INSTRUCTOR USE ONLY — Two sample questions with Excellent / Adequate / Weak answer variants
Use this to calibrate TA grading and to show students what different quality levels look like
| Criterion | Max |
|---|---|
| States Hebb's rule and links it to the NMDA receptor as its molecular implementation | 0.5 |
| Explains the dual requirement: glutamate binding (presynaptic activity) + postsynaptic depolarization | 1.0 |
| Correctly describes the Mg²⁺ block at resting potential and its voltage-dependent removal | 1.0 |
| Identifies Ca²⁺ influx through the open NMDA channel as the trigger for downstream plasticity | 0.5 |
| Explains why this matters: coincidence detection ensures only correlated pre/post activity strengthens synapses, filtering noise and preserving meaningful patterns | 1.0 |
Hebb's rule says that when a presynaptic neuron repeatedly contributes to firing the postsynaptic neuron, the connection between them strengthens. The NMDA receptor is the molecular implementation of this rule because it functions as a coincidence detector — it will only open when two conditions are met simultaneously.
First, glutamate must be released by the presynaptic neuron and bind the receptor, confirming presynaptic activity. Second, the postsynaptic membrane must already be depolarized (typically by nearby AMPA receptors responding to strong or convergent input). This second requirement exists because at resting potential (~-70 mV) a magnesium ion physically blocks the NMDA channel pore, preventing ion flow even when glutamate is bound. Only when the postsynaptic cell is sufficiently depolarized is the Mg²⁺ expelled by electrostatic repulsion, unblocking the channel.
Once both conditions are met, the NMDA channel opens and allows calcium to flood into the postsynaptic cell. This calcium influx is the critical signal — it activates CaMKII and other kinases that phosphorylate AMPA receptors and trigger their insertion into the membrane, strengthening the synapse (early-phase LTP).
The beauty of this mechanism is that it ensures only synapses where presynaptic and postsynaptic activity are correlated get strengthened. A presynaptic neuron firing randomly won't depolarize the postsynaptic cell enough to remove the Mg²⁺ block, so random noise is filtered out. Only input patterns that actually matter — that reliably drive the postsynaptic cell — produce plasticity. This is what makes the NMDA receptor a true Hebbian coincidence detector.
| Hebb's rule linked to NMDA | 0.5 / 0.5 |
| Dual requirement explained | 1.0 / 1.0 |
| Mg²⁺ block mechanism | 1.0 / 1.0 |
| Ca²⁺ influx as plasticity trigger | 0.5 / 0.5 |
| Functional significance (noise filtering, meaningful patterns) | 1.0 / 1.0 |
| TOTAL | 4.0 / 4 |
Hebb said neurons that fire together wire together. The NMDA receptor does this because it needs two things to happen at once to open. Glutamate has to bind it from the presynaptic side, and the postsynaptic cell has to be depolarized.
Normally magnesium blocks the channel so even if glutamate binds, ions can't get through. When the postsynaptic neuron is depolarized enough, the magnesium comes out and calcium can enter. The calcium then causes the synapse to get stronger.
This means that only when both neurons are active at the same time does the synapse change, which prevents random activity from changing connections.
| Hebb's rule linked to NMDA | 0.5 / 0.5 |
| Dual requirement explained | 1.0 / 1.0 |
| Mg²⁺ block mechanism — correct but lacks detail (doesn't mention voltage-dependence or electrostatic basis) | 0.5 / 1.0 |
| Ca²⁺ influx as plasticity trigger — mentioned but no downstream detail | 0.5 / 0.5 |
| Functional significance — states coincidence but thin on why this matters for meaningful vs. noisy patterns | 0.5 / 1.0 |
| TOTAL | 3.0 / 4 |
Hebb's rule is about how neurons learn. NMDA receptors are involved in learning and memory. They need glutamate to work. There is also something with magnesium that blocks them. When they open, calcium comes in and that strengthens the synapse.
This is important because it's how the brain learns new things and forms memories.
| Hebb's rule linked to NMDA — mentions both but doesn't connect them as implementation | 0 / 0.5 |
| Dual requirement — mentions glutamate but omits the critical postsynaptic depolarization requirement entirely | 0.5 / 1.0 |
| Mg²⁺ block — mentions magnesium but doesn't explain the voltage-dependent mechanism or why it matters | 0.25 / 1.0 |
| Ca²⁺ influx — correctly stated | 0.5 / 0.5 |
| Functional significance — only a vague statement about "learning"; no explanation of coincidence detection or noise filtering | 0.25 / 1.0 |
| TOTAL | 1.5 / 4 |
| Criterion | Max |
|---|---|
| Framework: Clearly explains predictive processing / Bayesian inference — brain generates predictions, compares with sensory input, transmits prediction errors | 5 |
| Evidence breadth: Uses multiple specific examples from the question prompt to support the framework (at least 3–4 phenomena explained through the lens of prediction) | 6 |
| Evidence depth: Demonstrates understanding of the neural or mechanistic basis (e.g., mechanoreceptor types, homunculus distortion, efference copy, cortical remapping) | 5 |
| Advantages & vulnerabilities: Discusses both why predictive perception is adaptive AND how it creates systematic failures | 5 |
| Argument quality: Answer reads as a coherent argument, not a list of facts; draws connections between phenomena rather than treating them in isolation | 4 |
The predictive processing framework proposes that the brain is not a passive receiver of sensory data but an active prediction engine. At every level of the sensory hierarchy, cortex generates top-down predictions about expected input and compares them with bottom-up sensory signals. Only the difference — the prediction error — is transmitted upward. This is computationally equivalent to Bayesian inference: the brain maintains prior beliefs about the world, updates them with incoming sensory evidence (likelihood), and the result (posterior) is what we consciously perceive. Perception, on this view, is the brain's best guess, not a readout of reality.
This framework unifies phenomena that otherwise seem unrelated. Mechanoreceptor specialization (Merkel cells for edges, Pacinian corpuscles for vibration) makes sense as efficient encoding of the features the brain's generative model most needs to predict. The cortical homunculus — with its grotesquely enlarged hands and lips — reflects that body regions requiring finer predictions get more cortical territory, not that those regions are "more important" in some abstract sense. In the rubber hand illusion, the brain's prediction about hand location is updated by correlated visual-tactile input until the fake hand is incorporated into the body model; this works because the brain always infers the most probable cause of correlated multisensory signals.
Phantom limb pain reveals the dark side: after amputation, the cortical body model still predicts sensory input from the missing limb. With no sensory evidence to correct the prediction, the model generates pain as an unconstrained prediction error signal. Ramachandran's mirror therapy exploits this by providing visual evidence of movement, allowing the prediction model to update and reduce pain — a direct clinical application of predictive processing. Sensory attenuation (you can't tickle yourself) follows from efference copy: when you initiate a movement, the brain predicts its sensory consequences and suppresses them, reducing prediction error to near zero. Someone else's touch generates a large prediction error because it was unpredicted.
The advantages of predictive perception are clear: speed (predictions arrive before sensory data), efficiency (transmitting only errors saves bandwidth), and robustness (priors fill in missing data — why you don't notice your blind spot). But these same features create vulnerabilities. When priors dominate over weak sensory evidence, you get hallucinations. When the body model fails to update after injury, you get chronic pain. Cultural differences in susceptibility to illusions like the Müller-Lyer (stronger in "carpentered" environments) show that priors are learned, not innate — the brain's predictive model is shaped by experience, making perception both powerful and systematically biased.
| Framework clearly explained (prediction, error, Bayesian logic) | 5 / 5 |
| Evidence breadth (mechanoreceptors, homunculus, rubber hand, phantom limb, tickling, Müller-Lyer, blind spot) | 6 / 6 |
| Evidence depth (efference copy, cortical remapping, mirror therapy mechanism) | 4 / 5 |
| Advantages & vulnerabilities (speed, efficiency, hallucinations, chronic pain, cultural bias) | 4 / 5 |
| Argument quality — excellent; each phenomenon flows naturally from the framework | 3 / 4 |
| TOTAL | 22 / 25 |
The Bayesian brain hypothesis says the brain doesn't just take in sensory information passively. Instead it makes predictions about what it expects and then checks them against actual input. The formula is posterior = likelihood times prior, meaning your perception is a combination of what you already believe and what the senses tell you.
There are many examples of this. The skin has different receptors like Merkel cells for texture and Pacinian corpuscles for vibration. The homunculus in somatosensory cortex is distorted, with big hands and lips because those areas are more sensitive. In the rubber hand illusion, people start to feel the fake hand as their own after synchronized touching, because the brain updates its prediction of where the hand is.
Phantom limb pain happens when someone loses a limb but still feels pain because the brain expects signals that aren't coming. Mirror therapy can help by giving the brain visual feedback. You can't tickle yourself because your brain predicts the sensation and cancels it out.
The advantages of this system are that it's fast and efficient. The problems are that it can lead to hallucinations when predictions are wrong, and chronic pain when the body model doesn't update. Different cultures are more or less susceptible to visual illusions depending on their environment.
| Framework — states the idea but doesn't explain error transmission or hierarchical prediction | 3 / 5 |
| Evidence breadth — touches on many phenomena but surface-level treatment | 4 / 6 |
| Evidence depth — mentions receptors and mirror therapy but mechanistic detail is thin (no efference copy, no cortical remapping explanation) | 2 / 5 |
| Advantages & vulnerabilities — listed but not explained in terms of the framework | 3 / 5 |
| Argument quality — reads as a list of examples organized by topic, not as an argument flowing from the framework | 3 / 4 |
| TOTAL | 15 / 25 |
The brain uses Bayesian inference to make sense of the world. This means it combines prior knowledge with new information. It is a good system because it helps us perceive things quickly.
There are different kinds of receptors in the skin that detect different things. The cortical homunculus shows how the brain maps the body. The rubber hand illusion is where you can trick the brain into thinking a rubber hand is your own. Phantom limb pain is when people still feel pain in a limb that is gone. You can't tickle yourself because your brain knows you're doing it.
This system can go wrong and cause hallucinations or pain. Overall the Bayesian brain theory is a good way to explain perception.
| Framework — mentioned by name but not explained; no prediction errors, no hierarchy, no actual Bayesian logic beyond "combines prior with new" | 1 / 5 |
| Evidence breadth — names several phenomena but doesn't explain any through the predictive lens | 2 / 6 |
| Evidence depth — no mechanistic detail at all; each example is one sentence with no neural basis | 1 / 5 |
| Advantages & vulnerabilities — "fast" and "hallucinations" mentioned but without explanation | 2 / 5 |
| Argument quality — a disconnected list of definitions, not an argument; no phenomenon is explained by the framework | 2 / 4 |
| TOTAL | 8 / 25 |