How We Learn

Doing crossword puzzles helps stave off Alzheimer's...right? It's one of those things everyone just knows, and it follows the conventional wisdom of use it or lose it. But let’s actually look at the data behind this claim, as some researchers finally did, and see if it holds up.

The study that launched this idea followed 488 elderly people, of whom 101 eventually developed dementia, and only 17 of which regularly did crossword puzzles. The finding in this study was that these 17 people scored better on memory tests for longer, meaning specifically that their mental decline appeared to accelerate about 2.5 years later than those who weren’t doing crosswords. Which sounds promising, except that once the decline began, these 17 people fell off a cliff faster, which meant that by the time dementia was actually diagnosed, both groups ended up at essentially the same place. So the people doing crosswords didn’t actually decline less, they just looked fine on the tests longer before crashing harder.

And the reason this fooled the researchers is that it turns out the memory tests used to assess these people’s decline were made up of word recall tasks, which crossword puzzles are specifically training people to do. Which means the people who do crosswords regularly were already better at the thing the test measured, and they also happened to have demonstrably higher verbal IQ scores to begin with. So they weren't actually protecting their brains from Alzheimer's at all, they were just better at the test being used to assess the decline, and which just so happened to overlap with their hobby. It’d be like concluding that playing basketball helps you keep your physical coordination longer, and then testing physical coordination by assessing whether or not you can throw your trash into the bin effectively from across the room. Those basketball players could probably perform the task better longer, but if their physical abilities were really declining, there would come a point where they would completely lose the ability to do so. This is what we’re seeing here with the crossword puzzle people. So what the data actually shows is that doing crossword puzzles just…makes you better at doing crossword puzzles.

So if crossword puzzles don’t help, what does? What if the real protection against decline is actually more simple than targeted treatments, or puzzles, or diet, or pills. What if it lies in understanding how patterns get wired into our brains - and more importantly, how we can wire in new competing patterns if needed? For over a century, scientists have been documenting exactly this. Piece by piece, lab by lab, they've been mapping how learned responses form, why they persist, and how they can be changed or overridden. They were doing the same thing as medicine, documenting pieces of the elephant without seeing the whole, but all the while building an instruction manual telling us how we get sick, and how we can heal.

Learned responses

In the late 1800s a physiologist named Ivan Pavlov was studying the neural control of gastric secretions in digestion, using dogs as his subjects. He built an elaborate sound and vibration proof lab that got coined The Tower of Silence, in an effort to isolate as many variables as possible. To study this he surgically created openings in the dogs' digestive systems, which allowed him to collect and measure their gastric secretions in real time when they ate. To picture this, you can imagine his dog test subjects walking around living their regular dog lives, with collection tubes hanging from their mouths. What he started noticing was that the dogs' digestive responses, the gastric secretions, were starting before food even arrived, which was messing with his meticulous measurements and driving him crazy. These dogs would start salivating just from seeing the lab assistant who usually fed them, or even from hearing footsteps in the hallway. They were salivating in anticipation of the food’s arrival (sound familiar? Seems suspiciously similar to what happens with placebo effects, physiological processes being triggered by expectation…). This "psychic secretion," as Pavlov called it, was contaminating his measurements of the purely biological digestive response.

What researchers normally do in this type of situation is to control for the contaminating variable and move on with their original research, as medicine has been doing with placebo for 70 years. But Pavlov instead became obsessed with understanding what was actually happening. He saw a surprising response, and instead of labeling it “paradoxical” or a “confounding variable”, he got curious. And what he realized was that something was causing these dogs to respond physically to something that had no inherent biological meaning. Footsteps obviously don’t contain any nutrients, and yet the dogs' bodies were responding as if they did. He decided to pivot his research and study where the data led him instead of holding rigidly to his original hypothesis.

His new experiment was really simple; he would tap a buzzer, then give the dogs food. The food triggering salivation was normal and expected, but he found that after pairing the buzzer with food enough times, the buzzer alone started to trigger salivation, even if there was no food present at all. The dogs' brains had created a connection between a completely arbitrary sound and a biological response. He found he could make them salivate to metronomes, visual stimuli, even electric shocks; anything that he repeatedly paired with food became a trigger for the digestive response. Pavlov had built his Tower of Silence to eliminate every possible contamination - and then data contamination in his experiment ended up leading to one of the most important discoveries in the history of learning.

What Pavlov had discovered was that the brain creates associations that can produce physical responses. And associations like these are actually something we utilize all the time without thinking about it. As an example, flashcards work just like this. You look at a word on one side, the definition on the other, and as you repeatedly expose your brain to those two things in rapid succession, just like Pavlov did with the buzzer and the food, your brain links them together, so seeing one automatically brings up the other. And it makes sense evolutionarily why the brain does this. The faster your brain can make these associations, the faster it can trigger your reactions, the more likely you are to be the animal that makes it out alive. If you hear crinkling leaves, and then a tiger appears, after this happens just a couple of times, your brain will start to ready the fight or flight response just from hearing crinkling leaves, before a tiger comes into view. And it's exactly what was happening with the paper rose triggering an asthma reaction. Her body paired the image of the rose with the reaction enough times that the image alone could produce the response. Medicine saw this and it didn't fit their framework so they concluded this wasn’t real asthma but something else. But it makes evolutionary sense, right? Better to produce the full immune reaction in case the rose was real, than not produce it and end up in trouble. This is actually a really smart evolutionary adaptation. Until it starts linking things you don't want linked.

In 1920, an American psychologist named John Watson set out to prove that all human behavior was learned, not inherited. He was convinced that he could mold anyone to do or be anything. And for some reason he chose to start this crusade of proof by terrorizing a nine month old child named Little Albert. Before this experiment little Albert had no fear at all of rats, but when Watson presented him with a white rat, just as baby Albert started reaching for it, he struck a steel bar with a hammer directly behind Albert's head. The baby cried and fell forward, his threat response triggered by the sound. You can actually watch videos of this experiment on YouTube. It’s…disturbing. Watson shows no emotion at all about the child’s distress. Then Watson repeated this experiment seven more times over two sessions. By the end of these seven exposures, Albert would cry and try to crawl away just from seeing the rat, without any noise at all. The fear association to white rats had been wired in.

Five days later, they found that Albert was now also afraid of a white rabbit he'd previously happily played with. Then he cried at a white dog, a fur coat, and even a Santa Claus mask. Watson's own white hair caused Albert to cry when he leaned in close. The fear of the white rat had generalized, causing baby Albert to be afraid of anything that looked similarly white and furry. His brain had taken one learned association and attributed it to anything that seemed remotely similar. Watson had only repeated this experiment seven times, and in that short time he had created a network of fear responses in Little Albert to an entire category of objects. Watson had proven that emotional responses could be created through simple association, and that you could wire fear to literally anything if you paired it with something scary. It’s not known for sure, but there’s speculation that Little Albert’s mother learned what was happening to her son, because she pulled him from the study before Watson had the opportunity to see if he could reverse this conditioning, and despite a lot of people trying, no one has been able to track down Little Albert to see what came of him in the aftermath of this. And Watson, caught having an affair with one of his students and fired from Johns Hopkins, never did get to finish his crusade to prove his full theory.

Seven pairings. That's all it took to create a fear that spread to everything white and furry. And this is exactly the pattern we saw earlier with the asthma patients whose triggers kept multiplying - first cats, then dogs, then dust, then cold air, then exercise. Or the IBS patients whose safe foods kept shrinking. Medicine notes these expanding lists and thinks, yep, the disease is progressing as we expect it to. But the learning science is clear; the brain learns one association and then generalizes it to anything that seems similar. It's trying to protect you by casting a wide net - and once that net catches something, it becomes part of the pattern. The groove gets deeper. The net gets wider. And the system keeps pulling you back into the same response. When you wire in one association, if it’s strong enough, it can generalize to other related seeming things.

Systematic sensitization

In the 1930s a researcher named B.F. Skinner was running low on food pellets one weekend and didn't want to go make more. So he set his apparatus to only reward the rats in his experiment sometimes instead of every time, which led to one of the biggest discoveries in learning and behavioral research. Skinner had been studying how rewards shape behavior - when a rat presses a lever, they get food. But when he came back Monday after his lazy weekend adjustment, the rats were pressing the lever like maniacs, far more obsessively than when they got food every single time.

Skinner had assumed that continuous reinforcement, giving the animals food every time they pressed the lever, would create the strongest behavior. But what he found was that what is known as intermittent reinforcement, when you get the reward randomly (the rat equivalent of a slot machine), this made the behaviors way more obsessive and sticky, meaning more likely to persist. Even when Skinner stopped providing food entirely, the animals that had been on intermittent reinforcement would continue pressing the lever hundreds, even thousands of times before giving up. The ones who'd always gotten food would stop after just a few unsuccessful tries. In one experiment, Skinner gave pigeons food at random intervals regardless of what they did. And the pigeons started developing elaborate "superstitious" behaviors - one would spin in circles, another would thrust its head into corners, another would do a pendulum motion. Each pigeon had decided that whatever it happened to be doing when food arrived had CAUSED the food, and kept doing increasingly elaborate versions. The rituals became increasingly complex and rigid, which illustrates how false associations can create elaborate behavioral protocols based on coincidence. So Skinner had accidentally discovered two more crucial pieces: that unpredictability creates the most persistent patterns, and that associations can also encourage false explanations for what’s causing our experiences (more on this piece in the next chapter.)

When a behavior is rewarded unpredictably, the brain never knows when the next reward might come, so it keeps trying, the uncertainty itself drives persistence. If you never know when the next meal will come it makes sense to keep checking. And it's not just rewards that work this way. The same mechanism applies to anything unpredictable - including threats. When danger appears sometimes but not always, the brain can't afford to let its guard down. It stays vigilant and keeps scanning for the threat, keeping the threat detection system running. And anyone dealing with symptoms that come and go knows exactly what this feels like. The pain that appears sometimes but not always, the bad day that arrives unpredictably after a string of good ones, these create neural patterns far stronger than consistent experiences would. The brain can't stop checking, so it becomes hypervigilant.

And that vigilance changes your actual physiology. When you're constantly scanning for symptoms, you're training your nervous system to detect smaller and smaller signals. The same synaptic strengthening that lets you learn a language or memorize facts is now making you better at feeling pain, or gut sensations, or whatever you're monitoring, just like any skill gets honed through practice and attention. Remember the monkeys in Merzenich’s experiment who were learning to touch a spinning disc? The region in their brain for that specific finger exploded, they actually grew more neuronal connections in that part of their brain to enhance their ability to perform that skill. The same thing is happening here. Signals that used to be too weak to reach conscious awareness now register loud and clear. Medicine sees this hypersensitivity and assumes it's causing the condition - but you weren't born with intestines that feel everything, or hypersensitive lungs. The sensitivity is a product of the vigilance, not the other way around. You literally trained yourself to feel more by paying so much attention.

So now that we understand how these patterns and even symptom associations form, spread, and dig themselves in deeper the more we try to monitor them, it begs the question: is that it then? Are we stuck with them forever as medicine seems to think? To answer that, let’s look at what they were documenting on the other side of the equation: what we might actually be able to do about it.

Unlearning responses

In the 1970-80s, Mark Bouton at the University of Vermont wanted to know if we could reverse the kind of conditioning that Watson did to Little Albert; if we could extinguish (get rid of) or undo that kind of fear response. So he first taught rats to be afraid of a specific tone by pairing it with a shock, similarly to how Watson paired the white rat with the loud noise. Then he played that same tone repeatedly without the shock until they stopped showing fear. The prevailing assumption in neuroscience is still “use it or lose it” to this day, so the assumption here was that the fear memory in these rats had been erased once they stopped showing fear. Or that it had at least been weakened. But instead of extinguishing the fear and calling it done, Bouton kept testing that assumption. And what he discovered was that the original fear was actually still there, fully intact, and could be resurfaced if given the right conditions.

He showed that when he extinguished a fear completely in one location and then tested the animal in a different room, the fear was still there, as strong as ever. The animal hadn't unlearned that the tone was dangerous; it had only learned that the tone wasn’t dangerous in that specific room. So then he tried testing another assumption: he would extinguish a fear completely, and then he would wait a few weeks without any training or exposure, and do the test again. He found that the fear returned completely, as if the extinction had never happened, even in the learned location. And finally he showed that when he extinguished a fear, and then exposed the animal to mild stress that was completely unrelated to the original fear conditioning, the fear came roaring back. So what he was seeing was that extinction wasn't actually erasing anything - the original fear learning remained intact - it was just temporarily being suppressed under specific conditions.

By the 1990s, brain imaging technology finally let researchers see what was actually happening during extinction, and it confirmed what Bouton's behavioral experiments suggested. When they looked at the brain activity during extinction, they found that the fear neurons in the amygdala still initially activated when the stimulus appeared. The amygdala was still recognizing "this is the thing that was dangerous," but for responses that have been extinguished, neurons in the prefrontal cortex would quickly activate and inhibit the fear response before it could fully develop into the physiological and behavioral fear reaction. Which makes a lot of sense. If you learn that snakes are dangerous, and then you experience some snakes in the wild that don’t attack you, you don’t want that fear circuit to disappear completely, you just want slightly more nuanced activation. Because some snakes ARE dangerous, and your brain needs to know that.

So what Bouton had shown was that fear learning doesn't get erased, it gets suppressed. The original circuitry stays intact. And this turns out to be true not just for fear, but for learning in general. We don't prune anything. Which is why you may remember your childhood phone number, or the red socks you wore to your 3rd grade school concert. These details weren't important, they just still... exist. Because that's how our brains work.

And the prefrontal activation that the scans show inhibiting our pathostatic responses to fear is actually the same thing we saw in Chapter 2 with placebo. When someone believes they're getting treatment, their prefrontal cortex provides the safety signal that allows the fear response to quiet down. And then over time, with enough safety experiences, the prefrontal inhibition can become faster and more automatic, so the fear response gets suppressed before it fully develops. This might explain why sometimes drugs like SSRIs can work long term; not through the serotonin mechanism medicine assumed, but because taking them every day creates a repeated safety signal that activates the same prefrontal inhibition we see in extinction. We are constantly reminding our body, every single day, I’m being helped here, I’m being taken care of. But the original fear circuits still remain structurally intact, all the synapses and connections still there. Which is why under stress, or in new contexts where the safety learning hasn't occurred, the original fear response can return at full strength, as if the extinction had never happened.

This split between what we consciously know and how our body automatically responds becomes clearer when you look at how memory actually works. Researchers discovered that we have two completely separate memory systems that encode experiences in fundamentally different ways. The explicit memory system, centered in the hippocampus, creates conscious memories you can deliberately recall and describe. This type of memory contains the story you have of what happened to you. And the implicit memory system, running through the amygdala and other subcortical structures, creates unconscious emotional responses and physical reactions that fire automatically without any conscious recall or sometimes even awareness that you're remembering something. This is where our 500 million year old threat detection system lives.

Most of the time these systems work together seamlessly; you have an experience, and you get both the story and the feeling, tagged to each other. You remember your wedding day and the joy comes with it. You remember an embarrassing moment and feeling it elicited comes up too. The explicit memory calls up the implicit response and visa versa. But when something traumatic happens sometimes these two systems can encode completely different aspects of the experience, and they don't necessarily communicate with each other. Your explicit system might remember the facts of a car accident, the color of the other car, the intersection where it happened, while your implicit system remembers the feeling of your chest tightening, the sound of screeching brakes, the sensation of losing control. Later, you might hear brakes squeal and feel your whole body tense without consciously thinking about the accident at all. The implicit memory fires before the explicit system even knows what's happening.

And in people with chronic conditions, this imbalance compounds over time. Cortisol impairs hippocampal function while leaving the amygdala fully active — or even hyperactive. So the system that should be contextualizing and 'date-stamping' experiences, that could help you understand 'that was then, this is now,' gets weaker. While the automatic body responses get stronger. Day after day, the implicit system is reinforced while the explicit system is suppressed, making the automatic threat loops stronger, and the conscious override weaker. This is why chronic conditions can seem to take on a life of their own.

This is why you can know intellectually that you're safe while your body is reacting as if you're in mortal danger. And this is why most therapeutic approaches fail to create lasting change - talk therapy engages the explicit system but often can't reach the implicit body memories. Somatic approaches might calm the implicit system temporarily but don't update the explicit understanding. Medicine doesn't address either system, it just tries to suppress the downstream symptoms. To actually update these patterns, you need both systems engaged simultaneously, which rarely happens by accident.

Contextual learning

If you study for a test when you’re tipsy, you’d be best served by being tipsy while you take the test too. In the 1960s researchers found that when people learned something while they were drunk, they could remember what they learned better when they were drunk again, rather than sober. And when they studied this weird discovery further, it turned out to be true for every state they tested. Information learned while anxious was best recalled during anxiety. Skills practiced while calm were most accessible when calm. Pain states, and mood states, and even specific body positions all created state-dependent memory networks that were most available when those states returned. And this understanding actually helps Bouton's findings make a lot more sense. The rat didn't learn 'the tone is safe now.' It learned 'the tone is safe in this room.' The safety learning was tagged to that specific context, which is why it disappeared the moment you changed the room. If you change the context, you change the learning. Which makes sense evolutionarily. Your body could learn, tigers can’t get me in this cave, but when I go outside all bets are off. We had to be able to make context dependent threat assessments, so we could let our body relax when it was safe, and could stay vigilant when the possibility of threat demanded it.

In the 1980s a researcher named Gordon Bower at Stanford hypnotized subjects to feel either happy or sad, then had them learn lists of words. Later, he tested their recall in either the same mood or the opposite mood. People who learned words while sad remembered them better when sad again. People who learned while happy recalled better when happy. The brain wasn't just storing the information; it was storing the information tagged with the internal state present during learning. This means that all the coping skills you learn while feeling calm and safe in a therapist's office are neurologically tagged as "calm state" information, making them least accessible precisely when you need them most, during the stressed, symptomatic states where the maladaptive patterns live. It’s why meditation is often just like doing crossword puzzles. You are learning to be calm WHILE you meditate, in those specific circumstances. When you are annoyed with your spouse because they didn’t empty the dishwasher, that meditative state you’ve been practicing isn’t going to come up unless you make it come up. And both of these state dependent findings, mood states and state-dependent learning, provide a lot of context for how illness is self perpetuating. When you're sick, you're learning all your patterns in the sick state. Every symptom, every fear response, every coping attempt gets tagged to that pathostatic state. So being sick becomes the context that triggers being sick - your body learns to be a certain way when it's getting specific stimuli, and the symptoms and physiology that state creates actually create a feedback loop that becomes more and more sticky and thus chronic.

And just to round out how far these associative findings go, let’s look at the research on drug overdose as it relates to location. A researcher named Siegel showed that situation-specific drug tolerance is capable of preventing fatal overdoses. There was one man who received the same morphine dose four times a day for four weeks, always in his dimly lit bedroom, where he was bedridden. But one day he for some reason dragged himself out of his bedroom into the brightly lit living room. When his son gave him the exact same dose in the living room he died from an overdose. His body didn’t mount the tolerance response it had learned, without the context of the room it had learned it in. And if context can determine whether or not a morphine dose is fatal, imagine what it's doing to every other physiological process. Pain researchers discovered that chronic pain patients often felt better in novel environments like hospitals or vacation spots, only to have their pain return the moment they got home. The brain had learned not just "pain" but "pain in my bedroom," "pain at my desk," "pain at 3 PM," creating elaborate networks of association that turned entire environments into symptom triggers. This should have changed or at least inspired research on how we think about physiological responses to medicine, physiological responses to pain, and really, everything else. But instead, medicine just noted it and filed it away.

Hermann Ebbinghaus figured out another piece of the puzzle when he sat alone, memorizing nonsense syllables like 'DAX' and 'BOK' for hours every day, using himself as a test subject. This obsessive self-experimentation led him to discover what he called ‘the spacing effect.’ He learned that repeated practice done over time led to better retention than a lot of practice all at once. Cramming doesn't work. Spreading practice over time does. Which is another reason one-off therapy sessions or weekend retreats don't create lasting change.

This is what Skinner had been seeing with his intermittent reinforcement schedules; when rewards came unpredictably, each instance created a separate learning event. Which shows why the very unpredictability that makes symptoms so maddening also more deeply embeds them, not by creating one strong pathway but by linking the same pattern to hundreds of different states and contexts - creating a web of triggers that can activate it from almost anywhere in a person's life. State and context learning also explain really clearly why meditation often leads to you…being more calm when you’re meditating, but doesn’t necessarily lead to you being more calm in your everyday life. If you consciously take those meditation principles and apply them to how you show up moment to moment, that is when meditation becomes an impactful practice. But if you are only applying the meditation when you’re actively meditating, only applying the principles when you’ve shut out the world, that doesn’t end up translating to being something that helps when you’re activated and stressed. This is the difference between what are called state changes and trait changes. Meditating in the way most people do it creates a state change: calm while meditating. Vagal toning exercises like voo breathing, or cold plunges - same thing. You may be momentarily changing your nervous system state, but it only applies to that moment. Breath work is another example. In order to create a trait change, something that changes the way you show up in your life, you need to use practices throughout your day and your life in every single context. Across all contexts over time.

So now that we know isolated practices only create isolated change, let's look at what the research shows is actually required to create lasting updates.