413 ‒ The neuroscience of emotion: challenging conventional theories of emotion and much more
In a Nutshell
Lisa Feldman's core argument is that emotions are not hardwired, universal, or biologically distinct categories with objective biomarkers. Instead, the brain constructs emotional experiences through predictive processing, using past memories to anticipate and interpret sensory signals in service of metabolic regulation. This challenges conventional emotion theories, explains why individual variation is the norm rather than the exception, and suggests that emotional change comes from practicing new predictive patterns rather than willpower or insight.
These notes were generated by AI and may contain inaccuracies.
Peter Aia welcomes Lisa to the Drive podcast. Both grew up in the same city and have lived in their current homes longer than their previous homes.
Lisa attended graduate school to become a clinical psychologist rather than going to medical school like her peers. She began learning about emotion immediately in her studies and initially believed she had a solid understanding of emotions and how they work.
Lisa was conducting PhD research on a topic other than emotion but needed to measure emotion using standard methods - asking people how they felt. She attempted to replicate published peer-reviewed experiments as part of her graduate training but was unable to replicate the findings. Over a three-year period, she tried to replicate eight different experiments multiple times without success.
Lisa's initial response was to conclude she was not cut out to be a scientist. However, upon closer examination of her data, she discovered she was consistently replicating her own findings, which suggested something interesting was occurring. Her research advisors were willing to be persuaded by her observations.
The studies involved asking participants to rate emotions using adjectives or statements about symptoms. Lisa was particularly interested in anxiety and depression, using clinical scales including the Beck Depression Inventory and Beck Anxiety Inventory. Participants rated the extent to which statements described their experience.
The experiments she attempted to replicate examined self-concept, comparing beliefs about who one is to who one ideally wants to be or who one ought to be. The theory predicted that failing to match one's ideal self would produce depression, while failing to meet "ought" standards would produce anxiety.
Lisa found that participants who reported feeling anxious also reported feeling sad or depressed, and participants who reported feeling happy reported absence of anxiety and feeling calm. Despite having the opportunity to be granular and specific, participants essentially reported feeling bad or feeling okay.
Lisa had been taught that emotions like anger, sadness, fear, depression, and anxiety are biologically distinct categories. The prevailing assumption was that human infants are born with circuits embedded in ancient brain regions for specific emotions. The belief held that these are universal, inborn states that can be objectively measured without relying on self-report.
The assumption was that objective measurement meant not using the subject's experience. Scientists believed they could measure blood pressure, heart rate, brain activity, or facial movements to determine emotional states. The theory posited that these biologically distinct categories have boundaries in nature that humans discovered rather than created.
After examining the emotion literature directly rather than relying on summaries, Lisa realized there are no objective measures for diagnosing emotional states. Scientists have been debating this question for 150 years, with philosophers debating it even before the scientific revolution. When a question persists for such a long time, it often indicates the question itself is wrong.
Lisa's clinical training proved valuable because therapists learn to identify when people are lying to themselves. She recognized a massive case of a story not adding up - a set of beliefs about emotion in the research literature that contradicted the actual evidence in results sections.
Lisa changed her dissertation topic entirely, taking an extra year to finish. She abandoned her plan to become a therapist and focused exclusively on research.
Research meta-analyses show that in Western cultures, adults in urban settings scowl about 35% of the time when angry. This means 65% of the time people do something else with their face when angry. Conversely, half the time people scowl, they are not angry.
Using scowling as an objective measure of anger has poor predictive value - missing 65% of cases and being wrong half the time. The same pattern holds for every emotion category regardless of measurement method.
Amygdala activity increases during fear about 30% of the time, but often does not. Amygdala activation occurs in many instances of emotions that are not fear and across other phenomena.
When angry, people may scowl, frown, smile, laugh, remain silent, or plot revenge. Blood pressure may go up, down, or stay the same depending on the situation and what action comes next. Anger can sometimes feel pleasant and can either distance or bring people together. Cultural differences add another layer of complexity.
Scientists simplify complex reality to study phenomena. One approach captures as much variance as possible and models its structure. The other approach limits variability - the typical method in emotion research up to that point.
Experiments limited the number of emotion categories studied, the number of instances studied, the sampling of people and stimuli, and used statistical modeling to remove variance as error variance.
The first question is whether emotions can be objectively measured without self-report. The second question concerns what objectivity means - measuring something in a subject that the subject does not report.
Measurements are never truly objective because a scientist's observation is the scientist's experience. All experiments are theory-laden in their design and statistical analyses. Double-blinding cannot remove theoretical views from studies.
Consider an experiment testing whether a pill lowers cholesterol using 100 participants, half receiving the pill and half receiving placebo. Even with randomization and blinding, other factors like pill color, size, subject sleep, diet, and other variables affect outcomes.
Randomization removes certain variance from modeling but does not remove variance from the causal complexity causing the phenomenon. Effects observed in large samples may not predict outcomes for any individual subject.
Rather than relying on averages, researchers can quantify levels of factors thought to matter and run experiments at different values of those factors. This reveals how treatments work under specific conditions.
Immunologists, psychologists, and physiologists placed participants in hotel rooms, quantified everything about their condition, and exposed each to identical viral loads. Between 20-40% showed symptoms of respiratory illness, meaning 60-80% did not become ill despite exposure.
A virus is a necessary cause of respiratory illness but not a sufficient cause. Most people think in terms of simple causes, but complex causation involves multiple necessary but insufficient factors.
In a hypothetical clinical trial where 30% improve on placebo and 40% improve on treatment with no statistical difference, the conclusion might be that the drug is ineffective. However, this could be wrong if the drug works for 35% of people due to complex causation.
Lisa shared a personal medical experience involving thyroid disease that illustrates the limitations of relying on group averages in clinical research. When diagnosed with Hashimoto's thyroiditis, she discovered through her own research that synthetic thyroid hormone could arrest disease progression in some individuals, despite the absence of large-scale clinical trials. After discussing this with her physician, they conducted a personal experiment with low-dose levothyroxine. The treatment not only arrested the disease progression but reversed some damage, including the disappearance of thyroid nodules and a reduction in antibodies. This outcome transformed her condition from a progressive disease requiring eventual thyroid removal to a manageable borderline case requiring only a low maintenance dose.
Lisa emphasized this was not presented as a universal solution or magic cure, but rather as evidence that averages hide meaningful individual variation. She stressed that clinical trials remain valuable, but that the approach of examining variation rather than compressing it through group averages represents a different scientific paradigm.
Peter Attia described his framework of Medicine 2.0 versus Medicine 3.0, where Medicine 2.0 relies on randomized controlled trials that homogenize heterogeneous patient populations, apply uniform treatments, and derive recommendations from averaged outcomes. While acknowledging this approach is superior to previous medical practices, both speakers agreed that the critical limitation is the loss of individual variability and nuance.
Peter provided the example of ezetimibe (Zetia), a cholesterol-lowering drug that blocks intestinal and hepatic cholesterol transporters. Clinical trials showed underwhelming results for ezetimibe as monotherapy, leading to debate about its evidence-based use. However, understanding the drug's mechanism allowed identification of biomarkers for cholesterol absorption, enabling targeting of high cholesterol absorbers who respond exceptionally well to monotherapy. This represents precision or evidence-informed medicine that accounts for individual variation rather than dismissing treatments that don't work on average.
Lisa explained that unlike cholesterol absorption or other measurable physiological markers, the science of emotion lacks biomarkers entirely. There is no objective measurement analogous to the cholesterol example discussed. She argued that brilliant researchers could develop alternative scientific approaches that don't rely on group averages, similar to how the Large Hadron Collider represented massive international cooperation and funding to answer fundamental physics questions.
Lisa noted that behavioral science and neuroscience lack the resources and coordination seen in physics projects like the Higgs boson discovery. She suggested that current technology could enable different experimental approaches if the scientific community had the ideas, will, and funding to pursue them.
Two independent observations emerged from Lisa's research: emotions cannot be measured objectively through biomarkers, and emotions are not hardwired universal states identical across individuals. These conclusions required approximately 20 years of empirical investigation, including multiple retrainings in measurement techniques.
Lisa described her methodical approach to published research: reading only the hypothesis, methods, and results sections while discarding introductions and discussions to avoid theory-laden interpretations. She personally collected data rather than relying on other scientists' summaries.
Lisa's lab conducted extensive fieldwork across diverse populations to test emotion universality claims. Research sites included Namibia with the Himba people, Tanzania with the Hadza hunter-gatherers, and rural China. Multiple measurement methods across these populations consistently revealed that variation is the norm in emotional expression and experience, though this variation contains meaningful structure rather than being random.
Peter drew a parallel between emotion biomarkers and aging biomarkers, noting that while aging is universally acknowledged and people can intuitively assess relative aging rates, no definitive individual-level biomarker exists to measure aging rate or intervention effectiveness. On average, certain changes occur with aging (bone density loss, muscle mass reduction, insulin resistance), but individual measurement and intervention assessment remains impossible with current methods.
Lisa proposed that personalized AI systems could potentially learn to predict an individual's specific emotional vocabulary across anger, sadness, fear, and other categories through repeated measurements across many situations. This approach would require modeling individual variability first, then empirically identifying similarities across people rather than starting with group-level simplifications.
She noted that while generalized algorithms might eventually emerge, the starting point must be individual modeling rather than variance compression. This represents a more expensive and cumbersome research approach with significant implications for scientific incentive structures and funding priorities.
When asked to define emotion, Lisa responded that the question itself may be problematic. Instead, she described the brain's primary function as regulating and coordinating bodily systems efficiently. This regulatory function, rather than any psychological purpose, represents the brain's most important job, with all psychological processes serving this metabolic coordination.
The brain receives signals from the body's sensory surfaces and runs predictive models of these signals rather than directly modeling the external world. Since the brain only accesses signal outcomes rather than their causes, it must make inferences about what generated each signal - whether a loud bang, glucose change, or other input.
Lisa expressed reluctance to speculate extensively about consciousness, preferring to avoid unsubstantiated claims about how electrical and chemical changes produce sentience and conscious content. She noted that many questions about emotion can be addressed without resolving consciousness questions.
However, she emphasized that most bodily processes remain unconscious, which is necessary for functional attention to external events. The brain receives compressed, gist-level information about metabolic and interoceptive signals rather than detailed visceral mapping. Unlike the retinotopic mapping in primary visual cortex (V1), no equivalent spatial map of the body exists in the brain.
Lisa concluded by noting that the traditional distinction between exteroceptive (world-originating) and interoceptive (body-originating) signals may be less important than whether senses are proximal or distance-based in nature.
Distance senses like vision and audition undergo different signal processing and compression compared to proximal senses such as olfaction or gustatory signals. Gustatory signals from the gut are considered interoceptive. Some signals like proprioception do not fit cleanly into the external versus internal distinction. When focusing on signal processing properties rather than traditional categories, different distinctions become important for understanding how thoughts and feelings arise.
No existing framework explains the transformation from high-dimensional chemical and electrical signals to low-dimensional mental features like pleasantness or unpleasantness. Claims to have solved this transformation are likely self-deception.
Fear is often considered essential for species survival and predator avoidance. However, the assumption that fear is inherently unpleasant is challenged by evidence that people voluntarily seek fear-inducing experiences such as roller coasters and horror movies. The speaker ran a haunted house for charity for 13 years, offering three levels of scary experiences.
Animals engage in play behaviors that may serve as practice for foraging and innovation. While it is impossible to access what animals subjectively feel, behavioral evidence can indicate unpleasant experiences. Studies show animals will work hard and expose themselves to predators to obtain opioids after tissue damage, suggesting they experience something unpleasant that they seek to reduce.
The prototypical fear response varies across species and contexts. While Western theory emphasizes fight-or-flight, some cultures respond to fear by going to sleep. Invertebrates in water may play dead by slowing heart rate to avoid shark detection. The brain does not react to stimuli but instead makes predictions that begin as motor predictions, including visceral motor predictions to support skeletal movements.
An emotion represents a set of predictions telling a story about what sensory signals mean in a particular situation. This definition applies equally to emotion, cognition, and perception. Some cultures do not separate emotions from cognitions or bodily states, treating them as integrated categories rather than distinct phenomena.
Western culture maintains firm distinctions between emotional and physical states, reflected in separate medical disciplines like neurology, internal medicine, and psychiatry. Some cultures view thinking and feeling as unified features of the same state, comparable to how blue and green together create turquoise. Legal systems in some cultures do not distinguish between culpability based on feelings versus actions.
A valid theory must explain how brains and bodies create minds across all human cultures, not just Western categories. Emotions, thoughts, and perceptions may be phenomena requiring explanation rather than explanatory categories themselves. The brain's primary priority appears to be supporting the body physiologically and metabolically, functioning as a predicting machine where emotions arise from predictions about physical sensations.
Evidence from anatomy, physiology, brain evolution, evolutionary biology, developmental neurobiology, and signal processing engineering all converge on the conclusion that brains predict rather than react. This concept appears across cultures and has existed in Western thought for centuries.
The experience of the brain as a reaction machine, metaphorically described as riding a bull, an elephant, or controlling horses, feels intuitive but lacks empirical support. The triune brain concept has been contradicted by molecular genetics evidence for at least 50 years. Experience itself is not evidence of how brains work, representing a form of naive realism that science progressively moves beyond.
Primary visual cortex receives only 5% of its synapses from thalamic inputs carrying retinal signals. The remaining 95% come from other cortical areas, representing top-down rather than bottom-up processing. When rodent retinas are ablated, V1 neurons initially quieten but return to normal activity levels after 24 hours, driven by motor-related signals from areas like the anterior midcingulate cortex rather than visual input.
Action potentials lack intrinsic meaning—their significance depends on which neurons send and receive them. The same signal functions as a motor command in the spinal cord but as a sensory prediction in V1. Studies with blindfolded humans learning Braille demonstrate that V1 processes tactile signals during reading, and disrupting V1 with TMS impairs Braille reading ability.
A documented case involves a woman congenitally lacking a somatosensory system who could only eat by watching herself in a mirror, using visual signals to control motor movements including swallowing. This illustrates that sensory signals primarily serve motor control rather than simply providing knowledge about the world.
Motor signals and their copies propagate throughout the brain, beginning to construct experience before other sensory signals arrive. The brain can generate an experience of red even when no corresponding wavelengths of light hit the retina. This occurs through prediction based on countless statistical regularities accumulated over time.
A demonstration involves presenting a traffic light with the red position appearing gray. The brain predicts red due to learned patterns, and viewers experience red until adjacent gray boxes are added, revealing the actual gray, then removed to restore the red perception.
Prediction signals involve the brain using its own intrinsic signaling to change the firing of sensory neurons in preparation for expected inputs. When predictions match incoming signals, the corresponding neurons stop firing, and experience is generated. Only unexpected signals propagate forward through the visual system from V1 to higher brain regions.
Professional sports demonstrate predictive processing because athletes cannot react fast enough to wait for sensory confirmation. A baseball pitcher throwing at 90-95 miles per hour creates conditions where batters must swing based on predictions about ball trajectory rather than visual reaction.
Predictions derive from memories reimplemented as signal patterns from past experiences for present purposes. Every experience combines remembered past with sensory present, where the sensory present either confirms or corrects predictions. Learning is defined as changing the remembered past through prediction errors.
Practice builds skills through repetition with concrete feedback. Interval training provides an example where unpredictable changes every 30 seconds prevent prediction, making it metabolically more demanding than predictable exercise. Prediction errors require more metabolic resources than actions arising from predictions.
Repetitive practice alone is insufficient without concrete feedback. This principle applies equally to emotional skills, where developing greater emotional granularity requires unambiguous feedback that is difficult to obtain.
Every mental event originates as a motor plan. Predictions begin as motor plans anticipating bodily needs and preparing to meet them before they arise. These motor signals create thoughts, feelings, and sensations through preparation for metabolic and skeletal movement.
The sequence is motor first, then mental experience. Changing actions or circumstances alters thoughts and feelings because these arise from motor preparation rather than preceding it. Shifting attention or moving the eyes can change mental states by altering the underlying motor predictions.
The book "Principles of Neural Design" by Peter Sterling and Laughlin demonstrates that predictive processing is necessary given energetic constraints on nervous systems across different organisms. The brain's high metabolic demand, consuming approximately 20% of energy budget despite being 2% of body weight, reflects this predictive architecture.
Human brains are not more developed than other species but differently developed. The cerebral cortex shows an architectural gradient from primary sensory areas with many small neurons to limbic areas with fewer but larger neurons having extensive connections. Humans demonstrate greater capacity for dimensionality reduction and compression compared to other primates.
This enhanced compression enables seeing similarity across differences, allowing abstraction such as recognizing different objects as serving the same function or understanding metaphorical concepts like social ladders. However, this comes with reduced ability to retain fine-grained episodic detail compared to other primates.
Cortisol functions as a glucocorticoid involved in glucose metabolism rather than being a stress hormone. A cortisol surge occurs when anticipating energy expenditure, such as waking in the morning.
Allostasis is predictive and concerns metabolic efficiency, while homeostasis is reactive and returns systems to set points. Allostasis can be maintained during activities like running a mile but not during extreme disruptions like running a marathon.
A moment of stress metabolically occurs when the brain predicts a significant metabolic outlay, anticipating bodily needs and preparing to meet them. This predictive process, rather than subjective experience alone, accounts for physiological stress responses.
Mood is fundamentally related to metabolic status. When the body spends energy and replenishes what it spends while predicting accurately, distress does not necessarily occur. Distress emerges when metabolic costs cannot be recovered, when prediction fails, or when people are in a sea of uncertainty. Chronic stress does not harm health immediately, but it will eventually harm health over time, like a slow drip that eventually bores a hole in a steel pipe.
A child being abused faces two major sources of distress: unpredictability about when violence will occur and a lack of agency to act. However, there is an additional factor: humans are social animals who manage their nervous systems through each other. We are the caretakers of each other's nervous systems under best circumstances. The best thing for your nervous system is another human, and the worst thing is also another human.
When things work well, adults maintain allostasis for children. Babies cannot nurse without being taught and cannot control their limbs. Adults regulate children's temperature, feed them, clothe them, and cuddle them, thereby regulating their nervous systems and metabolism. Children slowly learn to take this regulation on themselves, but humans continue to regulate each other throughout life.
We create savings for each other using a body budgeting metaphor where the brain runs a budget for the body, managing resources like salt and glucose. People coordinate their actions, breathing, and movements. When people trust and like each other, they coordinate heart rates. We can also disrupt each other, and children in abusive situations do not receive the savings that most people in safe environments benefit from.
In abusive situations, everything costs more metabolically. Experiments show that people carrying a backpack full of rocks burn fewer calories when with someone they like and trust compared to being alone or with a stranger. This means the child in an abusive situation will be metabolically less resilient. Adding vigilance and actual harm creates a perfect storm of metabolic burden that may not show up until adolescence or adulthood.
Adverse childhood experiences (ACE scores) predict bad health outcomes including metabolic illnesses like diabetes and heart disease, though not perfectly. This is a slow process where chronic stress may not harm health immediately but will inevitably harm health eventually.
Modern life is designed to bankrupt a human body budget. We are not all equally affected, but everyone is affected. There are no fight or flight circuits in the brain. The periaqueductal gray contains circuits for regulating the autonomic nervous system throughout life, not specifically for fighting or fleeing. These circuits are active during memory tests that involve no stress in the conventional distress sense.
Modern life creates expanded niches with more things that concern us and affect metabolism due to news and the internet. This creates expanded uncertainty and expanded costs, making people more vulnerable to metabolic illnesses of all sorts.
Individual differences in susceptibility to adversity arise from prediction and memory. Nervous systems adjust themselves to their environment. The brain wires itself to its world, which includes the body. Chronic stress or adversity changes how the brain regulates the body, and the body then sends signals back to the brain that require adjustment.
If there is literal tissue damage and metabolic burden from healing, the brain may continue predicting signals even after they disappear because learning is expensive. This is chronic pain where the tissue damage is gone, but pain persists because all sensations are experienced in the brain, not in the body.
Everything experienced is a combination of the remembered past and the sensory present. To change current experience, one approach is to change the remembered past, though this is difficult. Memory is not different from perception. When remembering something, it is a current experience of something that happened in the past, though not identical to the original experience.
A more effective approach is to change something now and practice that change so the brain can predict differently in the future. Like driving, investing energy to learn something new through practice leads to automatic behavior. When feeling overwhelmed, forcing oneself to play soccer with a child can become easier with advance practice so the brain predicts differently in the heat of the moment.
The example of practicing getting out of a tailspin on ice illustrates preparing for difficult situations in advance. Moments can be taken advantage of to practice making meaning differently. One person practiced experiencing awe every day, starting with a billboard of a baby orangutan, then moving to waves, flowers, and crickets. Eventually, a weed or dandelion poking through a cracked sidewalk could inspire awe by representing nature's power to be unconstrained by human control.
After months of daily practice, experiencing awe becomes possible when most needed, such as before important meetings, when stuck in traffic, when a computer fails, when needing to stay calm, or when sleep-deprived with a child acting out. The practice is not about picturing the dandelion but finding a moment in the present to experience awe about, such as the miracle of talking across distances when the internet fails.
Control and agency do not look the way expected. The ability to act in the moment requires practice. Standing up to give a lecture appears spontaneous but results from practicing 30 times beforehand. Control is harder to get than expected, and not everyone has the same opportunities. Even a monk being tortured can exercise some control, meaning everyone has a little more control over how they experience themselves in the world than they think, and sometimes that little control is enough.
Every morning, walking past an office creates the temptation to check email for just a minute, knowing it will become an hour. Practicing not doing the thing one feels compelled to do in order to do the thing one will later be glad about builds the skill of making meaning differently.
The approach is not about willpower. Willpower is putting a bowl of potato chips in front of yourself and trying not to eat them. The alternative is constructing the niche so the bowl of potato chips is not present. To change how one feels in the moment, one can change the remembered past, change the sensory present, or move in some way, which usually involves changing the sensory present.
Over 10,000 emails from people who read the author's books report finding ways to use the scientific findings in their own lives, gaining a sense of agency and hope. People have written from battlefields and all walks of life. The work on metabolism and meaning-making came after initial work on emotion and the brain.
Grief involves losing a source of body budgeting impact. When losing someone loved, one loses a person who was regulating one's nervous system. A daughter as a baby was regulating her mother's nervous system as much as the mother was regulating hers.
When you lose someone you love, it can feel like you've lost a part of yourself because you've lost body budgeting support and someone you were physically connected to. Love and safety involve being physically linked to another person in a way that impacts metabolism mutually. Texting three words to a friend in Belgium can affect her heart rate, breathing rate, and mood, demonstrating that words and actions toward others are powerful. Losing someone means losing this connection, and you feel it physically.
This doesn't explain why someone died or make it seem fair, but it helps understand what the sensations mean. Evidence suggests that approximately 70% of people who don't experience complicated grief will have their symptoms go away within a couple of months, while complicated grief like the loss of a child can last longer. Eventually, within a year or two, people may still feel sadness but won't suffer - it becomes discomfort rather than suffering.
Research indicates that people who recover faster from loss of a loved one are those who remember the positive meaningful times together. Cultural traditions like sitting shiva in Jewish mourning practices or wakes help by remembering things with others, bringing the person closer and more alive.
An example involves memories of a daughter at age two that parents share but the child herself doesn't remember. The parents had a relationship with someone who's not here anymore, and they sometimes miss that exuberant two-year-old. This extends to parents missing their children's younger selves - the child doesn't remember those early experiences, but the parent does, creating a shared memory that can be revisited through photos, videos, journals, and text messages.
Memories are predictions rooted in the brain reinstating patterns from the past for the purposes of the present. When reading a memoir, you're reading how someone experienced what happened to them, not literally what happened. When you have a memory, you're not literally remembering what happened - you're constructing an experience in the moment informed by signals from the past.
Memories are not like files in a file drawer and are not veridical. You are always reconstructing them when you remember something. Sometimes when an intrusive thought comes to mind, you can literally say out loud "Nope, not that past" to redirect the reconstruction.
The work done through experimental research unable to replicate certain findings led to a new career direction that ultimately has potential for greater impact in clinical psychology than would have been possible as a practicing clinical psychologist. The conversation covered complex neuroscience concepts that may require multiple listens to fully understand.
Keep Peter Attia MD in your library
Save the videos and channels worth coming back to, and find them again in one place.
