Skip to content
Cognitive neuroscienceThe Neural Mind
Contents and reading path

Chapter 2

Sensation, Perception and Action

  1. 2.1How does the world become an experience?
  2. 2.2From stimulation to nerve signals
  3. 2.3Vision and visual perception
  4. 2.4Hearing and the direction of sound
  5. 2.5When the senses work together
  6. 2.6The cup in a noisy café
  7. 2.7How does a movement take shape?
  8. 2.8Prediction, feedback and learning
  9. LabThe visual lab

Perception and motor learning – in preparation

Reading path
Coming chapters (planned content)
  1. 1. Introduction to Cognitive Neuroscience
  2. 3. Attention
  3. 4. Memory and Learning
  4. 5. Language
  5. 6. Thinking and Executive Functions
  6. 7. Decision-Making
  7. 8. Emotion and Motivation
  8. 9. Social Cognition
  9. 10. Metacognition and Consciousness

Chapter 2 · Section 8

Prediction, feedback and learning

How can prediction help when feedback takes time, and what changes between one attempt and the next?

As the hand reaches towards the cup, both the body and sensory information change. Feedback takes time: a change must affect receptors, signals must be transmitted and information processed. If control only waited for each new signal, it would always work with an older state. Prediction can help the nervous system estimate what is happening in the meantime.

What does a forward model predict?

A forward model predicts sensory consequences from motor-command information and the body’s current state. An efference copy is a copy of command information, not a copy of a completed movement. If the hand is here and the command changes muscle activity in a particular way, what sensory information should follow?

Predicted information can be combined with incoming vision, proprioception and other sensory signals. Prediction does not replace observation or magically know about unexpected external events. If someone moves a cup you have not yet touched, a model of your own movement cannot by itself reveal the cup’s new position.

A conceptual forward modelMotor-command information branches to movement and, through an efference copy, to a forward model. Current state also contributes to the model. The model predicts sensory outcome; incoming sensory information gives observed outcome. Observed minus predicted outcome is a sensory prediction error that can contribute to updating. Arrows are conceptual, not specific neural tracts.Motor commandefference copyMovementForward modelObserved outcomeincoming sensory informationPredicted outcomeComparisonobserved − predictedSensory prediction errorcan contribute to updating across attemptscurrent state
Figure 2.10. Prediction and incoming information. A conceptual forward model: command information branches to movement and prediction. Comparing observed and predicted outcome differs from comparing an outcome with the goal. Arrows do not identify specific neural tracts.Original schematic illustration.

Two different questions about error

Sensory prediction error here means observed sensory outcome minus predicted sensory outcome, as a conceptual difference. Goal error instead asks: how does the outcome differ from what I wanted to achieve? The hand could arrive at its predicted position yet be beside a displaced cup: the goal is missed without the hand’s sensory outcome necessarily differing from its prediction.

When the cup moves

Vision can provide new information about a displaced object. Even with vision blocked, proprioception remains, but by itself it does not reveal that an untouched cup has moved. Predictions need new information to account for an unexpected external event.

Vision condition

Reader’s overview — not the mover’s visual field

Physical cup position and available informationVision available. The reach begins. The cup and hand are initially visible. The reach starts from their estimated positions; proprioception contributes information about the arm and hand.Seen cupHandSymbolic positions — no movement trajectory

Stage 1/4 · The reach begins

Vision available · Proprioception intact

The cup and hand are initially visible. The reach starts from their estimated positions; proprioception contributes information about the arm and hand.

Figure 2.11. Original information schematic, not a measured or simulated movement trajectory. The overview shows physical positions that are not always known to the mover.

Correct now, learn for next time

A change to the hand during an ongoing movement is a correction within that attempt. Motor adaptation is an adjustment across attempts when conditions change. Correcting a movement and changing the next movement are therefore not the same observation. One can occur without the other following in precisely the same way.

An unexpected sensory consequence can contribute to changing a prediction or control strategy. Skill learning also involves other processes, such as finding suitable strategies and developing more stable coordination with practice. Adaptation is not the whole of learning.

What does the evidence show?

Saunders and Knill (2003) perturbed visual information about the hand during reaching. Participants’ corrections show that ongoing visual feedback from the hand is used for regulation during movement. The study concerns this contribution, not vision alone controlling every movement.

Tseng and colleagues (2007) studied adaptation under visuomotor perturbations of reaching. Their results support a contribution of sensory prediction errors to cerebellum-dependent adaptation even when immediate corrections are distinguished from learning across attempts. This is evidence from a particular experimental situation, not proof that the same mechanism explains all learning.

Corrections and changes across attempts are observations. Forward models are one way to explain how prediction can contribute. The cerebellum is an important part of the networks under study, but neither the model nor the structure is a conscious little planner or one uniquely proven comparison centre. How different signals, networks and learning processes interact remains a research question.

Connect this to visual processing, to multisensory processing and to the cup in the café. The same action may require estimating, predicting, observing and reassessing — without these forming a rigid sequence of isolated stages.

Sources for this section
  • Scientific referenceWolpert, D. M., Miall, R. C. & Kawato, M. (1998). Internal models in the cerebellum. Trends in Cognitive Sciences, 2(9), 338–347. DOI: 10.1016/S1364-6613(98)01221-2.A model-based account of prediction and cerebellar contributions, not a direct anatomical map. Open the source ↗ (new tab)
  • Scientific referenceShadmehr, R., Smith, M. A. & Krakauer, J. W. (2010). Error correction, sensory prediction, and adaptation in motor control. Annual Review of Neuroscience, 33, 89–108. DOI: 10.1146/annurev-neuro-060909-153135.Review of feedback, sensory prediction and motor adaptation. Open the source ↗ (new tab)
  • Scientific referenceTseng, Y.-W., Diedrichsen, J., Krakauer, J. W., Shadmehr, R. & Bastian, A. J. (2007). Sensory prediction errors drive cerebellum-dependent adaptation of reaching. Journal of Neurophysiology, 98(1), 54–62. DOI: 10.1152/jn.00266.2007.Sensory prediction errors and cerebellum-dependent adaptation under visuomotor perturbations of reaching; not all forms of learning. Open the source ↗ (new tab)
  • Scientific referenceSaunders, J. A. & Knill, D. C. (2003). Humans use continuous visual feedback from the hand to control fast reaching movements. Experimental Brain Research, 152(3), 341–352. DOI: 10.1007/s00221-003-1525-2.Visual feedback from the hand during fast reaching movements. Open the source ↗ (new tab)
  • Figure provenance and model limitations are described on the sources page. About sources and figures