get reward. Specifically we’re interested if those neurons have to learn to remember -- if there are certain things that are only remembered when it becomes clear they are relevant to understanding how the world works.

The short answer so far is: the same group of prefrontal cortex neurons remember many things at the same time, some of which they can remember without learning. They remember those things from the very first time they happen (like, in our case, whether the rat chose to go right or left at a junction).

But learning only seems to happen once those memories get synchronized. The group of neurons synchronize their activity just before the rat’s “aha!” moment of learning. It’s as though the key to learning any complex action is to take the memories of the individual relevant things (“I turned right”, “I got reward”) and synchronized them (“I turned right *and* got reward”).

* * * *
Thank you Dr. Humphries. I will look forward to reading more of your writings in the year ahead. 

Humphries Lab is currently located in the School of Psychology at the University of Nottingham.

Key Links
Lab website: https://www.humphries-lab.org/
Neuroscience news for all: https://medium.com/the-spike
Twitter: @markdhumphries
Sleights of Mind: What the Neuroscience of Magic Reveals About Our Everyday Deceptions

Meantime, life goes on all around you. Engage it.