How Dopamine & Serotonin Shape Decisions, Motivation & Learning | Dr. Read Montague

KEYPOINTS

• 01

  Dopamine is primarily a learning signal, not a pleasure chemical

  Montague, a computational neuroscientist, says the old 'dopamine equals pleasure' view has given way to dopamine as a learning signal. Its fluctuations control learning, and it also plays roles in motivation, movement and possibly feeling states. He stresses that how dopamine's effects on learning relate to how you feel is much less well understood.

• 02

  Dopamine encodes changes between successive predictions, not just outcome versus expectation

  The 1972 Rescorla–Wagner rule compares an outcome to an expectation, but reality mostly gives long stretches with no feedback. Rich Sutton and Andy Barto's temporal difference rule learns from the gap between one prediction and the next. Using Sutton's example: on Wednesday you predict two inches of rain on Saturday, and on Thursday you update to ten inches. That change in expectation, before any outcome arrives, is what dopamine appears to encode, and the same rule lets learning chain back through cues.

• 03

  The brain's learning algorithm is the one behind modern AI breakthroughs

  Around 1990, Montague, Peter Dayan and their adviser Terry Sejnowski found that Wolfram Schultz's recordings of dopamine neurons matched Sutton's algorithm 'figure by figure'. The same temporal difference approach later underpinned DeepMind's Go programs, including AlphaGo Zero, trained from scratch without expert input. Montague calls it the only case he knows of an algorithm 'crawling out' of the brain into software that then surpasses us. It is seen in honeybees, fruit flies and songbirds as well as humans.

• 04

  Dopamine both teaches and sets how motivated you should be

  Following Todd Braver, Jonathan Cohen and Matt Botvinick, fast dopamine fluctuations carry the prediction errors that drive learning, and a slower envelope across them can indicate how much to want something. Huberman notes that infinite-scroll platforms exploit this by constantly updating expectations so you stay in foraging mode. 'Dopamine hits' are real, but Montague calls the phrase 'blunt' and incomplete.

• 05

  Parkinson's shows dopamine's role in valuing actions

  Each side of the brainstem has roughly 80,000 dopamine neurons that make hundreds of millions of connections. By the time Parkinson's symptoms appear, 70–75% are lost, so the signal becomes too noisy to read. Montague describes Parkinson's as a 'flat value function': nothing looks more valuable than anything else, and since moving costs energy, the system 'actively freezes'.

• 06

  Dopamine and serotonin act as opponents in humans

  Montague's group says it is the only one recording sub-second dopamine and serotonin levels in conscious humans, piggybacking with consent on electrodes placed in patients for surgery. In economic games, when things go well dopamine rises and serotonin falls; when they go badly the reverse happens. Serotonin seems linked to learning about negative events and to 'active waiting', and Rob Malenka's rodent work shows learning requires this opponency.

• 07

  SSRIs may push serotonin into dopamine terminals

  A roughly 20-year-old paper in Neuron (senior author John Dani) found that after weeks on an SSRI, rodents' dopamine transporters were pulling serotonin into dopamine terminals. Montague's hypothesis is that putting the 'negative juice' in 'positive' terminals could blunt reward from good events. He emphasizes that SSRIs have helped many people and are a life-changer for some, that the response is highly variable, and that human follow-up is lacking.

• 08

  Hunger and severe stress can flip dopamine's meaning

  Mark Andermann's rodent work at Harvard shows that in starvation states, dopamine in the amygdala encodes punishment prediction errors rather than reward. Montague frames this as rational in an emergency, when staying alive matters more than chasing rewards. He also cites an Israeli study showing judges' rulings varied depending on whether they had eaten.

• 09

  Harsh feedback is a poor teacher because it causes overgeneralization

  A cataclysmic negative event teaches the nervous system to fear anything resembling the context, which is the PTSD pattern. Montague argues that rulers on knuckles or very negative training of animals teach badly; small nudges work better. His abused rescue dog's world stayed 'inverted', with basic safety acting as the only reward.