🧠Your Brain Is Lying to You (And That's Why You're Still Alive)

📖 Read the full essay

I've been thinking about coffee shops lately. Not in the precious, writerly way where I romanticize the smell of roasted beans and the clatter of ceramic cups. I mean the moment right before you walk in—that electrical jolt when you round the corner and see the familiar sign glowing in the window. That tiny spike of pleasure that happens before you've tasted anything, before the caffeine has touched your bloodstream, before the reward has actually arrived.

That feeling? It's your brain time traveling. And according to new research from McGill University, it might be the most important thing your brain does.

The research discussed here is from "Predictive Coding of Reward in the Hippocampus" by Mohamed Yagubi and colleagues, published in Nature. For those interested in the technical details, the paper provides remarkable evidence for how biological neural networks implement reinforcement learning at the cellular level—a finding that bridges neuroscience, psychology, and artificial intelligence in profound ways.

Predictive Coding of Reward in the Hippocampus

This is Heliox: Where Evidence Meets Empathy

Independent, moderated, timely, deep, gentle, clinical, global, and community conversations about things that matter.  Breathe Easy, we go deep and lightly surface the big ideas.

Transcript

0:25

Welcome back to the Deep Dive. Today we are going to try and do something that, frankly, sounds physically impossible. We are going to watch a thought travel backward in time. It really does sound like science fiction, doesn't it? Like something out of a movie. It sounds completely made up. But the paper we're going to be covering today, and this is a dense, absolutely fascinating piece of work, it suggests that this time travel isn't a gimmick. It's actually the fundamental mechanic of how your brain learns. It's how you learn, how you build goals, how you navigate the world. And in the process, it might just force us to completely rewrite what we think memory even is. Right. I mean, we all tend to think of memory as a record, you know. It's a library of the past, a kind of perfect video recording. Exactly. I remember where I left my keys. I remember the plot of that movie I saw last week. It's an archive. It's historical fact. But the research we're diving into today argues that the brain isn't an archivist at all. It's a fortune teller. A prediction machine. A prediction machine. It uses the past, sure, but only as raw material. It's just training data to construct a better, more accurate simulation of the future. what is the hippocampus actually doing? A question we thought we had a pretty good answer to for a while. Right. Most people, if they've heard of it, know the hippocampus as the memory center or maybe the brain's GPS. Exactly. Since the 1970s, the dogma has been that the hippocampus creates your cognitive map. It tells you where you are in space. John O'Keefe won a Nobel Prize for discovering place cells, which are the foundation of that theory. does the map itself change based on what you want, based on what you value in that environment? So in other words, they wanted to see if the brain maps not just where we are, but what good things are about to happen at those locations. And the answer they found is, well, it's why we're doing this deep dive. It's genuinely fascinating. Okay. Before we get into the really heavy lifting, and I want to warn you, there is some heavy lifting today. We're going to be talking about calcium imaging, AI algorithms, even a little bit of mouse brain surgery. Yeah, it's just a little. I want to start with a feeling, a really specific human feeling. Okay, I'm with you. The coffee shop feeling. Oh, no. Well, too well, probably. So picture this. You're walking to your favorite coffee shop. You've had a long morning. You're tired. You just, you need that caffeine hit. Yep, been there. You walk down the street, you turn that familiar corner, and then you see it. The sign. That little neon sign in the window. And you feel that little lift, right? That little jolt of something. Exactly. That's the thing. You aren't drinking the coffee yet. The caffeine is nowhere near your bloodstream. But the moment you see the sign, or maybe even the moment your hand touches the cold metal of the door handle, your brain goes, Huh, yes. Here we are. Good stuff is coming. You get the reward before you actually get the reward. And that right there, that is the phenomenon in a nutshell. That's what this paper is all about. But the question is, what is actually happening in your neurons at that moment? Is your brain just, you know, generically excited? Or is it doing something far more structured, more specific? And what Yagubi's paper shows, with incredible precision, is that the very neurons responsible for the feeling of enjoying the coffee, the reward neurons, they don't just stay put. Over time, as you learn that route to the coffee shop, those specific neurons physically shift their activity backward in time Wait, say that again? They shift their activity? They do. They stop firing when you're taking a sip of the coffee and they start firing when you see the sign. They literally move from the consequence to the cause. You propagate backward. Yeah. And this back propagation, as they call it, seems to be the key to how our brains learn causal relationships in the world. But to really, really get it, you can't just talk about coffee shops. We have to look at the experiment itself. Right. Because this is not a survey of coffee drinkers. Yeah. This is some of the most high-tech neuroscience happening today. Okay, let's set the stage then. Section one, the setup. an iPad for mice, and a window into the brain. Because this whole experiment, it relies on technology that, frankly, didn't even exist a few decades ago. Oh, absolutely not. This is truly cutting-edge stuff. And I think to understand the data, you really have to visualize what the researchers are looking at. If you look at the paper, specifically Figure 1b, There's this image. I'm looking at it right now. It looks like a window into deep space, doesn't it? A star field. That's exactly what I was thinking. It's a galaxy. It's a black void just filled with these flashing, twinkling, little green stars. But we aren't looking at stars, are we? We're looking at a mouse's brain. We're looking at the CA1 region of the dorsal hippocampus, to be precise. And each one of those stars is a single neuron firing. A single brain cell lighting up in real time. How? How do they even see that? You can't just point a camera at a brain and see neurons firing. I mean, brains are, they're gray and mushy, they don't glow. That is very true. Normally they don't. To get them to glow, you have to do something incredibly clever, a technique called calcium imaging. Okay, let's unpack this for a minute because this tool, it really feels like it makes this whole study possible. It's the foundation of it all. So we've known for a long time that neurons communicate using electricity, these little spikes called action potentials. Right. But you can't see electricity with a normal microscope. It's invisible. However, biology gives us this amazing little loophole. When a neuron fires that electrical signal, It opens up tiny channels in its outer wall, and a flood of calcium ions rushes inside the cell. Okay, so a firing neuron is a neuron that's suddenly full of calcium. Exactly. So what the researchers do, and this is the genetic engineering part of it, is they inject a specially designed virus into the mouse's brain. A virus. That sounds... bad. Well, it sounds scarier than it is. It's a viral vector. Think of it like a harmless delivery truck. And this virus is carrying a very specific piece of genetic code for a protein called GCAMMP6. Catchy name. I know, right? But here's what GCAMMP does, and it's brilliant. It But when calcium rushes in because the neuron just fired, the G-Campi protein binds to that calcium and it fluoresces. It glows. It glows bright green. So you've essentially managed to turn an invisible electrical signal into a visible light signal. Precisely. You're turning the process of thought into flashes of light. And then to actually see that light, they mount a tiny, tiny microscope right on the mouse's head. And this is the famous miniscope. The UCLA miniscope. It's an amazing invention. It weighs about three grams. It's like a tiny GoPro for the brain. And that part is so crucial, isn't it? because in the old days if you wanted to record from a brain the animal had to be completely stuck in a frame head fixed right head fix unable to move which is fine if you're studying say vision you can just show it pictures on a screen but if you want to study the hit a campus the GPS the navigation center the animal needs to navigate it has to move it needs to run and turn and explore the miniscope is revolutionary because it lets the mouse wear the camera like a little hat and move around freely, behaving naturally. So we have our subject, a mouse wearing a microscope hat with a brain literally full of flashing green stars. Where do they put him? What's the task? They put him in what I like to call the office. It's an automated touchscreen chamber. If you can, imagine a box, but it's shaped like a trapezoid. A trapezoid? Wow, a trapezoid. That seems like an odd choice. It's actually a very smart design. It helps the mouse orient itself. If the box was a perfect square, every corner would look identical. Ah, I see. In a trapezoid, the geometry itself is a landmark. The long wall is different from the short wall. It gives the brain's internal map some stable cues to lock onto. Got it. Okay, so what's in the box? On one of the walls, there's a touchscreen. It really looks like a little tablet for mice. And then on the opposite wall, all the way across the chamber, is a reward port. It's a little hole where the good stuff comes out. And the good stuff is strawberry milkshake. Specifically strawberry milkshake, it's right there in the methods section of the paper. They use the Ensure brand strawberry flavor.- I love that some of the most advanced neuroscience on the planet runs on the same stuff you'd give your grandma.- Hey, it's highly palatable. You need the mouse to want to do the work. If you just use water, they do a few trials and then get bored. They love the sugar. It's a powerful motivator.- So the mouse is in this trapezoid, motivated by milkshake. What's the job? What does it have to do? The task is called TUNL. It stands for Trial Unique Non-Batching to Location. TUNL. It's basically a memory game. And I want to stress, this is not an easy game. This is a real cognitive workout for a mouse. Okay, walk us through a single round, a single trial.

Step 1: 9:57

The Sample Phase A white square appears on the touch screen. Let's say for this trial it appears on the left side. The mouse has to run up to the screen and poke that square with its nose. Simple enough. Step two, the delay. This is the really hard part. After poking the square, the mouse has to turn around, run all the way to the back of the chamber and poke a little port there to initiate a delay period. It has to wait. And how long does that have to wait? It varies between two and six seconds. Six seconds? That sounds really short to a human. To a mouse, it's an eternity. Think about it. Mice are these hyperactive, constantly moving creatures. To ask a mouse to hold a specific piece of information in its head, the square was on the left, for six full seconds, while also running across a room, that is a massive cognitive load. You're really making sure the hippocampus is fully engaged. You have to. You're forcing it to use its working memory. Alright. Okay, so he waits. Then what happens? What's next? Step three, the choice. The screen lights up again, but now there are two white squares, the original one on the left and a new one on the right. And to win, he has to pick. The new one, the non-matching location. So he has to remember, OK, I touched the left square a few seconds ago, so now I need to touch the right one. And if he gets it right. If he remembers correctly. A pleasant tone plays a little ding. And the strawberry milkshake is dispensed at the reward port back at the other end of the room. And if he gets it wrong, puts the same square twice. The house lights go out for a few seconds, a kind of timeout. And a correction trial begins. No milkshake. So the stakes are high, at least for a mouse. Milkshake or darkness. Very high. Now this brings us to a really critical problem with doing an experiment like this over a long period of time. It's what neuroscientists call the time versus skill problem. Right. Because you want to see how the brain changes as a mouse learns. But learning, by definition, takes time. Days. Weeks. Exactly. So if I look at a mouse's brain on day one and then I look again on day 20 and I see a big change in the neural activity, I have a dilemma. I have a confound dilemma. Did the brain change because the mouse got better at the task because his performance improved? Or did it change just because 20 days passed? You know, brains are plastic. Things drift over time. Biology is messy. How do you separate getting smart from just getting older? So how did Yagubi and the team solve that? They did something very clever and actually a little bit mean to the mice. They kept moving the goalposts. That does seem mean. But it's scientifically rigorous. As soon as a particular mouse got really good at the task with a two-second delay, say, it was getting it right 85% of the time, they would bump the delay up to four seconds. Making it harder. Much harder. And if it got good at four seconds, they'd bump it to six. They actively manage the difficulty to keep the mouse's performance in a sweet spot, around 70-80% correct. So the mouse never really gets to coast. It never masters the task and then just does it on autopilot. Exactly. The mouse is always being challenged. It's always in a state of learning. This means their performance stayed relatively constant throughout the weeks of the experiment, but time, of course, kept moving forward. And that allows you to mathematically separate the variable of performance from the variable of pure experience. And that turns out to be a key distinction because, spoiler alert, the big effect they found, this backward time travel of the signal, it correlates with experience. with time spent in the box, not just with how good the mouse is on any given day. It's a slow, cumulative change. It is. It's a deep, structural reorganization. Okay, so the stage is set. Let's look at the data, section two, the baseline map. Before we get to the wild time traveling stuff, what did they see at the beginning? What does the hippocampus map look like when the mouse is just running around this trapezoid? Well, the very first thing they checked was the most obvious. Are the place cells working? Is the GPS online? And just to refresh everyone's memory, a place cell is basically a neuron that acts like a you are here pin on a map. That's a perfect way to put it. If you had a map of your house in your brain, you'd have one neuron that fires only when you're standing in the kitchen and a different one that fires only when you're in the bedroom. They are the building blocks of the cognitive map. So did the mice have these place cells for the trapezoid? They absolutely did, but they were weird. Weird how? Well, usually when we neuroscientists study place cells, we put a mouse in a big empty open box and just let it run around. And in that setting, the place fields, which is the specific area in the room that triggers a given neuron, they tend to be big fuzzy blobs, about 20 centimeters wide on average. Okay. That sort of low-resolution pin on the map. Exactly. But in this cognitively demanding touchscreen task, the place fields were tiny, incredibly small. On average, only about 3.9 centimeters. Wow. Wait, so the map was, it was sharper, more precise? Much sharper. It was a high-definition map of the space. Why? The difference. Why would the map zoom in like that? The researcher's hypothesis is that the nature of the task matters. In a big empty box, it doesn't really matter if you're standing at coordinate X or coordinate Y. There are no consequences. But in this task, being in the exact right spot to poke the exact right square is, you know, life or death. Or at least milkshake or darkness. Right. The task demands spatial precision. So the brain responds by increasing the resolution of its map. The map becomes more detailed where it needs to be. The brain is allocating resources to what's important. Always. And they proved how good this map was. They used a tool called a Bayesian decoder. That sounds like a device from Star Trek. It kind of is. It's just a fancy math algorithm. They would take the raw firing data from hundreds of neurons, just those flashing green lights, and feed it into this computer program. And they asked the computer a simple question. Based purely on which neurons are flashing right now, can you tell me where the mouse is standing? And could it? It could. The computer could pinpoint the mouse's physical location in the box with an accuracy of about 4 centimeters. That is literally mind reading. It is. They could watch a little dot move around on their computer screen that was tracking the mouse's position, not from a video camera, but purely by reading its mind. The algorithm had a harder time figuring out where the mouse was. The brain's representation of its own location was blurry. So a fuzzy map leads to a bad decision. That's exactly what it suggests. It's not just a passive record of location. It is the active tool the mouse is using to solve the problem. And if the tool is dull, the work suffers. The map's clarity predicted the mouse's success. Okay, so we've established the GPS is working, and it's a very, very good GPS. But now we need to find the gold. We need to find the milkshake and the map. And that moves us to Section 3, the investigation. So they went hunting. They sifted through all the hundreds of neurons they were recording, looking for reward cells. Did they find any? They did. They found a very specific population of neurons. It turned out to be about 8.5% of all the cells they recorded. that were completely unimpressed by the running around. They didn't care about the touchscreen. They only cared about one thing. Milkshake neurons. The milkshake neurons. But here's where the detail just gets amazing. They weren't just one big, simple yum signal. The firing wasn't a single burst when the reward arrived. It was a complex, structured sequence. Like a relay race or something. That's a great analogy. Exactly. So imagine the mouse arrives at the port and starts drinking the milkshake. That whole process takes about three to four seconds. What they saw was that neuron A would fire right at the beginning. Then it would go quiet and neuron B would fire. Then neuron C. These cells tiled the entire duration of the drinking experience. So the brain isn't just encoding reward. It's encoding the experience of the reward over time. A little story of the reward. Yes. A beginning, a middle, and an end. And it got even more specific. This is the part that really proves we're in the hippocampus, the memory region. Some of these cells were context dependent. Meaning? Meaning a cell might fire in the middle of drinking the milkshake, but only if the mouse had just correctly chosen the square on the left. You're kidding. What if it came from the right? If it came from the right and got the milkshake, that same neuron would be completely silent. But it's the same milkshake. The same reward port, the same taste, the same action of drinking. Same everything. But the journey mattered. The history of how the mouse got there was embedded in the reward signal itself. That is the absolute definition of an episodic memory, isn't it? It's not just what, it's what, where, and when. It's the story. which is basically just a way to ask, how much does the firing of this neuron tell us about the reward? And day by day, that number just kept dropping. The reward information was vanishing from the brain. Okay, wait. If I was the scientist looking at that data, my first thought would be, oh, the mouse is getting bored. He doesn't like the strawberry milkshake anymore. That is the most logical assumption. It's what anyone would think. But the masses' behavior said the complete opposite. They were still running just as fast to the port, drinking just as eagerly. They clearly still loved the milkshake. But their hippocampus was... acting like it didn't care anymore. It had stopped representing the actual consumption of the reward. Why? Why on earth would the brain stop tracking the most important thing in the room? Because it wasn't the most important thing anymore. It had found something better to pay attention to. to. Which brings us to section four, the discovery. This has to be the aha moment of the whole paper. This is the back propagation. This is the core finding. Okay, let's go slow here. Explain this. I don't know. I'm a freshman psychology student. What does it mean that the signal back propagated? It means the firing activity of these specific reward cells moved backward in time. Okay, walk us through the timeline. Let's take one hypothetical neuron. Perfect. Okay. Day one. You are a brand new naive mouse in this experiment. You eventually figure out that if you poke the right square, you get milk at the back of the room. Oh, milk. Your reward neuron fires while you are drinking at the moment of the consequence. Got it. It's a reward consumption cell. Simple enough. Now fast forward to day 10. You know the drill now. You know that poking the screen leads to a trip to the port, which contains the milk. Now, those same specific cells, and this is critical, the mini-scope lets them track the same individual cells across many days. They stop firing during the drinking. They've gone quiet. So where are they firing now? They start firing while you are running toward the port, during the reward approach. Okay, so it's shifted from consumption to anticipation. Exactly. And now, let's go to day 20. You're an absolute expert at this task. those same cells have moved again they are now firing the instant your nose pokes the correct square on the screen so the reward signal has moved all the way from the reward itself back to the approach and now all the way back to the earliest predictive trigger it moved from the consequence to to the action that precedes it all the way back to the cue that started it all. That is just, it's wild. It's like, okay, let me try my own analogy. The ice cream truck. Yeah, let's hear it. When you're a little kid, the first few times you get ice cream, the peak moment of joy, the reward signal in your brain is when you're eating the popsicle, the cold, the sugar. That's the moment. Sure, makes sense. But after a whole summer of chasing that truck down the street, the reward signal in your brain shifts. You get that same dopamine hit, that same neural spike of excitement, the moment you hear that faint music three blocks away. That is a perfect analogy. Absolutely perfect. The brain has learned a pausal relationship. once I hear that music the popsicle is basically guaranteed so why wait until you're eating it to celebrate the brain being efficient moves the value signal to the earliest possible predictor but the paper shows this happened really slowly right it wasn't an instant switch it was incredibly slow almost glacial they were able to calculate the rate of this backward shift it was about 0.1 seconds per 0.1 seconds. So to move the reward signal back by just one full second in time. You would need about 10 days of training, 10 separate sessions in the box. This isn't just a thought changing on the fly. This is a slow structural reorganization of the mind. It's the physical wiring of the hippocampus adapting over time. And the data clearly showed that as the number of these classic reward cells went down, the number of screen cells, the neurons that fire when the mouse touches the cue, started to skyrocket. Yes, the population shifted. The brain was actively transferring the value. It was taking the goodness of the milkshake and metaphorically pasting it onto the image of the square. This just explains so much about modern human behavior. You know, why we get addicted to our phones, why the sound of a text message notification is sometimes more exciting than the actual content of the message. We condition ourselves to value the predictor over the reward. And there's a deep, powerful survival reason for this. We choose. Efficiency. Just think about it from the mouse's perspective. Once it has arrived at the port and is drinking the milkshake, does it need to learn anything new in that moment? No, I guess not. The problem is solved. It's just. eating. Right. The task is over. The danger has passed. The critical moment, the point in the trial where the mouse could screw up and lose everything is the choice. The moment it pokes the screen. So the brain essentially shines its cognitive spotlight on that moment of decision. Exactly. Exactly. The AI connection. The AI connection. So in the world of artificial intelligence, specifically in a subfield called reinforcement learning, there is a foundational algorithm called TDRL. TDRL. Temporal Difference Reinforcement Learning. Now, this isn't some obscure algorithm. This is the fundamental logic used by systems like AlphaGo, the Google AI, that famously beat the world champion at the game of Go. Okay, so this is serious stuff. How does TDRL work in simple terms? It's all based on something called prediction error. The AI is constantly making a guess. Based on the state of the game board right now, how much reward do I expect to get in the future? Okay, it's making a prediction. And at the very start of its training, it's terrible at this. The only time it gets any concrete information about a reward is at the very end of the game. It sees a screen that says you won. So initially all the value is at the end. Makes sense. But to get better at the game, the AI needs to know which move it made 10 minutes ago was the one that caused the win. So the TDRL algorithm is explicitly designed to mathematically shift that final value signal backward in time, step by step. It learns that the move I made in this position led to a position that led to a position that eventually led to the end. this win. So it assigns a little piece of that final value to that much earlier move. It back propagates the value from the outcome to the actions that predicted it. Exactly. Computer scientists and mathematicians wrote this code decades ago, not because they were looking at brains, but because it is the mathematically optimal way to solve a problem with delayed rewards. It's just pure logic. And now? And now Yagubi and his team peer inside a living mouse's hippocampus and they see the biology doing the exact same thing, step by step, session by session. The mouse's brain is running TDRL. The mouse's TDRL. Or, to be more accurate, TDRL is a clumsy, simplified imitation of what the mouse's brain has been doing for millions of years. That is a huge, huge validation for that entire field of AI theory. It's massive. It suggests that intelligence, whether it's evolving in a silicon chip or in a wet, squishy brain, might follow the same fundamental convergent laws. To master the present, you have to be able to predict the future. But there's a bit of a twist here when it comes to the brain, right? Because when you hear the term prediction error, my mind immediately jumps to one thing. dopamine. You are spot on. And for years, that was the consensus in the field. The model was the VTA, the ventral tegmental area, which is the brain's dopamine factory, does the prediction error calculation. And the hippocampus just, it just draws the map. So the VTA is the fortune teller. And the hippocampus is the cartographer. A perfect division of labor. That was the model. Yeah. But this paper really blows that up. It provides incredibly strong evidence that the hippocampus, the cartographer itself, is doing its own predictive coding. The map is not just a map. The map is alive. The map is alive and is constantly trying to predict the future. And this leads to this really cool visualization they did using a tool called CEBRA. CEBRA, not Z-E-B-R-A. C-E-B-R-A. It's a newer machine learning tool. It's a way of taking all this messy, high-dimensional neural data, hundreds of neurons firing over time, and projecting it down into a simple visual shape, a kind of shape of the thought. A manifold. Yes, exactly. A low-dimensional manifold. And when they applied this to their data, they could see the brain's activity organize itself into distinct clumps. There was a clump four at the screen, a clump four on the approach path, and a clump four at the reward port. And over the weeks, they could literally watch the reward clump shrink and the screen clump grow and become more prominent. It's a visual proof of the value transfer. You can see the brain's priorities shifting. You can see it clear as day. So this is probably a good time to move into section six, the implications. Because if the map is constantly shifting its focus based on value, is it really a map of space anymore? That is the big philosophical question that hangs over this work, isn't it? The authors argue that it is still a map of space, but that we should start calling it a predictive map. A map colored by desire. Precisely. And they have data for this too. they found that the map over-represents the reward location. What does that mean, to over-represent a location? It means that there are physically more neurons, more place cells, dedicated to mapping the small area right around the milkshake port than there are for, say, the empty corners of the box. Huh. It's like that famous New Yorker's map of the world where 9th Avenue is huge and then China is this tiny little speck across the river. That is exactly it. It is a distorted, biased map. The things that matter to you take up more real estate in your brain. The important landmarks get more neural resources. And they actually calculated a score for this distortion, right? He did. They called it the reward over representation score. And it was significant and it grew over time. As the mouse learned that the port was the most important place in its universe, the density of its mental map physically shifted to provide a higher resolution picture of that goal location. Wow. Does the brain know how good the reward is? Or is it just a binary system, good thing here versus nothing here? Oh, it knows. It's an analog system. They did a really neat control experiment that's mentioned in the extended data of the paper. On some trials, instead of giving the full, satisfying strawberry milkshake, they gave the mouse just a tiny little drop. An incentive, they called it.- A tease. - A total tease. And the neurons fired differently, The predictive signal that had shifted back to the screen was weaker on those trials. So the brain isn't just predicting a reward is coming. It's predicting a huge reward is coming versus a small reward is coming. It is quantifying the expected value. This changes. I mean, this fundamentally changes how I think about my own memory. How so? Well, I think we all tend to trust our memories. We think of memory as an objective recording of what happened. This is the way it was. But this research suggests that my memory is actually a highly edited, deeply biased, distorted tool that has been shaped for one purpose to get me to the next milkshake. You are hitting on the big final outro thought here. Let's go there. Let's sum this all up. We've been on a real journey today. We started with this humble mouse in a box. We saw the raw data from its brain, that galaxy of flashing green stars. We saw how its internal map of the world wasn't static. how it sharpened up and focused on what was important. And then we watched as the reward signal itself did the impossible. It traveled back in time, shifting from the reward to the cue, all driven by the brain's relentless need to predict the future. And in the process, confirming that our brains are effectively these beautiful biological AIs running a version of a reinforcement learning algorithm. So what? Of all of this, the big takeaway for everyone listening is that we really need to stop thinking of the brain as a historian. It's not a history. It's a survival engine. And every single memory you have, your memory of your childhood home, your wedding day, the route you take to work, it is not preserved in your mind for the sake of nostalgia. It is preserved because your brain has made a calculation. There might be information in this data that I can use to predict a reward or avoid a punishment in the future. It's all just training data. It's all just training data. And like any good machine learning model, the brain edits and curates that data. It emphasizes the wins. It downweights the losses. It physically distorts the map to make the goals look bigger and closer. So the provocative thought to leave our listeners with is this. Are any of our memories a faithful record of the past? Or are they all just subtly rewritten, constantly re-edited stories that our brain is telling itself to help us get what we want in the future? I think we are living inside the map, not inside reality. And honestly, that's probably a good thing. A feature, not a bug. If we saw reality exactly as it was, flat, objective, with no part more important than any other, we'd be paralyzed. We wouldn't know where to go. We wouldn't know what to do. The bias is what gives us direction. The bias is what makes us move. The bias is what makes us alive. That is a beautiful if, yeah, slightly disorienting place to land. It is, isn't it? So next time you're walking to that coffee shop, pay attention to that little spark you feel when you see the sign. That's not just excitement. That's your hippocampus time traveling. That's your brain back propagating the joy. Thanks for taking this deep dive with us. And a huge thank you to Mohamed Jagoubi and the entire team at McGill for giving us this incredible window into the mind. It was a real pleasure. See you on the next deep dive.