What this covers

Nicole Rust discusses why modern neuroscience, despite decades of progress in understanding brain function, has delivered surprisingly few cures for brain disorders. She and Sean Carroll examine the mismatch between our expanding knowledge of genes, brain imaging, and neural mechanisms on one hand and the modest clinical wins in treatment on the other. The core argument traces this gap to a flawed foundational model: the 1990s vision of the brain as a chain of dominoes, where finding and fixing one broken domino solves the disorder. Rust contends that the brain operates instead as a [complex adaptive system](concept:Complex Adaptive Systems) with interlocking feedback loops across multiple levels, making single-cause thinking obsolete. She grounds this in concrete failures—the amyloid hypothesis driving Alzheimer's treatments that add only months of benefit—and in structural challenges like the chicken and egg problem: conditions like depression or insomnia cluster diverse biological causes under one symptom-based name, making it impossible to group patients by etiology.

Rust proposes that progress requires two shifts. First, brain disorders should be reframed as control problems—moving a complex dynamical system from one state to another—which radically changes how we think about interventions and their limitations. Second, treating subjective phenomena like mood demands epistemic iteration: jumping in with imperfect measurements and letting theory and measurement refine each other over time, rather than waiting for perfect understanding before measuring. She illustrates her case with successes that worked differently than expected: the discovery-driven orexin blocker suvorexant, the neuroimmune model explaining NMDAR encephalitis, and a Parkinson's drug that succeeded by modeling the disease as a dynamical system rather than a linear chain. The conversation also touches on plasticity, allostasis, and the adaptability-fragility tradeoff that makes adaptive brains vulnerable to burnout.

Sharpest takeaway

Rust argues that neuroscience has failed to cure many brain disorders because it relied on an oversimplified 'find the broken domino and fix it' model; the brain is a complex adaptive system with feedback loops at and between levels, so treatments must be reconceived as a control problem and progress must proceed through epistemic iteration.

  • The amyloid hypothesis of Alzheimer's led to plaque-clearing drugs that barely slow decline, showing single-cause models fail
  • Feedback loops and emergent properties mean disorders cannot be reduced to one broken gene or brain area
  • Measuring subjective phenomena like mood requires committing to iterative refinement of measurement alongside theory

The argument · threads9 threads · 38 claims
0.81

Brain function operates at the population level across interdependent neural systems whose adaptability creates both learning and fragility.

6 pointscentrality 5/5
  • The 2012 deep neural network breakthrough in computer vision, developed to solve object recognition for engineering purposes, turned out to bear striking resemblance to how the brain performs object recognition, so these models became state-of-the-art models of brain visual processing and now extend to auditory and memory systems via the task-optimized framework.

    “in 2012, there was a breakthrough in computer vision... it turned out that the way that those models worked, bear is like a striking resemblance in how the brain does it too. So those actually became our state of the art models of how our brains perform the same thing.”

  • Treating the brain holistically means capturing the interactions between its parts by measuring the parts all at the same time rather than one at a time, and approaching problems from the top—starting with a phenomenon that needs explaining and asking how the brain gives rise to it—rather than only building up from how individual neurons work and hoping mood emerges.

    “measuring the parts all at the same time, as opposed to one at a time, and also approaching it from the top. So here's a phenomenon that needs to be explained, like some brain function. And so, how does the brain give rise to that, as opposed to if we just understand how the neurons work, then from the bottom, we can kind of build up and then out will come the mood.”

  • Neuroscience has shifted from focusing on what individual neurons do to appreciating that brain function resides in the pattern of activity across entire populations of neurons; models align with the brain at this population level, where similar stimuli (a horse and a cow) produce more similar population responses than dissimilar ones (a horse and a basketball).

    “We really used to focus on the brain and looking at what the individual neurons were doing, but now we've kind of taken a step back and really started to appreciate that it's really how an entire population of neurons, it's the pattern of activity across a big population.”

  • Adaptability is the brain's superpower as a complex adaptive system—we learn new languages, our bodies shift modes to flee a tiger—but this adaptability comes at the price of fragility, because the system's many interdependencies mean that adaptive mechanisms like chronic stress activating cortisol can burn out, interfere with sleep, and damage structures like the hippocampus where memory and cognition reside.

    “adaptability is really the brain's superpower... But this adaptability also comes at the price of fragility. And some of those fragilities can be linked to the consequences of the adaptability.”

  • Labeling something a complex system does not guarantee good modeling, because all models are wrong but some are useful, and threading the needle between oversimplification and useful abstraction is one of the big challenges of the era and of all science, not unique to brain research.

    “all models are wrong, but some are useful. The famous phrase... And I think getting that right, that threading that needle, that is one of the big challenges of our era.”

  • Allostasis differs from homeostasis: where homeostasis reacts to a change that has occurred (sweating once you produce heat), allostasis anticipates a change about to happen and adjusts physiology in advance—such as anticipatory insulin release before habitual meal times, or raising blood pressure before standing up so blood doesn't drain from the head—requiring prediction of the future.

    “Allostasis is anticipating a change that is about to happen and changing your physiology in order to anticipate that change... we can have anticipatory releases of insulin... when we are lying down and we get up, we have to increase our blood pressure... that actually happens in an anticipatory way”

0.76

Single-cause and domino-chain models fail to explain brain disorders, which arise from interactions between multiple levels and feedback loops.

5 pointscentrality 5/5
  • John Hardy, who discovered the gene and originated the amyloid hypothesis, now believes the single-toxic-protein story is a massive oversimplification and that Alzheimer's must be understood through how proteins, neurons, and organ systems interact and how toxicity travels through the brain—a multi-level complex dynamical systems approach.

    “the individual who discovered the gene, his team is John Hardy, and he came up with the amyloid hypothesis. And now he really thinks that this is a massive oversimplification. He thinks that we have to be thinking not just about proteins, but how proteins and neurons and organ systems and how toxicity travels through the brain. A really multi-level, complex, dynamical system type of approach is what he's proposing”

  • The 'grand plan of modern neuroscience' is the broad-strokes description of how researchers in the 1990s planned to get from where they were to treating disorders: by sequencing genes and imaging brains with new technology like fMRI to find either a mutated gene (molecular medicine) or a brain area with aberrant activity, identify the cause of disorders, and thereby arrive at treatments.

    “in the 1990s, there was tremendous enthusiasm that if we just sequenced genes and we imaged brains, so we had this new technology, FMRI so we could look there under the skull and see what's going on, that we would figure out the causes, of what causes brain disorders and then we would be able. That would lead us to treatments. And that just hasn't really panned out.”

  • Over 30 years ago researchers found the APP gene mutation that causes Alzheimer's with certainty and coding for amyloid protein, and inferring amyloid was the cause rather than consequence, they engineered drugs to clear amyloid plaques; but the latest approved plaque-clearing drugs only add about eight months to the roughly eight-year disease course and are clearly not the magic bullet, refuting the simple find-the-gene-fix-it expectation.

    “those drugs don't really seem to slow cognitive decline in the way that we hoped they would. So they do slow it down. I think in a disease that has a course of about eight years, they add about eight months to the progression of the disease... But they're clearly not the magic bullet.”

  • The 1990s model treated the brain as a domino chain of causes leading to effects, where you find the one broken domino and fix it, but the brain is actually a complex dynamical system full of feedback loops both within and between levels—where brain area A sends to B and B sends back to A—so its emergent, surprising malfunctions follow from interactions between parts, not from individual broken dominoes, just like weather, ecosystems, and nuclear reactors.

    “the brain is full of these big feedback loops at all levels... if you have brain area A that sends information to brain area B, brain area B sends information back to A again. So now you have a system where you don't just have causes leading to effects, but you have those effects feeding back again on the causes. So this is a whole different type of system. It's a complex dynamical system.”

  • The brain is held up as the most complex thing we know about in the universe, which is why a domino chain, even with many levels, is an inadequate model of it.

    “The brain is held up as the most complex thing we know about in the universe. And so a domino chain is not super complex, even if it has a whole lot of levels to it.”

0.58

Measuring mental states requires epistemic iteration, where theory and measurement interact over time rather than waiting for perfect instruments.

4 pointscentrality 4/5
  • Epistemic iteration resolves the bootstrapping problem that you need measurements to create understanding but understanding to create measurements: as Hasok Chang showed with temperature progressing over 250 years from crude thermometers to near-absolute-zero precision, science must jump in with a best-guess measurement, commit to refinement, and let measurement and theory interact to make progress—and the same applies to measuring mood.

    “the problem of epistemic iteration is, once we understand how something works, we'll know exactly how to measure it. But you need measurements to create understanding and you need understanding to create those measurements. So where do you even begin?”

  • Clinical trials for antidepressants still use a depression measurement developed in the 1960s, despite enormous technological progress since, and although 250 different depression scales exist, the next breakthrough will not be a 251st scale but a genuinely new way to measure depression.

    “in clinical trials for antidepressants, the measurement we're using for depression was developed in the 1960s... there are actually 250 different scales for depressing... I don't think the next big breakthrough in that research program is gonna be creating the 251st scale. I think we need a new way to measure depression.”

  • Mood is harder to study than memory because memory tasks have an objective ground truth—show pictures, ask if they were seen before—whereas mood is a purely subjective experience that can only be measured by asking how happy someone is, making rigorous scientific study difficult without becoming fuzzy or unscientific.

    “when we study memory, there's actually a task we can... I can give you a memory task and there's an objective ground truth to the questions I ask... But when I ask you about mood, that's a purely subjective experience. The only way I can measure that, is I have to ask you, how happy are you?”

  • Rather than locking down knowledge and moving on to new questions, neuroscience keeps asking the same questions about the same brain areas again and again; epistemic iteration reframes this hopefully, suggesting the questions stay the same while the answers progressively get better and better.

    “we just keep asking the same questions again and again, like some brain area. We're just stuck in this one brain area... epistemic iteration, it suggests like, yeah, we're gonna continue to ask the same questions, but we're gonna refine the answers. So the answers are gonna get better and better.”

0.58

Treating brain disorders as control problems in complex dynamical systems favors behavioral and talk therapy over single interventions or drugs.

4 pointscentrality 4/5
  • Treating a disorder like depression amounts to a control problem—shifting a complex dynamical system from one state to another—but controlling such systems is extremely difficult (we can't control the weather), and depression likely requires reprogramming the brain by adjusting hundreds of thousands of connection strengths between cells, which no drug or stimulation can do, so behavioral or talk therapy that reprograms the brain through learning is currently the best tool.

    “what you're trying to do is really influence the brain to shift it from one place to another... That's what engineers would call a control problem... I think there's a lot of evidence that what will be required is doing something like reprogramming the brain. It's the adjustment of thousands, to ten thousands, to hundreds of thousands of strengths between brain cells”

  • Although the standard view is that one side of the brain controls the opposite side of the body, there are also signals on the same side of the brain controlling the same side of the body; the IpsiHand stroke-rehabilitation therapy decodes these ipsilateral signals through a robotic-assisted device so patients learn over an hour a day to use residual brain capacity to move a paralyzed hand, working for some who haven't fully recovered—an example of harnessing the brain's plasticity.

    “there are also signals on the same side of your brain to control the same side of your body. So this therapy called IpsiHand, the way it works is it decodes those signals on the same side. And then you learn through a robotic assisted device that's guiding your hand... it actually works for some individuals that can learn how to use that stuff that's residual in your brain.”

  • Treating a system as complex—focusing on the fact that it is a complex system rather than its specific particulars—can be more useful for controlling or fixing it than focusing on its specificity, and a key open question is when general complex-systems principles transfer across domains (slime molds to economies, or vision to mood) versus when lumping is a stretch.

    “to what degree are there gonna be general principles or tools that are gonna help us... if we understand slime molds, is that gonna help us understand the economy?... in what situations should we be lumping versus splitting or be inspired by? And I think that's a really important question.”

  • For a subset of Parkinson's tied to a mutation in the GBA1 enzyme, researchers first developed a drug using simple domino-chain reasoning (A to B) which failed in clinical trials; a second group modeled the whole enzymatic process as a complex dynamical system with feedback loops across compartments, saw exactly why the original drug failed and what to do next, and that new drug is now in clinical trials expected to complete around 2026—exemplifying the complex-systems control approach.

    “That drug failed in clinical trials. And so then, another group of researchers said... it's actually this big feedback loop... So they modeled the whole thing as a complex dynamical system. And they could see exactly why that drug that was originally formulated wasn't working, and they could see exactly what to do next.”

0.54

Psychiatric and neurodegenerative disorders prove particularly resistant to the find-the-cause-fix-it approach, unlike neurological conditions.

3 pointscentrality 4/5
  • The field divides brain disorders into neurological and psychiatric conditions, and for the most part something gets classified as psychiatric specifically if we don't yet have any biological test for it, making the division partly a reflection of current ignorance rather than a fundamental distinction.

    “There's an interesting distinction there, where for the most part, something gets piled in the psychiatric condition pile, if we don't have any sort of biological test for it.”

  • NMDAR encephalitis manifests with psychosis resembling schizophrenia, but in 2008 researchers discovered it is a neuroimmune condition where the immune system creates antibodies attacking a brain receptor, causing brain swelling and psychosis; once the biological cause was identified, treatments could be developed, demonstrating how finding a biological marker can lead to a cure.

    “it was in 2008, researchers discovered it's actually a neuroimmune condition. So for some reason, our immune systems create these antibodies. We just have this big immune system attack on a certain type of receptor in our brains, that causes brain swelling and causes a psychosis. But once they realize that that was the problem, they could go in and come up with treatments for it.”

  • Certain disorder classes are particularly impenetrable to treatment: psychiatric conditions (whose drugs were mostly serendipitously discovered), neurodegenerative conditions (where we cannot slow degeneration in Parkinson's or treat Alzheimer's well), and neurodevelopmental disorders like intellectual disability and autism spectrum disorder; whereas conditions like migraine headaches are more amenable to the find-the-broken-domino approach.

    “there are classes of disorders that seem particularly impenetrable, and they are the psychiatric conditions for the most part... our neurodevelopmental disorders, so conditions like intellectual disability or forms of autism spectrum disorder. Those are the classes that seem to be really resistant to treatments, whereas other types of conditions seem to be more suitable for that, find the broken domino and fix it type of approach. One of those might be migraine headaches”

0.54

Most brain disorder drugs were serendipitously discovered rather than rationally designed, revealing the limitation of the bench-to-bedside model.

3 pointscentrality 4/5
  • Tracing the development of every existing brain disorder therapy reveals that few drugs were developed via the bench-to-bedside pipeline; many trace back to serendipitous discoveries in the 1940s and 1950s—such as the first antidepressant found when tuberculosis trial patients were dancing around happy, before the brain was understood—and were merely refined afterward, while genuine discovery-driven drugs are a rarer minority.

    “the first antidepressant that we ever discovered was a clinical trial for the lung infecting bacteria tuberculosis. And in that clinical trial, people were dancing around and happy. And so, they said, oh, well, maybe this would be great, as for mood and depression. And this is a time before we understood basically anything about the brain at all.”

  • The insomnia drug suvorexant exemplifies genuine discovery-driven development: decades of academic research uncovered orexin, a brain chemical that keeps you awake (opposite of melatonin), prompting pharmaceutical companies to hypothesize some insomnia stems from excess orexin binding, screen two million compounds to find a receptor blocker, and run clinical trials costing around a billion dollars over another decade to produce the drug.

    “this discovery of this drug, like I said, it was decades to get up to that point of understanding what was going on there... To do that, we had to screen two million different compounds to figure out one that would block the orexin receptors, and then put that through clinical trials. I believe it cost something like a billion dollars, but then the drug suvorexant was born.”

  • A way to know whether accumulated knowledge about how the brain works actually matters is to attempt a therapeutic intervention—if you understand enough to intentionally change things and it works, that validates the knowledge; this 'don't speculate, just show' ethos mirrors the engineering culture where talks end with a working demo, contrasting with neuroscientists who debate inferences without a definitive test.

    “one way to know that, might be a therapeutic intervention. Where if you can go in, and you know enough to very intentionally change things, then don't speculate, just show. Just get in there.”

0.44

Understanding feelings lags far behind understanding perception and motor control, creating critical gaps in psychiatry; neurodiversity frames some conditions as differences rather than disorders needing cure.

7 pointscentrality 2/5
  • Neuroscientific progress in understanding functions like seeing, remembering, hearing, and motor control is advancing rapidly, but understanding the neuroscience of feelings such as happiness is not progressing as quickly, which matters critically for psychiatric conditions like mood disorders and depression.

    “our progress and understanding how the brain does many things, like see and remember and hear and even motor control move, that seems to be progressing at a pretty rapid clip. But then when it comes to the neuroscience of feelings, like how happy are you, that's not progressing as quickly.”

  • The hope behind psychedelic drugs is that they enhance the brain's plasticity so that talk therapy works better or faster, or helps treatment-resistant patients, but this is risky because enhancing plasticity is just as likely to wiggle the brain out of a depressive episode as to make it worse, so people should not be self-experimenting.

    “that's really what the psychedelic drugs are all about. So that's the hope behind those is, they're gonna enhance the plasticity so talk therapy will work... you're just as likely to wiggle the brain out of a depressive episode as you are to make it worse. So you have to be really, really careful. People shouldn't be messing around with this stuff.”

  • One function of sleep appears to be reestablishing the balance of the brain via homeostatic plasticity, and many disorders seem to involve a brain that has become unbalanced; being able to measure whether a brain is in balance—imagined as a morning cap that reads how sleep-restored and balanced your brain is—could let us catch problems before they go too far.

    “one of the functions of it looks like it probably is reestablishing the balance of the brain. And so, being able to measure that, is a brain in balance?... how sleep restored is your brain this morning? So if that's going awry, we can actually catch it before it gets too far.”

  • Conditions labeled brain disorders, like ADHD, may partly be forms of neurodiversity that society has wrongly become intolerant of, and not everyone with ADHD wants or needs treatment; nonetheless there are subsets of individuals for whom it causes severe problems like insomnia or inability to hold a conversation who genuinely want treatment, so 'disorder' should be used empathetically to mean simply that better solutions are needed.

    “not everyone with ADHD will want or need treatment, but there are some individuals who really do. It can cause severe insomnia. It can cause the inability to hold a conversation... that's the way I use that term disorder, is just acknowledging we need better solutions. So it's very empathetic on my part.”

  • Measuring happiness raises the normative question of what we are actually trying to measure or optimize: rather than maximizing the mean happiness of a population, we may really care about reducing the variance or raising the lowest quartile, so the measurement target itself involves moral-philosophical choices.

    “maybe we're not trying to maximize the mean happiness of a population. Maybe we're really worried about the variance or the lowest quartile or something like that. So it's a discussion about what is it exactly we're trying to measure.”

  • While curiosity-driven foundational research remains essential because the manual of how the brain works is still full of empty pages, it need not happen in total isolation from end goals; researchers should also work backward from the nature of the problem to be solved—such as how to control a complex dynamical system—which requires importing physicists, mathematicians, and other disciplines.

    “I don't know that curiosity-driven research has to happen so much in the absence of consideration of what the end goals are... here's the nature of the thing we have to solve. Let me work backward and figure it out... We need to figure out how to control a complex dynamical system. That's really hard. We need physicists and mathematicians”

  • Writing a popular book for a general audience gives back to a scientist's own research program by reigniting excitement, providing new perspectives from pulling pieces together, and making apparent where the holes and poorly understood bits are.

    “it just gives back to the research program in so many ways. It can get you excited about things you haven't been excited about in a long time. It gives you new perspectives 'cause you have to just pull the pieces together... I don't think of writing as a service. I think of writing as kind of selfish on my part.”

0.36

Heterogeneous symptom-based diagnostic categories hide multiple underlying biological causes, making causal grouping and targeted treatment impossible.

2 pointscentrality 3/5
  • If one identical twin has schizophrenia, the genetically identical co-twin has only a 50% chance of having it, proving the condition isn't determined by genes alone even with the same womb; its incidence is also linked to developmental trauma such as famine during pregnancy increasing schizophrenia risk in offspring, though we cannot yet predict or explain the mechanism.

    “if one twin has schizophrenia, the chances of the other genetically identical twin having schizophrenia are only 50%. So right there, we know that this isn't just about genes... when a population goes through a famine on the other side of that, if women were pregnant, there is an increased risk in schizophrenia and their offspring.”

  • Conditions given a single name like depression or insomnia actually have a multitude of underlying biological causes, analogous to cystic fibrosis where every different mutation in a single gene requires a different drug; this creates a chicken-and-egg problem because you cannot group individuals by cause when the only way to define them is by symptoms, leaving one large heterogeneous group.

    “all the different causes of depression, all the different biological causes of depression, will require different treatments. But that creates this really complicated chicken and egg problem. So how do you find the cause of something if you can't group together individuals with the same cause, because the only way you can define it, is by their symptoms”

0.36

Gene therapies for brain disorders face practical barriers, while neuroscience still lacks consensus on fundamental questions about efficacy.

2 pointscentrality 3/5
  • There is essentially no consensus in brain research beyond the fact that the brain has neurons in it; debates remain about the efficacy of antidepressants and talk therapy, partly because of the difficulty of defining the right control group given that some depression resolves on its own and patients are still having experiences.

    “there is no consensus in brain research other than the brain has neurons in it. So I think that there's a lot of debate about all of these things, including how effective are antidepressants? How effective is talk therapy?”

  • Because you are born with essentially all the neurons you will ever have, you cannot fix a causal gene mutation by editing it the way sickle cell is treated by removing and reinserting blood cells, and current gene-editing techniques are not good enough to edit genes across the roughly 85 billion neurons in an adult brain, so correcting such mutations is likely a long way off.

    “you're born with all of the neurons that you have, and you're ever gonna have for the most part... you can't do, for example, what they're doing with sickle cell where they're taking the blood cells out of the body or the bone marrow, and doing something and then putting them back in. That's not gonna work for a brain.”