How Dopamine & Serotonin Shape Decisions, Motivation & Learning | Dr. Read Montague
These notes were generated by AI and may contain inaccuracies.
If any goal that you achieved, whatever it is, taking a drug, eating a food, getting a partner or whatnot, if that was enough for you, then you wouldn't keep living. You want that system to keep tracking and once it gets to one place, you want it to have another place to which it could go. Otherwise, you wouldn't live.
Welcome to the Huberman Lab podcast where we discuss science and science-based tools for everyday life. I'm Andrew Huberman and I'm a professor of neurobiology and ophthalmology at Stanford School of Medicine. My guest today is Dr. Read Montague. Dr. Read Montague is the director of the center for human neuroscience research at Virginia Tech. He is also an expert in the science of motivation, decision-making, and learning, and a pioneer in developing methods to directly measure levels of dopamine and other neuromodulators in humans in real time.
Today, you'll learn how dopamine really works. Not just to regulate your levels of motivation. We've all heard that before. But also to teach you things. Dopamine is involved in learning as well as persistence or lack of persistence. As Read will teach you, most of what we hear and know about dopamine is based on the idea that dopamine levels go up or down depending on our levels of expectations and then what happens. But as he explains, most aspects of life, work, school, relationships, our pursuit of money, etc. involve multiple milestones. We work, we wait, then we get an outcome that in turn informs the thing we do next. Or maybe dopamine arrives suddenly with no work involved at all. In other words, dopamine levels are constantly changing and that shapes not just what you do now, but how you think about your recent past and what you will do next. So when we say dopamine is involved in learning, today you are going to realize that dopamine is teaching you how to adjust your behavior.
We discuss how this knowledge can be leveraged for better motivation and decision-making, even better social interactions. We also discuss serotonin and how dopamine and serotonin work in sort of seesaw fashion and how serotonin in particular teaches you about unwanted outcomes. We have a discussion about SSRIs that you're going to find fascinating. As Read points out, SSRIs increase levels of serotonin, but often that serotonin gets used at the dopamine synapses to reduce the rewarding properties of dopamine. So today's discussion about dopamine and serotonin is going to be vastly different than any that you've heard or read about elsewhere. You're going to learn how those neuromodulators work, and you're going to learn how they impact your everyday life and decision-making.
As we all know, discussions about dopamine and serotonin are everywhere nowadays. But in today's episode, you're going to learn from a top expert in the field what these molecules truly do. And that's going to help you better leverage your efforts, introduce what we call deliberate delays, and how to use tools like AI to improve your levels of motivation and your ability to learn through neuroplasticity. Before we begin, I'd like to emphasize that this podcast is separate from my teaching and research roles at Stanford. It is however part of my desire and effort to bring zero-cost to consumer information about science and science related tools to the general public. In keeping with that theme, today's episode does include sponsors.
And now for my discussion with Dr. Read Montague. Dr. Read Montague, great to see you after all these 15 years. >> 15 years. You turned us down for a job offer then. >> I did. But we both turned out okay. >> Well, I hope so. >> Well, you certainly turned out okay and you look great. It's always great to see a colleague looking so fit and healthy who also raised five children successfully and all those things. We'll talk a little bit about your life and maybe your athletic life a little bit later, but I want to talk about dopamine.
The world is obsessed with dopamine. Now, until very recently, people thought about dopamine as a reward. Now, slowly, people are starting to understand that dopamine is involved with things other than feeling good such as motivation, movement, etc. How do you think about dopamine the neuromodulator and then we'll move into the context in which you study dopamine but when somebody says what does dopamine do how do you think and respond to that question.
It used to be that dopamine was thought to equal pleasure dopamine goes up you feel good dopamine goes down you feel less good okay there's been an explosion of work on it most of the new work that's not psychological has been out of the artificial intelligence world what's now called artificial intelligence. It's very clearly a learning signal number one. So dopamine fluctuations high and low control learning. It's also playing multiple roles. It plays a role in motivation and it may also play a role in the way you feel. It's less well understood how the sort of mechanics of what dopamine does for changing your nervous system relates to your feeling state. You can have a feeling state that's good and see things that don't correlate with dopamine being the cause of it.
Let's talk about dopamine in the context of learning because that's something that most people don't associate with dopamine. What are a few examples of what we know about dopamine and its role in learning?
That's a world I can't even summarize in a quick way. People that work on rodents now will take a genetically modified rodent and they will study the way in which dopamine release correlates with something the animal is learning. The animal may learn to turn left when it sees a light. It may learn to run toward food. It may learn to run down a maze. All kinds of learning tasks associated with the animal are associated with dopamine fluctuations in your brain. Now these aren't global. They're all over the place, but there are different kinds of signals that you can find in different spots in your brain.
We've begun to understand dopamine as a central player in the algorithms that your brain runs. And that's where people like me, computational neuroscientists, have made a connection. And that's the connection between the kinds of learning rules and learning procedures that are installed in your brain and installed in the brain of every mobile creature on the planet and dopamine fluctuations. So that's a strong connection that has been worked out over the last 30 years. The algorithms are well understood. What wasn't well understood 30 years ago was the kind of remarkable things those same algorithms can learn.
There have been a bunch of modern breakthroughs in what's called reinforcement learning. And reinforcement learning's main biology partner is dopamine. It's the first big hit. Now, you know, it's a area of science. And so, what happens when you have a big finding that looks like it explains a lot of things? Well, people come rushing in to sort of beat it up. That's their job. That's their job to hack away. Oh, is it really this? Does it work the same in this context and that context? But I think the description of what dopamine does as a learning rule is pretty much true.
Let me give you an example. Psychologists since the time of Pavlov have understood what it means for an animal to generate a prediction and to compare it to an outcome. Suppose today's Wednesday. Suppose I make a prediction today that it's going to rain two inches on Saturday. Now, we're going to fast forward to tomorrow and I'm going to update my prediction because I have new knowledge and it's going to say it's going to rain 10 inches on Saturday. There's been no reinforcing feedback. It hasn't rained yet because it's now Saturday yet. I'm making a prediction about Saturday. But there's a difference between this expectation and that expectation. Those differences are encoded by dopamine. It's called a temporal difference error. Dopamine seems to code that before you ever get to the terminal return.
Imagine that you were playing a game like checkers. You make a move in the game and you might make, 40 moves before you win the game. And suppose winning the game is the reward. Well, you may have some prediction. Your brain makes a prediction when you play board position to board position that you're going to win the game. And that's a fluctuating quantity. That's a different kind of learning rule.
The kind of learning rule that psychologists talk about that you think about in your everyday life is it's going to rain two inches today. How much did it rain? So that's a comparison between an outcome and your expectation. What Rich Sutton and Andy Barto did was said, well, what you really want to do is you want to stick between there your next prediction. So you want successive predictions. And why is that a good model for animals? Well, because if you're an animal and you're wandering around foraging, mainly you're not finding anything. You're going from position to position to position to position position, but you're learning and dopamine is encoding those signals.
I'm so glad you said the word foraging because I want to hover on the theme of foraging in the context of human decision-making and learning and behavior. To stay with your description, Saturday rolls around. Let's say it doesn't rain. Let's say the person doesn't want it to rain. They're not a farmer. They want to go to the beach on Saturday.
Now we can talk about reward prediction error, right? The difference between the expectation when it actually happens.
Okay, let me interrupt and correct that a bit. The reward prediction error that people talk about dopamine representing is the prediction error that you get for every single step whether or not you've received reward. That's kind of diffused out in the psychology literature as you have an expectation and you have a reward. It may be positive, negative or zero. And what you do is you make an error there. That was understood in the 60s and 70s. It's called the Rescorla-Wagner rule 1972. That's how the system should learn. The fact is though that doesn't model reality very well. Reality doesn't give you feedback like that every time. Reality often gives you long stretches of nothing.
The insight I think of Sutton and Barto in their algorithm was well a better algorithm for learning continuously is to take successive predictions and to say that's a learning rule. Obviously it's a learning rule of the outcome when you get an outcome when it's not zero. But it's successive predictions. It's like why that should be such a deep idea is not clear to me. What is clear to me from data is an algorithm based on that is installed in brains, C. elegans slug brains all the way up to human brains. There are these temporal difference reward prediction errors.
So I guess I'm sitting here trying to backwash the old version of it which is people say in a kind of vernacular way oh it's the difference between your expectations and the reward. Yes when that happens but most of the time that's not happening in which case it's the ongoing difference between your expectation and your next expectation. So it's fluctuations in your expectation as you move through the world.
The Deep Mind guys in London who beat the world Go playing champion and made AlphaFold and won Nobel prizes and I mean they're starting in 2015 they just had this unbelievable series of hits. They used the Sutton and Barto algorithm. They trained those systems where people would make the players computer players would make hundreds of board position changes before you ever got to the end of the game. and update and learn based on that.
So there's a difference. It's not just expectation and outcome. It's expectation next expectation current outcome and that is what rolls through and that is what we see installed in we have a paper this week coming out on honeybee brains where you can show the same sorts of learning rules in honeybee brains. In honeybee brains it's probably octopamine not dopamine. The other thing to say about dopamine is it's not just dopamine. It's very clear that lots of neuromodulators like that are fluctuating with learning and motivation and probably the whole symphony of them that creates motivation states and things like that.
I'd like to take a quick break to acknowledge one of our sponsors, David. David makes protein bars unlike any other. Their newest bar, the Bronze Bar, has 20 grams of protein, only 150 calories, and zero grams of sugar. These are the best tasting protein bars I've ever had, and I've tried a lot of protein bars over the years. These new David bars have a marshmallow base and they're covered in chocolate coating and they're absolutely incredible. I eat regular whole foods. I eat meat, chicken, fish, eggs, fruits, vegetables, etc. But I also make it a point to eat one or two David bars per day as a snack, which makes it easy to hit my protein goal of 1 gram of protein per pound of body weight. And that allows me to take in the protein I need without consuming excess calories. I love all the David Bronze Bar flavors, including cookie dough, caramel chocolate, double chocolate, peanut butter chocolate. They all actually taste like candy bars. But again, they have no sugar and they have 20 grams of protein with just 150 calories. If you'd like to try David, you can go to davidprotein.com/huberman. Right now, David is offering a deal where if you buy four cartons, you get the fifth carton for free. You can also find David on Amazon or in stores such as Target, Walmart, and Kroger.
Today's episode is also brought to us by Joovv. Joovv makes medical grade red light therapy devices. If there's one thing that I have consistently emphasized on this podcast, it is the incredible impact that light can have on our biology. In addition to sunlight, red light and near infrared light sources have been shown to have positive effects on improving numerous aspects of cellular and organ health, including faster muscle recovery, improved skin health and wound healing, improvements in acne, reduced pain and inflammation, even mitochondrial function, and improving vision itself. What sets Joovv lights apart and why they're my preferred red light therapy device is that they use clinically proven wavelengths, meaning specific wavelengths of red light and near infrared light in combination to trigger the optimal cellular adaptations. Personally, I use the Joovv whole body panel about three to four times a week and I use the Joovv handheld light both at home and when I travel. If you'd like to try Joovv, you can go to joovv.com/huberman. Joovv is offering an exclusive discount to all Huberman Lab listeners with up to $400 off Joovv products.
Okay. So, I want to pin up a few rules so that people can move along this because I think most people and including me who learned about dopamine through neuroscience textbooks and lectures and papers and so forth have been fed this overly simplistic model of expectation versus reward or lack of reward expectation outcome. So, just to remind people, dopamine reward prediction error. If you, the dopamine system loves novelty, especially positive novelty, right? You don't think you're going to have a great meal someplace that turns out to be spectacular versus you're really expecting a place to be great. Your friend says it's terrific and then it's okay. And you get dopamine codes for a lot of the expectation reward relationship.
What you're telling us is that in most scenarios, it's more interesting than that. there's an updating of expectation before the final answer comes in and dopamine is coding for that. I'd like to take this word foraging and apply it to a real world scenario in humans and then maybe we can use a combination of what's known and you'll also tell us where it might be conjecture to kind of paint this picture in an intuitive way for people.
I have a friend and she's on the dating market now. She will occasionally call me and ask me, "How do I decode this text message or this interaction?" I try and offer my support, where I can. The conversations often go something like this. Met so-and-so. They seem really great. They seem really busy and they set a plan for like a month from now. Is that weird? You're like, "All right. Well, you know," and I give my interpretation. I say, "Well, you know, he's nice. They've set a concrete plan. You know, this and that. Like person's busy, you know, this and that."
I also hear the, "Hey, you know, met someone, they're really, really terrific." And I say, "Hey, listen. The last time you said this, like two weeks later, it was how do I get out of seeing this person again?" So, like, go slow. Like, collect data slowly. And I'm not going to say I'm always right, but almost inevitably it's three days later or 3 weeks later it's like, "Oh my goodness, how do I get out of this thing?" Right? So, in some sense, it's what you're saying, right? There's a foraging for a healthy thing in life, a mate. This has happened since the beginning of time, although not with apps. There's updating of expectation based on experience and communication. And I think this is a really beautiful example of foraging in the context of updating expectations because and one could argue what is the final reward. Is it marriage? Is it whatever? Okay, that's subjective. But I think we all can intuitively understand this example either by experience or by observation.
So for someone, this person who gets excited about someone they just met, right? Then meets them and is increasingly excited, but it's unclear where it's going to go, then finds out as life goes that, oh, they're not perfect. There's this thing. Can I live with that? So there the I think of this as like a sawtooth of dopamine going through their system. Is that statement accurate that dopamine and other neuromodulators are encoding the sort of expectation of success or lack of success without actually knowing what the final end point is?
Exactly that and that's the insight of Sutton and Barto and when I first heard about this I learned about it from Peter Dayan when I was a postdoc and we both arrived at the Salk Institute together something about it captured me because all of a sudden it's not this you understand expectation and outcome I mean businesses understand that yeah you're disappointed you expected to have a quarterly return of X and you had Y? That's less. You expected it to be low. It was more. That's But that's really rare. I studied hard. I wanted an A. I got an A minus.
But the reality is embedded in this little simple continuous learning update rule. It's called temporal difference reinforcement learning is the fact that in the world these expectations are going through their own trajectory. And that's what dopamine is coding for. Any learning rule should code for the surprising outcome. You have an expectation for an outcome and either high or low of that. Every learning rule should do that. And the psychologist had that kind of figured out 40 years ago, 50 years ago. But it doesn't quite work because it won't account for the way animals learn. It won't let you chain events.
So for example, if I show a light and go to and train on a reward with an outcome and I use that expectation outcome learning rule, it won't chain back to something that predicts a light. Suppose a sound predicts a light and we know the light predicted the outcome. Now I ask the question, well, what happens to the sound? Well, we know people learn. They'll associate the sound with the outcome. It's Pavlovian. But those learning rules won't do that. They learn the wrong thing. It's just not well appreciated.
Now, back when we were trying to associate that learning rule with dopamine, there was we were mainly working on it in a theoretical way, like if you had a signal, what would it need to look like? Where might you find this in biology? I remember our adviser, Terry Sejnowski, who's been on your show, I think he said something like, "There are these diffuse ascending systems. They deliver these transmitters. you guys go work on that."
Sounds like his episode was spectacularly received.
Oh, great. Well, I mean, and he was it was the most open, inviting environment, but of course, all the problems given out were impossible to solve. And I remember just thinking, what? But the first inroad was realizing that it matched what Sutton had written down not so many years before. Sutton got his PhD I think in '84. I think he published the paper in '88. We were doing this in 1990 and we ran into a guy's data on dopamine signaling Wolfram Schultz. We didn't know him. We ran and we could explain every figure in every paper he published and we just thought, okay, that's not an accident.
Okay, fast forward. We're in generation three now. People doing very fancy, very detailed experiments in rodents where you can control where dopamine neurons are going to fire, when they're going to fire. You can control reward. You can just control a lot of things. And so, it's clearly more than that. It's that and some other stuff, but that's central core. I don't see any good reason to throw away that little explanation there.
Back in 1990, the complaint was, "Well, that's really cool. It matches these traces in a arcane journal of physiology paper. What good is that? Reinforcement learning like that can't learn anything." The problem with that was at the time it was right. Like there were no systems that had done anything amazing. Now they've done everything and it's insane how good it was.
You're talking about the AI. That algorithm that I just described with my hands waving is the same thing that David Silver and the Deep Mind guys did when they made the world champion Go playing program and it beat the world champion and that particular game had expert advice built into it. And they removed all that and then they trained it from scratch. It's called AlphaGo Zero. And then that game was amazing. this item. It's never been beaten. It basically beats the history of Go. And so that as an example, it's such an amazing that's a breakthrough. Anybody that knows that side now, that's the AI side. That's the algorithm. But that same algorithm is installed in your head.
It's installed in the head of a song bird. The interesting thing that's going on now is this kind of convergence, right? They're these little gremlins in your brain stem that run that algorithm. Okay, they've now been externalized and put into a computer program that now does things that supersede us. It's a little interesting convergence. It's the only thing I know of that's sort of crawled out of your mind into a program and now the program is doing things that we couldn't imagine before. And it matches the biology. I mean, you can see this in creatures as old as honeybees and Drosophila and whatnot.
Okay. So, a couple of things. One comment and a couple questions. First comment, everyone should know that when Read says dopamine is responsible for X, Y, and Z, there are many other chemicals in the brain likely involved as well.
Other chemicals. And dopamine has multiple functions.
Yeah. Like anything in biology. We should just embed that's up on the chalkboard now so that if you want to mention it again, you can, but don't feel obligated to. People, we're talking about dopamine through a narrow cone here, but certainly serotonin, acetylcholine, norepinephrine, peptides we haven't even discovered or understand yet are contributing. Dopamine is clearly a major player.
I want to step back to a human example, a non-AI example with the understanding of what you just said, which is that the algorithms that AI is running are based on the same algorithms that neurons in our brain stem are using to deploy
No known example exists of discovering the nature of a biological algorithm, externalizing it in code, and then achieving giant breakthroughs using that same algorithm, which pumps information back into our heads, creating an interesting recursion.
Biological learning rules given to computers allow computers to outperform human use of those rules, which is spectacular and a little scary.
Dating example: Interactions update knowledge about a person (e.g., new info on Thursday without seeing them, coworker info on Saturday), changing expectations. Dopamine encodes changes in expectation via positive and negative fluctuations.
Prediction errors from dopamine fluctuations signal how motivated to be for a thing, measuring activity across signals (Todd Braver, John Cohen, Matt Botvinick).
On different timescales: fast fluctuations correlate with changing expectations; slower envelope correlates with motivation in experimental psychology setups.
Dopamine drives forward or causes pause in social media, dating, financial environments—not just at the finish line.
Systems like social media or businesses constantly update expectations to keep users in foraging mode (infinite scroll, no final outcome) to sustain engagement.
Probably one insight into why it's in every mobile creature's brain on the planet.
Nervous system keeps pushing forward; achieving a goal (drug, food, partner) isn't enough.
Mapping to dopamine hits is not wrong but blunt; expectations form before big unexpected hits.
This is why I don't like the phrase "dopamine hits" because it implies it's like a reward that gets trickled into you.
Unexpected rewards augment learning; dopamine fluctuations encode updating/learning and motivation. Surprises provide extra hits—saying so is not wrong, just incomplete.
Speculation on ramping dopamine (e.g., stimulant, not drug of abuse): raises tonic (slower-changing baseline) level, like filling a well with more water so phasic hits sit on top.
Example: Spectacular show on date floods dopamine, raising tonic level, associating positively with person, shaping expectation/motivation.
Disappointing experience lowers levels, changing fluctuation impact on learning.
Parkinson's disease: By symptom onset, 70-75% loss of dopamine neurons in brainstem (only brain source except tiny hypothalamic/pituitary pathway; retinal dopamine unrelated, controls light adaptation).
~80,000 dopamine neurons per side (160,000 total), project throughout brain/spinal cord, making hundreds of millions of connections.
Loss makes signals very noisy: high noise floor relative to signal, no smooth changes.
Dopamine involved in valuing world (value of actions like grabbing/drinking). Parkinson's creates flat value function—can't differentiate value, everything equal, so system freezes (active freezing disease).
Nervous system avoids energy cost of transitioning if no higher value; doesn't habituate to keep behavior going or die.
Not coincidence dopamine involved in learning, motivation, feelings, movement.
Dopamine tonic baseline and phasic spikes predict physical movement (forward/back/side/stay).
Applies to thought movement: sense of urgency (persistent resilient motivation) as need/readiness to move body/thoughts in a direction.
I no longer like the word motivated. I decided I like the word a sense of urgency.
Dopamine stabilizes/sustains brain states (involved in seizures); provides dwell time for thoughts, enables sequencing.
Great groups exploring in mice/theoretical models.
In mobile animals, computes value, motivation, desire; now dissecting computationally (speed, updates) to build artificial systems.
2004 paper by David Redish: addiction as computational disease—drug blocks reuptake, unanticipatable dopamine chases endlessly.
ADHD drugs (Ritalin, Adderall) raise dopamine/norepinephrine, enabling narrower focus/selectivity, not random foraging.
Stabilizes brain states/thought sequences narrowly without diversion.
Bees example (work with Brian Smith, Arizona State): forager bees waggle dance communicates nectar location (decoded language using sun polarization).
Dopamine/serotonin/norepinephrine measured during odor learning.
Spectrum: ADD bees (high octopamine:tyramine ratio, like dopamine:serotonin) get distracted, explore more; concentration bees fly straight to nectar (exploitation).
ADD mode = explorers (new info/nectar); focused = exploiters (get/bring back).
Mind has both modes in same individual (multiple bees in head); distribution varies across people.
Valuable: companies keep lateral thinkers/explorers; need instruction-followers too.
Wiggle in software for motivation/learning effective: explore sometimes, stay on course others.
All have ADHD-like (explorer) and focused (exploiter) modes; some strongly one or other (e.g., linear task-based athletes set/reconstitute multi-scale goals persistently).
Social media short-form video (TikTok, YouTube Shorts) rapid stimulus turnover might strengthen rapid-updating circuits, potentially undermining long-haul goal circuitry.
Builds ADHD muscle; in AI training, balance chase-goal vs. chase-everything to avoid overtraining—impose constraints.
ADD valuable in combat, fighter pilots, situational awareness: train by practicing surprises, hunger, stress for rapid decisions (mental game; e.g., Navy captain examples).
ADHD individuals focus intensely on enjoyable things (e.g., video games); not lack of focus ability, but choppy terrain to narrow mode unless loved.
Rapid-fire decisions need trainable ADD-like capacity.
Overexposure to frequent media (YouTube Shorts: Jell-O house, parakeet) mind-numbing, little long-term learning.
Useful learning: reflect on it later.
Example: Steven Bartlett podcast guest on meaning of life/point of life: "It's to learn to enjoy the passage of time."
Add behaviors that buy more time without undermining the time piece. A short social media clip stuck: learn to enjoy the passage of time. Short clips rarely provide lasting entertainment or information, unlike books where 5-10 things per chapter are underlined for later reflection. Reading a book requires deliberative, intentional actions, unlike effortless scrolling on social media with short-form video content requiring no learning, just thumb movement.
Effortful, slower activities strengthen neural circuits or algorithms. Effort slows processes, aiding learning. In rodents, effort to accomplish tasks is studied. Humans can be set into mental states with words, hunger, or instructions, making human experiments preferable over rodents for this.
Read The Anxious Generation by Jonathan Haidt, who appeared on this podcast and collaborated on a MacArthur network on neuroscience and law. Social media enables rapid comparisons. Speaker has four girls and one boy, unsure of exact effects but suspects negative features. It's like holding back a tsunami; react fly-by-wire to issues like mood depression from specific features. Social media evolves independently.
Child got a cell phone at 13, last in her seventh-grade class, missing Snapchat discussions. Downsides to Snapchat abound. Now managing time requests. Haidt offers prescriptions, but collective action is hard due to defectors.
Train slow, effortful knowledge integration alongside social media. Ideal settings: entertainment vs. educate modes to track time. Social media yields learning, like enjoying passage of time.
In dating, advice to collect data slowly: see someone frequently for two weeks before deciding to avoid wasting time. Put social media on an old phone in a supermax lockbox for 22 hours a day. Paper shows phone upside down in the room lowers cognitive performance by pulling resources; in another room, performance returns to higher levels. Physical distance from distractions feels good.
Can dopamine system learn motivation and pleasure from resisting? Pathologic example: anorexia, where resisting food becomes rewarding, control feels good. Anorexia is the deadliest psychiatric illness. Healthy version: resisting phone for work and presence.
Dopamine encodes reward for resisting as for indulging. Speaker was an accomplished decathlete and wrestler: relished early bed, running hills while others slept, throwing up atop tennis court hill. Wrestling demands: stay calm when air cut off, manage weight, think strategically. Studiers aim to outperform non-studiers, healthily without comparing—live by personal standards quietly.
Kids' K-12 school bans phones during day, surrender on arrival, reactivate at 3:30 PM for rides. Best decision by head of school. Now wrestling with AI integration.
Dopamine fluctuates with constant expectation updating, reward or not. Serotonin often fluctuates oppositely to dopamine in animal and human studies.
Neuromodulators: dopamine, serotonin, norepinephrine, acetylcholine, histamine (15-20 total), plus peptides. Big three for learning/motivation: dopamine, serotonin, norepinephrine. Norepinephrine/epinephrine control attention; serotonin signals active inhibition, preparation to wait.
Example: rodent sees cheese on electrified grid (light on, not super hungry), waits despite desire—active waiting, learning about negatives. Dopamine: positive or absence of negatives.
Group records sub-second dopamine and serotonin in conscious humans during reward tasks, social interactions, visual/emotional tasks. Theme: dopamine and serotonin are opponent—dopamine up/serotonin down for positive/anticipation; reverse for negatives.
At own institution, Rob Malenka's rodent results show learning requires opponency. Humans complicate with ideas generating signals. First concurrent measures in humans confirm opponency everywhere. Old 60s/70s ideas: opponent affective processing, dopamine positive, serotonin negative. Opponency theme in nervous system (retina: color/light-dark to visual cortex).
SSRIs (e.g., fluoxetine, lexapro) block serotonin reuptake, clinical effect in weeks (heterogeneous, life-changing for some). Pushes serotonin into dopamine terminals. Dopamine transporter binds/clears serotonin less efficiently. May negatively condition pursuit of positives.
SSRIs help many but linked to accelerated suicidality, anhedonia tragedies.
Increasing serotonin (e.g., SSRI) makes coding negatives less likely due to less serotonin in serotonin terminals (pushed to dopamine terminals). Tough interaction encoded less bad with abundant serotonin.
20-year-old Neuron paper (John Dani, Penn): SSRIs in rodents → serotonin accumulates in dopamine terminals via dopamine transporter. Need direct human measures (done in surgical patients under ethics).
In economic games (natural rewards via interaction): strong opponency in amygdala/deep structures. Positive anticipation: dopamine up/serotonin down. Losing/uncertainty: serotonin up/dopamine down. Hunger induces state changes preserving opponency.
Hungry humans (epilepsy patients pre/post-meal): testing now. Rodents (amygdala): hunger flips dopamine to encode punishment prediction errors, not rewards—emergency state for survival.
Mark Anderman (Harvard): starvation makes dopamine encode aversive events/errors.
Feed partner before events for enjoyment. Israeli paper: judges' decisions better post-meal. Hunger (stress) flips dopamine to reinforce negatives for survival when positives fail (creek dried, fire ruined foraging).
Reinforcement learning for animal training (lab/dogs): heavy negative feedback bad—leads to overgeneralization (e.g., mall shooting → PTSD, fear spreads). Rational for cataclysmic events.
Bad teaching: ruler slap for wrong fraction denominator. Better: nudges.
A psychiatrist did a rotation with former victims of torture, leading him into the history of political torture. He explained that extreme stress contorts the dopamine reward contingency so dopamine no longer encodes positive things but tries to prevent things from getting worse. Torture methods involve hurting someone a little, threatening much worse harm unless they give information, then not following through. Hurting someone catastrophically, like breaking limbs or bringing near death, leaves nothing left and is ineffective for training. Under sufficient stress, incremental removal of threat becomes highly rewarding after making good on a small promise. This occurs in family dynamics with dialed degrees of punishment.
Excess dopamine from drugs like methamphetamine or cocaine makes everything seem like a good idea and fosters self-obsession. In a Grateful Dead documentary, cocaine was called a "me drug," explaining why music quality declined in the mid-80s. Overindulgence in food or dopaminergic activities resets expectations so few natural events exceed them.
A rescued dog, abused as a puppy, was never right—starting by biting due to inverted world where basic safety was reward and everything else signaled emergency. Extreme abuse leads to learned helplessness. Hurt people exhibit similar traits; drugs of abuse push people into states of poor decision-making despite knowing outcomes.
Dopamine encodes graded positive expectation and rewards. Serotonin encodes negative events. In conditions without true rewards, survival becomes the reward; dopamine adjusts baseline to fight for survival, providing positive pulses for predicting negative outcomes to learn and stay alive. Stress induces this emergency state where positive prediction errors apply to negative events.
Example: A friend in brutal relationships entered a peaceful one but felt like "a cat in a room full of rocking chairs"—tense in calm—for 3 years until safety allowed dopamine baseline to readjust for rewards.
Serotonin moves opposite to dopamine; best data from humans, less consistent in rodents. SSRIs increase serotonin by blocking reuptake; much enters dopamine terminals (2005 paper by John Dani showed 40% difference, blockable by dopamine reuptake inhibitors). This contaminates dopamine signals with negative valence, reducing rewarding properties of good things, impairing positive learning, and potentially registering negatives as rewarding, leading to depression.
Specialized electrodes for deep brain stimulation (DBS) treat Parkinson's or essential tremor (elective surgery via burr hole). Research piggybacks on extra contacts using neural network models to measure dopamine, serotonin, norepinephrine, pH, peroxide from deep brain structures without interfering clinically.
Christina Zelano (Northwestern) snakes FDA-approved ~1mm rubbery tubes up the nose to olfactory epithelium (high near eyeball) for electrophysiology in healthy consented people during eating, meditation, breathing, decision tasks, odors. Nasal recordings mirror midbrain: dopamine rises with positive expectation/reward, serotonin with negative.
Speaker gets up 3:30-4 AM, sleeps ~4 hours young (now less), naps 8-10 PM. First sleep grabs deep sleep/growth hormone; second provides REM. Decided as child to trust own energy needs despite norms. Science is a "contact sport" at the frontier—competitive, attacked ("reviewer two syndrome"), requires self-belief to push knowledge edges with tax money. Divorces (two marriages, five kids) taught non-judgment: nobody is one thing; everyone makes mistakes. Scars from soul-crushing events and science battles appreciated.
Science has a culture with punishing features that build resilience. It's one of the few professions where you must work exceedingly hard to secure resources just to do the work, like having two jobs wrapped into one. Most people, including the speaker at age 10 in Mon, Georgia, have no idea where knowledge and textbooks come from. Part of starting this podcast is to let people interface with scientists, learn from them, and understand the endeavor. Discovering things is awesome but requires real fortitude, which you're not told when joining. Science is an apprenticeship: you sit by someone great at X, absorb them doing little things— not like school grades.
You've benefited from firebrand intuitive people, typically not paying for yourself initially. Then you go out on your own, and it's bracing. The American system is hard on each other; Europeans on American review panels say Americans slap everybody on the back compared to Europe. On study section review panels, as a regular member, you eliminate 70% of grants, advocating for ones you like and finding reasons to dislike others due to insufficient funding. Things are changing with philanthropy and foundations, but it's not for the weak of heart. Minimal sleep need helps; no one succeeds in science without working really hard.
Becoming public-facing kicks your ass, but science makes it feel easier. Science is a thorough ass-kicking with occasional rewards. Plus, it's biology—even when right on Tuesday, things change.
Doing a PhD and biology experiments teaches a reward expectation-motivation contingency loop based on everyday things and long-term goals. Healthy humans register small rewards like a great cup of coffee and serotonergic setbacks like failed experiments, gaining motivation when things work without expecting the PhD as the only reward. Life now involves not expecting great things daily or weekly, trained by big rewards every couple years from publishing papers.
For typical people with mixed easy/hard experiences, registering wins sustains motivation, registering losses avoids stupid mistakes. You must sustain losses and get up again. Sports for kids teach effort-reward contingency and learning to lose even when giving your best—somebody's better. That's a template for lessons. In science labs, students show results, feel sad when failing, but evolve and get better.
Graduate students are worthless for a long time in terms of data output—they have much to learn before interpreting experience. But young fresh people are valuable for energy and great thoughts. Then they transition to the most valuable person in the lab, and six months later they leave, breaking your heart—just like they did to their advisors. Speaker's students: one at UT, one at University of Utah kicking butt, one in biotech, one on job market—all phenomenal.
Energy of youth pouring into something with focus, not for money—labs and professors don't pay much.
Insisting kids play at least one sport (speaker's kids did sports, one competitive dance) removes phone problem—you can't be on your phone. If really tired, hard to get in trouble; soccer players run and are shot at day's end. Sports challenge in ways other life elements don't. Wrestling teaches managing rising panic: don't panic, think about position. Only socially acceptable setting to learn that—nobody chokes you in office to teach thinking under stress.
Losing when you did your absolute best in front of people is amazing. Track and field: only one winner. Daughter at mid-Atlantic middle school meets with 40 teams, 60 kids per event—getting second is really good, but wanting to win eats at you; sport teaches managing it. These are tests modern world lacks; ancestors defended in bands, had to do awful things. Modern ills from ancient brain managing civilization stress.
Skateboarding taught pain, fail, pain—done in solitude, great for solo lab work. Graduate advisor alone in lab: don't burn down, kill yourself, drink tetrodotoxin. Had lab accidents from long late hours. Failure-discomfort-failure pattern like skateboarding: as hard as this is, not as hard as falling on concrete. Snowboarding: snow softer than concrete.
Expectation of effort worth depends on interpreting pain episodes or rewards before success. Cognitive control lets you intend action, inhibit instincts to quit, sustain aversion signals. Transfers when older; trained athletes succeed. School in Rowan Oak, Virginia has no-cut policy—45 kids on tennis team last year, all compete at their level, win/lose. Trains expectations, disappointment, elation, recovery.
Meditation: close eyes, focus internally on breathing or forehead (or walking/open monitoring). Perceptual exercise directing inward vs. outward. Consequences on dopamine and serotonin?
Experiments with graduate student Nishka Raheda (Ohio State, Jeff Schenberg's lab at NIDA intramural Baltimore): neural basis of mindfulness. She's a meditator. Did breathing experiments first: recording from amygdala, anterior cingulate cortex, hippocampus during structured breathing (inhale 1-2-3-4, hold, exhale) vs. free-form breathing in epilepsy monitoring unit at Banner Hospital Phoenix with Robert Bina.
Norepinephrine and dopamine cycle with breathing cycle from deep brain structures. Easy breathing like metronome; amplitude of neurotransmitter fluctuations follows inhale-exhale—feels like watching brainstem work. Structured breathing: transmitters wiggle/wobble, hard to follow.
In healthy people, probe up nose shows same. Instructed breathing engages cognitive control, differentially adequate.
People play ultimatum game (take-it-or-leave-it): $20 split, e.g., $8 you/$12 me—you accept (get $8) or reject (nobody gets anything, cost to you). At 80/20 split, 50% accept, 50% reject to signal inequity.
Nasal probe: breathing registers with peroxide signal (mitochondrial function proxy), dopamine, norepinephrine. During exchange needing learning/model update, max breathing tracks oxygen need for ATP to make dopamine-allowed update. Amazing data.
Dopamine as currency: assigns common value to dissimilar objects, like trading cups for windshields—easier with agreed currency (USD, euro, bitcoin irrelevant).
Provides resilience: online, people rob message with potshots. Science has legitimate/illegitimate critique. Don't let them rob your dopamine/motivation currency. Winner has energy for more winning; losers dissolve or rob others. Zero-sum vs. non-zero-sum games. Media/political/social dynamics: battle over resources ratcheting to this molecule.
Dopamine turns on mitochondria—binds outside to monoamine oxidase, jumpstarts electron transport, signals ATP availability. Direct connection. Neuroscience misses link between dopamine algorithms/computations and combustion.
Forehead warm to touch; server chip burns finger, no AC and bursts into flame—generate heat. Market for chips doing same computing on 40% power; server centers inefficient. Don't understand running brain on 20-23 watts.
Sleep/meditation rejuvenative: even low sleep needers become different after two sleepless days—adenosine/inflammation up, motivation down, dopamine dynamics change. What replenishes currency? Combination physiological responses and algorithmic cleanup. Computational device: erase, not save/consolidate all, homeostasis, recovery—recycles transmitters, rebuilds. Nothing like motivation after great sleep; changes event interpretation. All animals sleep (echidna has REM).
Dopamine and neuromodulators in time perception. Artificially increasing dopamine with drug changes time perception. Latent in dopamine descriptions (psychology/algorithms): timing. Learning = suppressing predictions of what/where/when/how much. Many clocks: every cell has multiple, not just suprachiasmatic nucleus for daylight cycles. True for dopamine signals.
Clocks must register time for events, learning particular times that require rejuvenation and reset. Rodent interval timing literature teaches animals to anticipate events in seconds; dopamine plays a critical role. London group uses dopamine manipulations and dopaminergic drugs in humans to examine time perception changes.
No simple statement on whether increasing dopamine pharmacologically makes time move faster or slower; no clear bromide. Cannabis smokers often think long time passed when it was short. People on Ritalin report losing time, likely concentrating longer, involving multiple systems.
Speaker is awful at tracking time on daily order or directions but orients using sun. Knows body position kinaesthetically as former competitive athlete. Stereoblind (about 5% prevalence); couldn't see images in Ramachandran's demo. Pitched baseballs until 15, hurdled, threw javelin, pole vaulted. Kinetic depth effect from head motion may compensate. Left-handed until age 8, mother forced switch to right due to classroom desks.
Driven, obsessive people (ADHD-like) track time poorly, immerse fully, discard world thoughts. Modern generation with social media/smartphones bombarded by info; emerging data shows altered daily time tracking and life arc envisioning, harder for long-term goals. Dopamine mapped to short-term contingencies. Speaker grew up wandering forests hours unsupervised, seeking info effortfully; calmer than kids' noisy environments. Not necessarily pathological; adults worry but unsure what to do.
Uses Claude AI for research: directs to literature, summaries, compare/contrast (e.g., Read Montague's dopamine view vs. others), five-person panels. Loves clean interface, aesthetics.
Speaker works at AI-neurobiology interface, using algorithms to organize biological observations. Neural network training scaled unexpectedly. AlphaFold (DeepMind, Google) solved protein folding; Jumper and Hasabis Nobel for mapping DNA sequences to protein structures, using reinforcement learning like AlphaGo/AlphaGo Zero (trained Go grandmaster, chess from scratch in 30-40 days). NIH spent ~$100B over 70 years on this. Historical breakthroughs; algorithms in our heads for handling reality.
Neural reinforcement learning will explode; engineer learning/motivation/reward/contingency/dopamine. Commercialize probes up nose (Seth Batten's Nebula Neuro) for personal serotonin/dopamine hacking via cell phone, watch responses to Scrabble, scenarios. Experiments: sentences with valence changes word-by-word, social exchange games thinking self/others.
Example: kids learning concentration; real-time readout of dopamine/serotonin, servo on release while reading passage, suggestions to pay attention or lower sigma. Take 1000 people, train neural net on performance/transmitters step-by-step. Will dominate neurobiology, especially mental illness. Networks learn unknown things, scary/exciting. Use on all neuroscience data, especially humans.
Reinforcement learning, AI, brain-machine interfaces mocked 30 years ago (few posters at SfN); now hot. Molecular/genetic tools awesome but AI/BMI rage. Engineer into brain, not just pills. Parallels: meditation, breathwork, psychedelics comeback (caution: psychotic episodes); peptides/GLPs. Whatever beaten now becomes next big thing; Hinton persisted despite mockery.
Neural nets transitioned from learning nothing to everything (e.g., describe images poetically). Proof: persist in loved areas, world aligns eventually. Hinton has psychology PhD.
Human brain adapts repeatedly; younger generation on phones does well (homework, sports, life) despite concerns.
Public dopamine hits: Unexpected rewarding causes dopamine fluctuation (true, but incomplete story).
Serotonin hypothesis depression, dopamine schizophrenia: Oversimplification; both fluctuate in both. Schizophrenia: blocking dopamine receptors reduces symptoms (e.g., voices); L-DOPA induces voices/paranoia at high doses in non-schizophrenics/Parkinson's. Ill-defined; recording transmitters in healthy/sick will tease apart.
Neurochemical drive to persist (dopamine pursuit) vs. valuation (serotonin). Expectations set/update state-to-state control fluctuations; poorly understood representations from memory/brain states. AI will help next 20 years.
Grok AI response: Dopamine drives grit via effort/reward anticipation; high levels trap in sunk cost. Serotonin balances valuation; low ratio unhealthy persistence. Cites studies on dopamine surges overvaluing sunk efforts.
Speaker: Grok did well but missed electrochemical nature, networks setting expectations defining release (half the story).
Dopamine ≠ pleasure.
Serotonin: Drugs heterogeneous, often pathological/toxic over decades; unfortunate treatment moment. Identify SSRI candidates ahead (many swear by, minimal sides; 50-80% placebo in psychotropics with good outcomes). Poor understanding of how beliefs marshal brain changes.
How dopamine changes removing external rewards, relying on internal (meditation, private drawing)? No data, but fantastic to measure, e.g., sensory deprivation generating chased internal states. Unpublished schemes for this.
Many asking: greater influence exogenous stimuli/feedback or psychological framework?
Ability to generate a clear expectation and hold it in mind consciously links to motivation. Questions about creating capacity for persistent motivation under conditions where things aren't going well. Example of running up a hill and puking as self-training. On Friday nights after football games, ran until vomiting: "snot and vomit," went until couldn't go anymore. Copied Russian training from books because Caucasian sprinters in Olympics from Eastern Bloc countries won; unaware they were on steroids. Did plyometrics, weight box jumps with weight vest until threw up. Worked out this morning, felt nauseated. Discussion of mellow workouts declined. Personal workout moments alone, unless with Dorian Yates.
Met Arthur Jones from Dand, Florida, at Nautilus place in Sandy Springs, Atlanta. Worked out with Isaac Hayes (Black Moses), who wore dark shades and giant gold chain. At age 12, allowed in Nautilus gym. Isaac Hayes nicest guy, recently deceased. High-intensity work encouraged by Arthur Jones best for stimulating hypertrophy via supersetting. People carried buckets because threw up so much on high-rep, high-intensity leg day.
Many questions about serotonin syndrome. People on SSRIs experience sexual side effects, anhedonia, motivational issues. Cause: drugs bind to all kinds of receptors. Many serotonin receptors (around 80), unlike fewer dopamine receptors. Leads to side effects. Idea of being on a drug to manipulate mood state itself affects mood.
Thank you for taking time from busy family, work, and workout schedule despite little sleep. Wonderful opportunity to learn about dopamine, serotonin, and neuromodulator dynamics from expert understanding past, present, and future. Conversation up-to-date and nuanced. Not complete, will have back. Thanks for pioneering: tackling technical challenges, AI, computational neuroscience, putting people in scanners, wires up noses, into brains. From 15 years ago, clear goal to solve questions by any means. Spirit of science at its best. Links to work, writings, talks provided.
It's a blast.
Thank you for joining discussion with Dr. Read Montague. Links in show notes. Subscribe to YouTube channel (zero-cost support). Follow on Spotify and Apple, leave up to five-star reviews and comments. Check sponsors. Questions/comments in YouTube comments (all read). New book: Protocols: An Operating Manual for the Human Body. First book, 5+ years work, based on 30+ years research/experience. Protocols for sleep, exercise, stress, focus, motivation with scientific substantiation. Pre-sale at protocolsbook.com. Follow Huberman Lab on Instagram, X, Threads, Facebook, LinkedIn for science tools. Neural Network Newsletter: zero-cost monthly with podcast summaries, 1-3 page PDFs on sleep optimization, dopamine, deliberate cold exposure, foundational fitness (cardio, resistance). Sign up at hubermanlab.com (menu > newsletter); email not shared. Thank you for interest in science.
Keep Andrew Huberman in your library
Save the videos and channels worth coming back to, and find them again in one place.
