Vision: How the Brain Turns Light into a Meaningful World

A white mug still looks white when you carry it from sunlight into shade. You recognize it when its circular rim projects an ellipse onto your retina, and you adjust your hand to grasp its handle. Vision solves all three problems: interpreting light, recognizing objects, and guiding action.

Learning goals and Feynman practice

Explain how light becomes a neural signal, compare rods and cones, trace the main visual pathway, and distinguish early processing channels from higher cortical streams. Explain why seeing is an active, evidence-constrained interpretation rather than a camera recording.

After each section, close the notes and teach the idea to a beginner. Give a fresh example, explain the mechanism, and state where the analogy stops working. When your explanation stalls, reread that part and simplify it. Attempt the self-checks before opening the answers.

1. Why vision is harder than taking a picture

Two major purposes are object recognition, identifying what something is, and spatial guidance, locating things so we can move and act. These are useful organizing principles, not the only functions of vision.

Light reaching the eye depends jointly on surfaces, illumination, viewpoint, and obstructions. Different combinations can produce similar retinal inputs. The visual system must estimate likely properties of the world from incomplete evidence.

Perceptual constancies help stabilize experience:

Constancy	Example	What the system must take into account
Lightness	A white page still appears relatively white in shade.	Surface reflectance versus illumination.
Colour	A familiar object looks broadly similar under different lights.	The illumination’s spectral composition and scene context.
Shape	A tilted door remains recognizable as rectangular.	Perspective and orientation.
Size	A friend walking away does not seem to physically shrink.	Retinal image size together with estimated distance.

Brightness concerns perceived light intensity; lightness concerns perceived surface reflectance. They are related but not interchangeable. Constancies are also imperfect: unusual lighting or misleading context can change appearance.

Think of vision as inference from clues. This does not mean a tiny observer lives inside your head, or that the brain invents anything it likes. Interpretation emerges from interacting neural processes constrained by incoming signals.

Teach it back: Why does a darker retinal image fail to prove that a surface itself has become darker?

2. The eye: focusing light before interpreting it

Light passes through the cornea, the clear front surface providing much of the eye’s focusing power. It enters through the pupil, an opening whose diameter is regulated by the iris. The lens adjusts its shape through accommodation to help focus objects at different distances onto the retina. The retina contains light-sensitive cells and neural circuits; it is not a passive projection screen. [1]

The pupil is an opening, not a focusing lens. Accommodation changes lens shape, not the location of the retina. These distinctions help explain the mechanism rather than merely recite anatomical names.

Transduction converts light into changes in cellular activity. In rods and cones, absorbed photons initiate a biochemical cascade that changes membrane potential and neurotransmitter release. Light generally hyperpolarizes these photoreceptors; they do not simply fire an action potential for each photon. Retinal circuitry transforms their signals before ganglion-cell axons carry action potentials toward the brain.

A useful first circuit is photoreceptor, bipolar cell, then ganglion cell. Horizontal and amacrine cells add interactions that shape contrast, timing, and other properties. Processing therefore begins before signals leave the eye.

3. Rods and cones: sensitivity and detail require different designs

Feature	Rod system	Cone system
Typical operating advantage	Very dim illumination.	Daylight and other sufficiently bright conditions.
Distribution	Numerous outside the central fovea; absent from its centre.	Present across the retina, densely packed in the fovea.
Detail	Lower spatial acuity, partly due to pooling.	High acuity especially in central vision.
Colour	Rod signals alone cannot discriminate wavelengths as colours.	Comparing cone classes supports colour vision.
Convergence	Many rods can contribute to shared output pathways.	Central midget pathways can preserve signals from individual cones; other cone circuits pool inputs.

Textbooks often give approximately 100–120 million rods and 5–6 million cones per eye; estimates vary across individuals and counting methods. The functional contrast matters more than memorizing a single count.

Pooling resembles several microphones feeding one alarm: a weak signal becomes easier to detect, but identifying its precise origin becomes harder. This captures part of the sensitivity–acuity trade-off. Retinal circuits are more complex than a single wiring diagram.

Cones have overlapping short-, medium-, and long-wavelength sensitivities, often labelled S, M, and L. “Blue, green, and red cones” is shorthand. No cone individually recognizes a colour; the system compares signals across receptor classes and later opponent channels. An L cone is not activated only by red light.

The claim that cones connect one-to-one with ganglion cells applies approximately to particular central pathways, not every cone everywhere. There are many kinds of ganglion cells carrying different combinations of information.

Teach-back check: why can a dim object be detectable without revealing fine detail or colour?

Rod-based circuits are highly sensitive and pool signals across receptors. Pooling helps detect weak light while reducing spatial precision. A single receptor class cannot independently distinguish a wavelength change from an intensity change, so rod-only vision does not provide ordinary colour discrimination.

4. The blind spot and light detection beyond images

At the optic disc, ganglion-cell axons leave the eye to form the optic nerve. This region lacks rods and cones, producing a blind spot in each eye’s visual field. Normally the other eye and contextual processing help prevent an obvious gap in experience. Contextual filling-in does not mean the missing receptors secretly detect the scene.

Some retinal ganglion cells contain melanopsin and are intrinsically photosensitive. These ipRGCs contribute to responses such as circadian synchronization and pupil regulation. They can respond directly to light, but also receive rod and cone signals. “Independent of rods and cones” describes their intrinsic light sensitivity, not total separation from retinal circuits. [2]

Teach it back: Explain how light can affect the body’s daily timing without requiring detailed recognition of an object.

5. From eyes to brain: visual fields, not one eye per hemisphere

In the main pathway, ganglion-cell axons form the optic nerves. At the optic chiasm, fibres from the nasal half of each retina cross, while temporal retinal fibres remain on the same side. Consequently, the left visual field is represented mainly in the right hemisphere, and the right visual field mainly in the left. Each eye supplies both hemispheres.

Signals continue through the optic tracts to the lateral geniculate nucleus, or LGN, in the thalamus, then through optic radiations to primary visual cortex, V1, in the occipital lobe. Other retinal projections contribute to eye movements, orienting, pupil responses, and circadian regulation.

The LGN does more than relay messages unchanged. It participates in selective transmission and receives extensive feedback. Separate channels already begin in the retina; the LGN does not create them from an undifferentiated picture.

Magnocellular and parvocellular channels

Property	Magnocellular, or M	Parvocellular, or P
Relative cell and receptive-field size	Larger.	Smaller, at comparable retinal locations.
Typical temporal response	Relatively fast and transient.	Relatively sustained.
Major strengths	Luminance contrast and rapid changes.	Fine spatial detail and aspects of colour, especially red–green comparisons.
Receptor input	Cone input as well as rod contributions through retinal circuits.	Prominent cone input; not an exclusively isolated cone wire.

M is not simply the rod pathway, and P is not the entire colour pathway. Koniocellular channels also matter, including for some S-cone signals. Neither a fixed M:P neuron ratio nor a claim about one pathway being uniquely present in all primates is needed to understand the functional comparison.

Teach-back check: why is closing the left eye not equivalent to removing the left visual field?

Each eye normally samples both sides of fixation. Routing depends on which retinal half receives the image; partial crossing sorts the two visual fields between hemispheres.

6. V1: maps and local features

Retinotopy means that neighbouring visual locations tend to map to neighbouring neural locations. Cortical magnification means the map devotes disproportionate tissue to central vision. These principles coexist: a map can preserve neighbourhood relationships while greatly enlarging an important area, like a city-centre inset on a transport map.

This helps explain why you move your eyes to inspect a word: placing it near the fovea takes advantage of dense receptors and substantial neural representation. Retinotopy is a mapping of visual space, not a full-colour miniature image watched by another observer.

Hubel and Wiesel’s recordings helped establish that many V1 neurons respond selectively to oriented edges or bars within particular receptive fields. A receptive field is a region of visual space where a stimulus can influence a neuron’s response. [3]

A neuron preferring a vertical edge is not a “door detector.” Its response depends on location and other stimulus properties. Object recognition requires patterns of activity across many neurons, along with recurrent interactions and context.

Visual information is transformed across many populations and areas. Exact neuron counts depend on species, boundaries, and measurement methods. The important idea is distributed processing, not an unverified claim that neuron numbers expand exponentially at every stage.

Teach it back: How can a small patch of central vision receive more processing resources than a much larger peripheral region?

7. Higher visual streams: recognizing and acting

Stream	Broad cortical direction	Useful question	Example
Ventral	Toward temporal regions.	What is this?	Recognizing your mug despite a changed viewpoint.
Dorsal	Toward parietal regions.	Where is it, and how should I act?	Adjusting the hand’s orientation to grasp the handle.

The dorsal stream is often called where/how, because visually guided action involves more than locating an object. Ventral processing supports increasingly complex representations, with some populations responding preferentially to categories such as faces or bodies.

Higher representations can tolerate changes in size, position, or viewpoint, but this tolerance is neither unlimited nor identical across neurons. Recognition generally depends on population patterns, rather than one neuron for every complete object.

Ventral processing often emphasizes central vision, and dorsal processing often emphasizes spatial and peripheral information. Claims that all ventral neurons have foveal receptive fields, or a fixed percentage of dorsal neurons are peripheral, are too broad without specifying the sampled area, species, and task.

The cortical streams interact. They also do not map one-to-one onto the earlier P and M channels: multiple inputs contribute, and information mixes. Use the two-stream model as an organizing guide rather than two sealed pipes.

8. Worked example: finding your mug in a dim kitchen

At first, rod-sensitive circuits may help detect a vague shape, without enough information for confident colour or detail judgments. As lighting improves, cone signals contribute more useful detail. Moving your eyes places the mug near the fovea. Retinal circuits and the LGN transmit selected information; V1 and other areas analyse spatial patterns. Ventral networks contribute to identifying the mug, while dorsal networks help guide your reach. Context helps estimate its surface properties despite shadows.

This is a teaching sketch, not a rigid queue. Processing occurs in parallel, with feedback between levels, and both streams help solve an everyday task.

Teach it back: Tell the story again using only these words as prompts: light, pooling, map, identity, action, context.

9. Feynman workshop and spaced review

Draw the main pathway from memory: label the eye’s optical parts, retinal circuit, optic nerve, chiasm, LGN, and V1. Add the two cortical streams. Beside each label, write the problem it helps solve rather than only its name.

Repair: “The eye sends a photograph to the brain”

The eye focuses an optical image onto the retina, but retinal circuits transform light-dependent signals into multiple patterns of neural output. Later processing estimates properties of objects and scenes. No intact photograph travels along the optic nerve.

Repair: “An M neuron must receive only rods, and a P neuron identifies an object”

M channels also receive cone signals, and P channels carry selected visual information rather than complete object identities. Higher object recognition combines activity across multiple stages and networks.

Apply: a tilted plate projects an ellipse. Why can it still look circular?

The visual system uses perspective, context, and other cues to estimate the underlying shape. Shape constancy allows a reasonably stable object interpretation despite a changing retinal projection, though misleading cues can defeat it.

Tomorrow, explain rods versus cones without looking. In a week, trace the pathway again and give a new example of constancy. Score each explanation: 0 for a remembered label, 1 for a definition, 2 for a mechanism, example, and limitation. Repair the lowest score first.

One-minute explanation: The eye focuses light onto a retina that already processes information. Rod and cone circuits trade sensitivity against detail and support different aspects of seeing. Partially crossed pathways route visual fields through the LGN to mapped cortical representations. Networks analyse features, recognize objects, and guide action. Constancy helps perception remain useful as lighting and viewpoint change.

Sources and further reading

The supplied outline’s numbered citations did not include a bibliography. These sources support eye anatomy, intrinsic retinal photosensitivity, and the classic cortical recordings. The everyday scenarios and self-checks above are teaching examples.

1. National Eye Institute. How the Eyes Work.
2. Berson, D. M., Dunn, F. A., and Takao, M. (2002). Phototransduction by retinal ganglion cells that set the circadian clock.
3. Hubel, D. H. (1981). Nobel lecture: Exploration of the primary visual cortex.
4. Lee, B. B., Smith, V. C., Pokorny, J., and Kremers, J. (1997). Rod inputs to macaque ganglion cells.