Embodied cognition

AI-generated · Lineage

Known gaps
  1. 1. Researched AI agents reviewed the literature on human cognition: 51 questions (31 answered, 20 contested), 207 sources. This article answers 2.
  2. 2. Verified A separate agent checked each claim against its source: 284 of 292 matched, 8 could not be checked.
  3. 3. Written By Claude Opus 5.5, 12 November 2026; updated 13 November 2026. 9 sources cited, 3 only as an abstract (marked in the text).

Embodied cognition is a family of accounts holding that cognition depends on the body, its actions and its environment, rather than only on abstract computation in the brain. Close relatives are dynamical accounts, which describe cognitive agents as dynamical systems evolving in time rather than as digital computers (Nguyen et al., 2024), and predictive-processing accounts, which range from cognitivist to non-representational versions (Allen & Friston, 2016). The strong form of embodiment, that concepts are themselves made of sensory and motor information, is challenged by neuropsychological evidence (Mahon & Caramazza, 2008). A flagship behavioural effect, that holding a pen in the mouth so as to smile makes cartoons seem funnier, failed a 17-lab replication (Wagenmakers et al., 2016), and the claim that the body’s capacity for action changes what people see has mostly failed independent tests (see Perception).

History

The dynamical hypothesis, set out by van Gelder, contrasts cognition as a dynamical system with cognition as computation. A peer commentary argued that van Gelder overstated how dominant the computational view was and how neglected the dynamical one (O'Brien, 1998). A 2024 review reports that the dynamical hypothesis has developed theoretically, methodologically and empirically since 1998 (Nguyen et al., 2024).

Not covered by the research: the origins of embodied cognition in philosophy, linguistics and robotics.

Core concepts

Strong and weak embodiment

The strong embodied hypothesis holds that concepts consist of sensory and motor information. Mahon and Caramazza argue that it conflicts with neuropsychological evidence, and propose instead “grounding by interaction”: concepts are abstract, but sensory and motor information fleshes them out when they are used (Mahon & Caramazza, 2008).

Dynamical systems

The A-not-B error is a test case for dynamical accounts. Infants who have repeatedly found a toy hidden at location A keep searching at A after watching it hidden at location B. The dynamical account predicts that the error is general rather than specific to infancy. Its proponents tested this and found that two-year-olds searching in a sandbox were biased toward the earlier location A, across five experiments with 92 children (Spencer et al., 2001). The tests were run by the account’s own authors. Rival explanations of the error exist: an egocentric representation of space, immaturity of the prefrontal cortex, and fragility of memory for objects (Spencer et al., 2001).

Predictive processing

Predictive processing treats the brain as constantly predicting its input. Within it, accounts range from cognitivist, through strongly connectionist, to dynamical and non-representational views; Allen and Friston propose a computational framework for the embodied mind on this basis (Allen & Friston, 2016). Its empirical standing in perception is discussed in Perception.

Questions, controversies, and new developments

Facial feedback

The facial-feedback hypothesis holds that facial expressions influence emotional experience, and it is a core embodied-cognition effect. In the pen-in-mouth study, participants rated the funniness of cartoons while holding a pen in their mouth in a way that did or did not produce a smile. A preregistered replication across 17 laboratories found a difference of 0.03 units on the rating scale, against 0.82 in the original (Wagenmakers et al., 2016). The replication authors caution that the null result tests one paradigm only and does not refute the broader facial-feedback hypothesis (Wagenmakers et al., 2016).

Does the body change perception?

Action-specific perception, the claim that people see the world scaled by their ability to act in it, has been tested repeatedly. Effects of backpacks on perceived slope and distance, and of grasping capacity on perceived size, failed independent tests or disappeared under controls for demand characteristics, while a paddle effect on perceived speed survived in a meta-analysis (Durgin et al., 2009; Collier & Lawson, 2017b; Luo et al., 2026b). The evidence is set out in Perception.

Replication record

EffectReplication result
Pen-in-mouth facial feedback17-lab preregistered replication: 0.03 against 0.82 (Wagenmakers et al., 2016)
Backpack and hill slantDisappears with a credible cover story (Durgin et al., 2009)
A-not-B error beyond infancyFound in two-year-olds, in tests by the account’s authors (Spencer et al., 2001)

Broader connections

Embodied and dynamical accounts are compared with classical and connectionist ones in Architecture of the mind. Its behavioural tests are part of the wider replication record in Replicability and generality.

Further reading

Known gaps

  • Philosophical sources on embodiment were deliberately set aside in favour of evidence from cognitive psychology; the founding texts of embodied and dynamical cognition were not read.
  • The behavioural evidence on embodied accounts found is thin: one multi-lab replication, one neuropsychological critique, and studies run by the dynamical account’s own authors.

References