Networks of Mind
A brain scan is an image full of colour and contrast, designed to look alive with signal. Bright reds and blues sit across a rendered brain, and it is easy to walk away thinking someone has looked inside the skull and seen thought itself. The image is a shadow: blood-flow changes, statistical residuals, group averages subtracted from group averages. We are not seeing the thinking. We are seeing what the thinking does to the machinery that carries it.
Shadows are not empty. Over the past two decades, functional MRI, resting-state connectivity studies, and large-scale meta-analyses have begun to map the architecture behind attention, mood, and creative work. The patterns are tentative, the effect sizes modest, and the causal claims sparse. They still point toward something usable. These capacities are networked processes, and some of those networks can be shifted by practice.
This chapter looks at what the science says, and at what it does not.
Overview
Neuroimaging has given everyday mental life a new vocabulary. Where older accounts spoke of focus, distraction, and mood, researchers now refer to the default mode network, the salience network, and the executive control network. These terms were not in textbooks a generation ago. They came out of data.
The interest for mental training is the possibility of intentional shifts among these networks. If different networks support different modes of thinking, training can be described more precisely than as the pursuit of better habits. It can be described as repeated practice that changes how these systems couple. Mindfulness, cognitive exercise, physical activity, and emotion-regulation strategies each have a research literature. Meta-analyses let us see which patterns recur and which do not.
What follows is an examination of three core networks, the emotion-regulation systems that cross them, and the creativity research that sits at their intersection. The claims are kept inside what the studies actually report.
The Default Mode Network
The default mode network was noticed almost by accident. In the late 1990s, researchers using PET and then fMRI found a set of regions that were more active when participants were told to rest than when they performed external tasks. The set was consistent across studies, internally connected, and reliably less active during focused external work.
The core regions include the medial prefrontal cortex, the posterior cingulate cortex and adjacent precuneus, and the angular gyrus. The network is engaged during self-referential thought, autobiographical memory, imagining the future, and thinking about other minds. It is not an idle circuit in the sense of doing nothing. It is the circuit the brain falls into when it is not occupied by an external task.
Killingsworth and Gilbert (2010) made the behavioural side of this concrete. Participants received prompts from an iPhone web app at random points in the day and reported what they were doing and where their attention was. Minds were wandering in 46.9 percent of samples. People were less happy when their minds had wandered, including during activities they otherwise enjoyed. The study did not scan anyone. It measured thought content and mood. The link to the default mode network is an inference from separate imaging work, not a result of that paper. Later research has also complicated the idea that default-mode activity is inherently an unhappy state. The network supports planning and social thought as well as rumination.
For training, the useful point is narrower. The network is not a defect. It supports memory, prospection, and a sense of self. Its bias toward past and future thought is what needs balancing by systems that hold attention on a present task.
Focused-attention meditation, including concentrative practices such as samatha, is associated with reduced activity or altered coupling in default-mode regions in several studies. Fox and colleagues (2016), pooling functional imaging of meditation, found dissociable patterns for focused attention, open monitoring, and compassion practices, rather than a single meditation effect. The mechanism is not a switch that stays off. Returning attention to one object, the breath or a sensation, is a repeated interruption of self-referential loops. The effect of a single session is small. What the training literature suggests is an accumulated change in how readily attention returns.
The Salience Network
If the default mode network carries internal narrative, the salience network is one of the systems that decides what gets priority.
Its cortical hubs are the anterior insula and the dorsal anterior cingulate cortex. The working account is that this network detects what is relevant, a sudden sound, a change in bodily state, a thought that stands out, and helps route processing toward internal reflection or toward goal-directed control. That switchboard description is a model. The regions are real and the connectivity is real. The clean handoff between networks is a simplification.
The anterior cingulate and the frontoinsular cortex also contain von Economo neurons, large bipolar cells with long axons, found in humans, great apes, some whales and dolphins, and elephants. They are not restricted to a few primate species, and they are not housed only in the dorsal anterior cingulate. Their long-range projections make them plausible contributors to rapid communication between these hubs. Claims that they explain meta-awareness, the noticing of one's own mental state, remain a hypothesis. The imaging case for meta-awareness does not depend on them.
Rahrig and colleagues (2022) meta-analysed resting-state studies of mindfulness training. Across 12 studies, training was associated with greater connectivity between the posterior cingulate, a default-mode hub, and a cluster in the left middle cingulate and paracingulate cortex, which the authors place in the salience network. The effect was small (Hedge's g = 0.234, uncorrected p = 0.0288). A reasonable reading is that the internal-narrative system and the relevance system become slightly better coupled. The further claim, that this makes it easier to notice a thought loop and step out of it, is an interpretation of that coupling, not a behavioural result of the meta-analysis.
Open-monitoring practices such as vipassana are often described as training this system, because attention is kept open to whatever arises without pursuit or rejection. Fox and colleagues (2016) found activation patterns for open monitoring that differed from focused attention, including regions involved in interoception and monitoring. That is not the same as a demonstrated increase in salience-network activity after a course of practice. The distinction between narrowing attention and monitoring whatever arises is still a useful training distinction. The network labels should stay looser than the instructions.
The Executive Control Network
The executive control network, often discussed alongside the frontoparietal control network, includes the dorsolateral prefrontal cortex and posterior parietal cortex. It is engaged during goal-directed effort: holding information in mind, planning, inhibiting a response, switching rules.
Cognitive training reliably improves performance on the executive tasks that were trained, especially working memory, flexibility, and inhibition. Transfer to unlike tasks is a different question, taken up below. Aerobic exercise also improves executive measures in many trials, with proposed contributors that include blood flow, neurotrophic factors, and, in animals, hippocampal neurogenesis. As an earlier chapter noted, adult human neurogenesis is contested, so it should not be treated as the mechanism in people.
Combined exercise and cognitive training is often reported to outperform either alone, but that claim depends on the outcome and the control condition. It is a reason to treat training as a set of practices rather than a single exercise. It is not a licence to say that any combination is superior.
Focused-attention meditation also trains the capacity this network is thought to support. Bringing attention back to a chosen object overlaps with the dorsal attention system as well as with executive control. People who practise this kind of meditation often perform better on attention tasks outside the session. The transfer is to tasks that resemble the skill, sustained and selective attention, more clearly than to intelligence or creativity in general.
Emotion Regulation
Emotion regulation, in this literature, is the interaction between regions that generate or detect affective signals, above all the amygdala, and prefrontal regions that alter how those signals are interpreted or held.
Berboth and Morawetz (2021) meta-analysed psychophysiological interaction studies of emotion-regulation connectivity. Across studies, regulation was associated with coupling between the amygdala and the left ventrolateral prefrontal cortex, largely driven by reappraisal tasks. When the analysis was limited to down-regulation by reappraisal, coupling also converged on the right dorsolateral prefrontal cortex and the dorsomedial prefrontal cortex. This is evidence of co-activation and task-modulated connectivity. It is not evidence that the amygdala seizes the prefrontal cortex during anger or fear. That remains a clinical description of a familiar experience, not a result of the meta-analysis. Kohn and colleagues (2014), in an activation-likelihood analysis of 23 cognitive-regulation studies, likewise mapped a prefrontal and parietal pattern for regulation and did not establish a hijack mechanism.
Two strategies have been compared directly. Cognitive reappraisal means changing the meaning of a situation. Acceptance means allowing the emotion without trying to alter it.
Monachesi, Grecucci, Ahmadi Ghomroudi, and Messina (2023) compared the two in an activation-likelihood meta-analysis. Both strategies increased activity in the left inferior frontal gyrus and the insula relative to conditions in which people simply experienced the emotion. Acceptance, at a more lenient threshold, additionally involved the claustrum and showed decreases in the posterior cingulate and in limbic regions. The authors argued that the strategies share an inhibitory circuit and differ in secondary pathways. That does not support a split in which reappraisal is effortful prefrontal override and acceptance is a passive process with no prefrontal role. Acceptance can feel less effortful. The imaging says the shared circuit is still an active one.
Both strategies are useful, for different situations. Reappraisal fits events whose meaning can be revised. Acceptance fits states that will not yield to a reframing, including some grief. Choosing between them depends on noticing what kind of state is present. That noticing is the meta-awareness the salience literature is trying to describe. It is not something a connectivity effect of g = 0.234 has demonstrated on its own.
Creativity
Creativity is not a network. The better-supported account is an interaction between default-mode regions and control regions. Default-mode activity is associated with associative and self-generated thought. Control activity is associated with evaluation and selection. Beaty and colleagues have argued that creative cognition involves coupling between these systems rather than a handover from one to the other. The sequential picture, generate freely and then evaluate, is a useful way to organise practice. It is not established that the brain runs the two modes in series, or that skilled creative people toggle them at will.
Cognitive-flexibility training improves performance on tasks close to the ones practised. Far transfer, to unlike tasks or to creativity in a real domain, is inconsistent. Being better at switching abstract rules does not by itself make someone more creative at their work. Baseline ability, working memory, motivation, and age predict who gains from this kind of training. Combined programmes look more promising in some reviews. That is a reason for broad practice, not evidence for a creativity network.
Structural Change
Functional scans are not the only evidence offered for training effects. Structural MRI studies have reported differences in cortical thickness and regional volume after mindfulness courses, often in prefrontal cortex, insula, hippocampus, and cingulate, and sometimes in the amygdala. The individual studies are small, the replications are mixed, and thickness is not a readout of better function. Stress, expertise, and learning can all change structure.
A 2023 anatomical-likelihood meta-analysis by Siew and Yu, often cited for a broad structural effect of mindfulness, should not be used. It was retracted. The unretracted analysis had also not found the prefrontal, hippocampal, and amygdala pattern sometimes attributed to it. Its only significant cluster was a volume increase in the right insula and precentral gyrus.
Behavioural effects of structured mindfulness programmes are on firmer ground than the structural images. Mindfulness-based stress reduction has small-to-moderate effects on stress, anxiety, and depression in meta-analyses. Mindfulness-based cognitive therapy reduces depressive relapse in people with recurrent depression, and in several trials has performed similarly to maintenance antidepressant medication for that specific outcome. Those are clinical findings. They are not proof that the brain thickened in the regions a scan happened to measure.
Practical Engagements
These exercises make the distinctions in the chapter concrete. They do not show you a network. They give you two ways of paying attention, or two ways of meeting an emotion, so the difference can be noticed.
1. Two modes of attention (10 minutes) For a few minutes, hold attention on the breath and return to it when it leaves. Then move attention through the body, noting sensations without following them. Alternate once more. Afterwards, describe the difference in your own words. The first mode is a model of focused attention. The second is a model of open monitoring. Neither minute of practice is evidence of a network change.
2. Generate, then select (20 minutes) Take one problem. For 10 minutes, write ideas without editing. For the next 10, keep the few that still look useful and revise them. The point is the shift in stance, from loose production to selection. It is a practice structure, not a demonstration that the default mode network ran the first half and the control network ran the second.
3. Two ways through a strong feeling (10 minutes) When a strong emotion is already present, try a reappraisal: a different, plausible reading of the situation. Note the effort. Then stop revising it and let the feeling remain for a short period without an attempt to change it. Note that difference too. Reappraisal is the more verbal move. Acceptance is not the absence of regulation. Use the one that fits the situation, and do not treat the exercise as a map of the prefrontal cortex.
4. A short check on where attention is (a few times a day) At a prompt, ask whether attention is on the past, the future, or the present activity. The aim is recognition, not suppression of past and future thought. Most samples will not be on the present activity. That is what Killingsworth and Gilbert found, and it is the tendency the rest of the practice is meant to make visible.
Limitations
Imaging studies show associations. A decrease in default-mode activity during meditation does not, by itself, show that meditation caused the decrease, or that the decrease caused any benefit.
Effects in this literature are small. A Hedge's g of 0.234 means the average trained group and the average control group differ by less than a quarter of a standard deviation. That can be reliable and still be a subtle shift.
Samples in individual fMRI studies are often 20 to 40 people. Meta-analyses help. They cannot repair weak designs underneath them.
People differ widely in resting connectivity. Two healthy adults can look unlike each other on these maps, which limits any universal prescription.
The network names are teaching tools. The brain does not switch cleanly between them. Boundaries overlap, and a cognitive process rarely belongs to one network.
Commercial claims that an app session activates a named network are not supported. A short session is not a network-level intervention.
Structural change does not equal functional gain. The brain restructures under many conditions, not all of them useful.
The creativity account is a research programme, not a settled circuit diagram.
These limits frame the findings. They do not empty them. The patterns that recur are worth training toward, at the size the evidence allows.
Connections to Other Chapters
An earlier chapter on attention covered sustained focus, selection, and the cost of switching. The executive and dorsal attention systems are the imaging vocabulary for those capacities, not a deeper explanation of them.
A chapter on creativity covered divergent and convergent thinking and the role of incubation. Coupling between default-mode and control regions is one current neural account of why free association and evaluation both matter. It does not replace the behavioural evidence.
A chapter on emotion regulation covered reappraisal, acceptance, and labelling. The connectivity and activation meta-analyses here add a constraint: the two main strategies share more circuitry than a passive-versus-effortful story suggests.
The practical implication is the same across these chapters. Training is a change in the relation between capacities, practised at the scale of weeks, with effects that are real and modest.
