Dr. Reed Montague is the director of the Center for Human Neuroscience Research at Virginia Tech and a pioneer in developing methods to directly measure dopamine and other neuromodulators in humans in real time. He joins Andrew Huberman to discuss the computational nature of dopamine, serotonin, and learning algorithms.

The conversation explores how dopamine functions as a learning signal that constantly updates expectations before final outcomes arrive, challenging the simple 'dopamine equals pleasure' narrative. Montague explains how the same temporal difference algorithms operating in our brainstem have been externalized into AI systems that now surpass human performance in games like Go and applications like protein structure prediction.

They discuss the opponent relationship between dopamine and serotonin, the surprising effects of SSRIs on reward processing, and how stress states can flip dopamine's role from encoding positive to negative prediction errors. The discussion also covers Montague's groundbreaking work measuring neurotransmitters through nasal electrodes in healthy humans, social media's impact on learning circuits, and the intersection of breathing patterns with neurochemical fluctuations.

Dopamine as Learning Algorithm, Not Just Pleasure Signal

Dopamine encodes temporal difference errors - the difference between successive predictions as you move through the world, not just final expectation versus outcome comparisons.

"Reality doesn't give you feedback like that every time. Reality often gives you long stretches of nothing" - Reed explains why continuous prediction updating is more realistic than simple reward-outcome learning.

The Sutton-Barto temporal difference algorithm, installed in brains from honeybees to humans, allows learning from ongoing expectation changes before any terminal reward arrives.

This same algorithm powers DeepMind's AI breakthroughs: "They used the Sutton and Bartow algorithm. They trained those systems where players would make hundreds of board position changes before you ever got to the end of the game."

Serotonin-Dopamine Opponent System and SSRI Mechanisms

Dopamine and serotonin work in opponent fashion throughout the brain: "When dopamine goes up, serotonin goes down. When serotonin goes up, dopamine goes down."

Serotonin signals active waiting and learning about negative events, while dopamine encodes positive expectations and rewards.

SSRIs push serotonin into dopamine terminals via dopamine transporters, potentially reducing the rewarding properties of positive events by flooding reward circuits with 'negative juice.'

A 2005 study by John Danny showed 40% of excess serotonin from SSRIs enters dopamine terminals, which could explain why some people on antidepressants struggle to feel rewarded by positive experiences.

Stress States Flip Dopamine's Role for Survival

Under extreme hunger or stress, dopamine switches from encoding positive to negative prediction errors: "If you make a rodent hungry, then you can show that dopamine will encode something like punishment."

This flip serves survival: "If you get to a state where you're really starving, things have not been going well for a long time. You've been making really bad decisions."

The system adaptively prioritizes threat detection over reward seeking when resources are scarce, using the same neural machinery but with inverted meaning.

Recovery from trauma or abusive relationships requires time for the dopamine system to readjust its baseline from survival mode back to reward-seeking mode.

Revolutionary Nasal Electrode Measurements in Humans

Montague's team developed minimally invasive nasal electrodes that can measure dopamine and serotonin in healthy humans during real-world tasks like economic games and breathing exercises.

"Dopamine and norepinephrine cycle with the breathing cycle. It is like a metronome. The amplitude of the neurotransmitter fluctuations follows the inhale-exhale cycles."

During economic decision-making games, breathing patterns track with peroxide signals (mitochondrial function), dopamine, and norepinephrine when people need to update their models of fairness and cooperation.

This technology could eventually allow individuals to monitor their own neurochemical responses to different activities via smartphone apps, revolutionizing personalized neurofeedback.

Social Media's Impact on Learning and Attention Circuits

Montague worries that constant exposure to short-form video content may strengthen ADHD-like circuits while weakening focused, goal-directed pathways: "You build your ADHD muscle."

He references The Anxious Generation by Jonathan Haidt, noting concerns about social media's effects on children's ability to envision long-term goals and tolerate delayed gratification.

Natural foraging requires both exploration (ADHD-like) and exploitation (focused) modes: "You need both. You need the explorer and the ones that fly right to the nectar source."

Activities requiring effort and slower processing, like reading books, may strengthen circuits for sustained attention and deeper learning compared to passive scrolling.

AI-Biology Convergence and Future Applications

The same reinforcement learning algorithms in biological brains have been externalized into AI systems: "It's the only thing I know of that's sort of crawled out of your mind into a program."

DeepMind's AlphaFold solved protein structure prediction using game-like reinforcement learning, a problem the NIH spent "probably $100 billion on for the last 70 years."

Future applications could include real-time neurofeedback for learning optimization: "You could give them that probe and they could read a passage and get real-time readout of dopamine and serotonin."

Neural networks will increasingly analyze human neurotransmitter data to identify patterns of optimal learning, attention, and decision-making across thousands of individuals.