How Science Works | Neuroscience — Neurons, Circuits, Brains, Behaviour and the Evidence Problem

1. The Scientific Job of Neuroscience

Neuroscience asks how nervous systems detect information, transform it, store it, coordinate action and change through development and experience. It spans molecules, cells, circuits, brain regions, whole organisms and behaviour.

No single scale owns the entire explanation. Ion channels explain membrane currents. Synapses explain communication between cells. Circuits explain coordinated patterns. Systems neuroscience links those patterns to perception and behaviour. Cognitive neuroscience asks how measured neural activity relates to mental functions.

2. A CivDJ Lens: Signal, State, Transformation and Output

Neural reasoning becomes clearer when we separate the signal, the state of the receiving system, the transformation performed by a circuit and the output that changes behaviour or another neural state.

The same sensory input can produce different responses depending on attention, expectation, fatigue, learning or internal state. That means input alone does not determine output. The receiver state is part of the mechanism.

3. Neurons Are Excitable Cells

Neurons maintain electrical potential differences across their membranes using ion gradients and selective channels. Changes in membrane conductance alter voltage. If conditions reach a threshold, many neurons generate action potentials that propagate along axons.

An action potential is not a miniature thought. It is a stereotyped electrical event that carries timing information. Meaning emerges from which neurons fire, when they fire, how strongly populations respond and how downstream circuits interpret that activity.

4. Synapses Are Conditional Handoffs

At chemical synapses, an arriving action potential can trigger neurotransmitter release. Transmitter molecules cross a tiny gap and bind receptors on the receiving cell, changing its conductance or biochemical state.

Synapses can excite, inhibit, modulate or change their own future effectiveness. Their effect depends on receptor type, timing, location and the receiving neuron’s current state. A synapse is therefore not a fixed wire; it is a dynamic interface.

5. Neural Codes Are Population Patterns

Neuroscience often asks how information is represented in neural activity. Some neurons respond strongly to particular features, but most behaviours involve distributed patterns across many cells.

Rate coding, timing, synchrony, population vectors and mixed selectivity are different candidate descriptions of how neural activity carries task-relevant information. The useful code depends on the system and question; there is no reason to assume one universal format.

6. Circuits Transform Rather Than Merely Relay

Neural circuits combine excitation, inhibition, recurrence and feedback. They can amplify differences, suppress noise, integrate evidence, generate rhythms and maintain activity after an input disappears.

This is why a brain cannot be understood as a cable map alone. Connectivity matters, but dynamic interaction decides what the circuit actually does.

7. Inhibition Is Computation, Not Merely Suppression

Inhibitory neurons regulate timing, gain, competition and stability. They can sharpen selectivity, prevent runaway excitation and coordinate oscillations.

Calling inhibition a simple “off switch” loses most of its function. Neural computation often depends on precisely balanced excitation and inhibition.

8. Sensation Begins With Transduction

Sensory receptors convert physical or chemical energy into neural signals. Light changes photoreceptor chemistry; sound moves mechanical structures in the ear; pressure deforms mechanoreceptors; chemicals bind taste and smell receptors.

Perception therefore begins before the brain “interprets” anything. The sensory apparatus already filters the world by what it can detect and how it encodes intensity, location and timing.

9. Perception Is Inference Under Constraints

Sensory signals are incomplete and noisy. The brain combines incoming evidence with prior structure, context and expectations to infer likely causes in the world.

Illusions reveal this process because the inference machinery works normally under unusual conditions. The error is not proof that perception is unreliable; it is evidence about assumptions the system usually uses successfully.

10. Attention Changes Processing Priority

Attention alters which inputs receive enhanced processing and which actions or memories gain access to limited resources. Neural responses can be amplified or suppressed depending on task relevance.

Attention is not one switch located in one place. It involves distributed control systems interacting with sensory and motor circuits. Different forms of attention can rely on partially different mechanisms.

11. Memory Is Not a Single Storage Box

Working memory, episodic memory, semantic knowledge, procedural learning and conditioned responses involve different processes and neural systems. Memory formation can alter synaptic strength, circuit organisation and systems-level interactions.

Retrieval is reconstructive. Remembering can depend on cues, context and current state. A memory is not necessarily replayed like a perfect recording.

12. Plasticity Lets Experience Change the System

Neural plasticity refers to lasting changes in neural function or structure associated with experience, development or injury. Synaptic strengthening and weakening are important mechanisms, but plasticity also includes changes in excitability, connectivity and network organisation.

Plasticity does not mean unlimited rewiring. Change is constrained by developmental state, genetics, prior structure and biological cost.

13. Motor Control Is Feedback Plus Prediction

Movement requires selecting goals, transforming them into motor commands, coordinating muscles and correcting error. Sensory feedback reports actual consequences, while predictive models can estimate expected consequences before slow feedback arrives.

This combination explains why skilled movement can be fast yet adaptable. Pure feedback would often arrive too late; pure prediction would drift without correction.

14. The Brain Is Not a Set of Independent Modules

Specialisation exists, but most complex functions arise from interacting networks. Language, memory, attention and decision involve distributed systems with dynamic coupling.

Brain maps are therefore useful when treated as network maps, not as labels claiming “this spot is where X lives.”

15. Lesions Reveal Necessity—With Caveats

Damage to a brain region can reveal whether that region is necessary for a task under particular conditions. But lesions often affect fibres, neighbouring tissue and network dynamics beyond the visible damaged area.

Recovery and compensation also complicate interpretation. A deficit can reveal causal involvement without providing a complete localisation of function.

16. Electrophysiology Measures Fast Neural Events

Electrodes can record membrane potentials, action potentials or population-level electrical activity. These methods provide excellent temporal resolution and, in some settings, highly local measurements.

The recorded signal still depends on electrode placement, reference choice, filtering and source geometry. “Electrical activity” is not a direct readout of a thought.

17. EEG Trades Spatial Precision for Timing

Electroencephalography records voltage differences at the scalp produced largely by coordinated postsynaptic activity in cortical populations. It can track millisecond-scale dynamics but has limited ability to localise deep or closely spaced sources uniquely.

This is an inverse problem: many different source configurations can produce similar scalp patterns. Source localisation therefore depends on models and assumptions.

18. fMRI Measures Blood-Oxygen Consequences

Functional MRI commonly uses blood-oxygen-level-dependent signals related to local changes in blood flow and oxygenation associated with neural activity. It provides useful spatial coverage but is indirect and slower than neural electrical events.

A coloured brain image is therefore not a photograph of thought. It is a statistical map of measured signal differences interpreted through neurovascular coupling and analysis choices.

19. Imaging Statistics Can Manufacture Confidence

Brain imaging analyses may test thousands of locations. Without proper correction, chance fluctuations can appear significant. Preprocessing choices, movement, smoothing and region selection can alter results.

Replication, preregistration, transparent pipelines and adequately powered samples are therefore especially important in neuroscience.

20. Correlation Is Not a Neural Mechanism

If activity in a region rises during a task, that activity may be necessary, supportive, compensatory or simply correlated with another process. Observational imaging alone often cannot distinguish these possibilities.

Causal strength improves when perturbation changes the predicted behaviour and when alternative explanations are controlled.

21. Perturbation Tests Causal Roles

Neuroscience uses lesions, electrical stimulation, pharmacology, transcranial magnetic stimulation and, in research animals, optogenetic or chemogenetic tools to alter neural activity. If changing a circuit predictably changes behaviour, causal interpretation becomes stronger.

Perturbation is still not perfect. Interventions can spread beyond the intended target or trigger compensatory responses. The strength comes from converging methods.

22. Animal Models Create Transfer Questions

Animal research permits invasive measurements and causal interventions that are not possible in humans. Shared neural mechanisms make many findings informative across species.

But translation is not automatic. Species differ in behaviour, development and brain organisation. A result must be tied to the conserved mechanism, not merely the label of the task.

23. Behaviour Is a Scientific Measurement

Reaction time, accuracy, eye movement, movement trajectory, choice and verbal report are observable outputs. Neuroscience becomes stronger when neural data are connected to carefully designed behavioural measurements rather than interpreted in isolation.

A neural difference without a behavioural or computational interpretation may be real but scientifically incomplete.

24. Computational Models Make Hypotheses Explicit

Models can describe evidence accumulation, reinforcement learning, sensory coding, network dynamics and control. A computational model specifies variables and rules precisely enough to generate predictions.

Several models can fit the same behaviour. Model comparison and out-of-sample prediction are therefore essential. A good fit is not proof of psychological truth.

25. Development Changes the Operating System

Brains change during development through cell growth, synaptic change, myelination, pruning and experience-dependent reorganisation. The same task can be solved by different neural strategies at different ages.

This makes developmental neuroscience state-dependent. Adult mechanisms cannot always be projected backward unchanged onto children.

26. Sleep, Arousal and State Alter Neural Computation

Neural processing depends on global state. Sleep, fatigue, stress, arousal and neuromodulatory systems change responsiveness, attention and memory.

Experiments that ignore state can mistake transient performance conditions for stable neural traits.

27. Worked Example: Seeing a Moving Object

Light from the object reaches the retina, where photoreceptors transduce it into neural signals. Retinal circuits transform contrast and temporal changes before information reaches the brain. Visual pathways then extract and integrate features related to position, motion and context.

The perception of movement is therefore not one neuron detecting “motion” in isolation. It emerges from layered transformations and population activity constrained by sensory input.

28. Worked Example: Learning a New Motor Skill

Early practice requires attention and produces variable performance. Feedback reveals error. Repeated practice changes motor commands, sensory prediction and neural plasticity. As performance becomes more automatic, less conscious control may be needed for routine components.

Improvement is therefore not simply “the brain forms a pathway.” It is a changing control system involving representation, prediction, error correction and plasticity.

29. Common Neuroscience Failure Modes

• Reverse inference: seeing activity in a region and naming a mental state from it without enough specificity.
• Blob psychology: treating colourful brain maps as direct locations of thoughts.
• Correlation inflation: turning co-activation into causation.
• Scale collapse: jumping from ion channels directly to complex behaviour without intermediate mechanisms.
• Animal-to-human overreach: assuming task labels guarantee identical mechanisms across species.
• State blindness: ignoring sleep, attention, arousal or development.
• Model fit worship: treating one well-fitting computational model as uniquely true.
• Neuroessentialism: assuming a neural description automatically explains more than behavioural or psychological evidence.

30. How to Think Like a Neuroscientist

Always ask what was measured directly. Separate electrical, metabolic and behavioural signals. Identify the scale. Use causal perturbation when possible. Compare multiple methods. Treat localisation claims as network claims unless evidence supports stronger specificity. Look for replication across laboratories, tasks and analysis pipelines.

Most importantly, do not let the biological vocabulary hide the inference gap. “Brain-based” is not a synonym for “proved.”

31. Neuroscience Connects Outward

Biology supplies cellular and organismal organisation. Chemistry supplies neurotransmitters, receptors and metabolism. Physics supplies electricity, diffusion and imaging principles. Psychology and cognitive science supply behavioural constructs that neural measurements must explain rather than replace.

Neuroscience is powerful precisely because it is a bridge discipline. It becomes weak when the bridge is mistaken for one shore.

32. The Frontier Is Better Cross-Scale Explanation

Modern neuroscience can record larger populations, manipulate circuits more precisely, map connectivity, model dynamics and combine imaging with behaviour at unprecedented scale.

The central challenge remains conceptual: how to connect molecules, cells, circuits and cognition without losing causality at the handoffs. Better measurement helps, but the frontier is ultimately an explanation problem.