- RNP
- Sensorimotor Loop and Motor Control
- Before the movement: prediction and motor decision
- Active inference (Friston)
Sensorimotor Loop and Motor ControlSeries 11/15
SeriesBefore the movement: prediction and motor decision11 of 15
Active inference (Friston)
The continuous minimization of prediction error that constitutes both perception and action.
In short. In the predictive coding framework founded by Karl Friston (free energy principle, 2010; active inference, 2016) and Andy Clark, active inference treats the nervous system as a hierarchical predictive engine. It continuously minimizes prediction error using internal models of the world and the body. It is not passive perception of the environment, and it is not an abstract, disembodied cognitive representation. Active inference is the alternative paradigm to Gibsonian ecological dynamics. LabO RNP integrates it and links it to the four functional levels.
Why does active inference matter to the human movement professional?
Active inference is the other major paradigm in contemporary movement science, and it stands in opposition to Gibsonian ecological dynamics. Gibson proposes direct perception without internal representation. Friston instead casts the nervous system as a hierarchical predictive engine that constantly builds internal models of the body and the world to minimize prediction error. RNP integrates this paradigm into NIT. It is one of the two paradigms an RNP professional must understand to calibrate interventions to the functional level the client has reached.
Active inference does not separate perception from action. Perception is inference about the probable causes of the incoming sensory data. Action is active inference that changes the world so the incoming data match the nervous system's internal predictions. Uniting perception and action through the minimization of prediction error is one of the most important conceptual advances in cognitive science from 2010 to 2025. See also: Clark: Predictive Processing and Surfing Uncertainty (2013-2016), Efference copy (von Holst-Mittelstaedt 1950).
Active inference sheds light on several clinical phenomena that RNP practitioners encounter every day. The pre-reflexive motor decision an RNP professional reads is the silent predictive work the nervous system does before any observable motor output. A transient compensation is a local predictive strategy that bypasses an integration lock without resolving it. Lasting calibration is the updating of hierarchical internal models through accumulated minimization of prediction error. See also: Motor decision: not a command, the pre-reflexive act you have 280 milliseconds to read, Calibration vs compensation: the same result never costs the same price.
Active inference has real practical reach. It is one of the major scientific paradigms that explain motor decisions and lasting calibration. It is also one of the two paradigmatic poles that RNP articulates sequentially in the pedagogical calibration of NIT.
How active inference applies in practice
Active inference applies in three areas: reading the pre-reflexive motor decision, calibrating internal focus at the early levels, and understanding resistance to behavioral change.
Reading the pre-reflexive motor decision. The motor decision can be read in the window of two hundred to three hundred milliseconds before observable execution. It reflects the silent predictive work of the nervous system (see Motor decision: not a command, the pre-reflexive act you have 280 milliseconds to read). A trained RNP professional reads this pre-reflexive decision by spotting the micro-behavioral cues that precede it: gaze, micro-tilt, tonic pre-activation. This reading is consistent with Friston's premise that motor behavior is continuous active inference.
Calibrating internal focus at the early levels. At level 1 (conscious control) and early in level 2 (automatization), predictive coding levers can be used (see Internal versus External Focus of Attention (Wulf)). A selective internal focus on one precise detail of the movement reduces local predictive uncertainty and stabilizes the emerging coordination. Augmented feedback (KR-KP) speeds up the updating of hierarchical internal models. These levers work in the early learning window. Later, dependence on augmented feedback becomes counterproductive.
Understanding resistance to behavioral change. When a client resists changing a postural or motor strategy, the client is not at fault. Internal predictive coherence is resisting the update. The nervous system prefers to keep predicting with its existing models rather than update them, because updating costs energy.
The RNP training space, free for life
Get access to RNP training, no strings attached.
What you hear people say
The first confusion comes from a disembodied cognitivist reading. You hear "active inference is cognitive, it doesn't directly concern the body." Friston's free energy principle is explicitly embodied. Hierarchical internal models minimize prediction error across the whole body, from the cortex through subcortical structures down to spinal sensorimotor loops.
The second confusion involves Gibsonian ecological dynamics. You hear "active inference and affordance are the same thing." The two paradigms are in explicit philosophical conflict over the nature of perception. Gibson rejects internal representation; Friston makes it the fundamental structure.
The third confusion comes from a simplified marketing reading. You hear "the brain predicts, so you can sell any protocol by talking about neuroscience."
Active inference is a major scientific paradigm that explains embodied predictive motor behavior. In RNP, it is linked to ecological dynamics.
Related reading
Predictive coding lineage: Clark: Predictive Processing and Surfing Uncertainty (2013-2016) · Efference copy (von Holst-Mittelstaedt 1950)
Paradigmatic tension: Variability versus repetition
Ecological dynamics: Ecological dynamics (ecological framework) · Affordance: what your athlete sees before moving (and that you never correct) · Perception-action coupling: why your athletes fail when the information moves · Direct Perception (Gibson) · Informational invariants
Motor decision and calibration: Motor decision: not a command, the pre-reflexive act you have 280 milliseconds to read · Sensorimotor loop: not a gesture, the operational unit that decides in its place · Calibration vs compensation: the same result never costs the same price · Prediction–reafference mismatch: why your system only learns when its forecast is wrong · Motor recalibration (learning mechanism)
Practical tools: Internal versus External Focus of Attention (Wulf) · Augmented feedback (KR and KP) · Guidance hypothesis: why your help works today and betrays you tomorrow
Functional levels: Motor automatization
Historical lineage: Somatic markers (Damasio) · Craig: interoception and the insula (2002)
Going further
Active inference rests on a solid mathematical and neuroscientific lineage, and its clinical reach has only recently been put into practice.
Karl Friston (Wellcome Centre for Human Neuroimaging, University College London, 2000s-2025) formalized the free energy principle as a unified principle of self-organization in living systems. His foundational publications include The free-energy principle: a unified brain theory? (Nature Reviews Neuroscience, 2010) and Active Inference: A Process Theory (Neural Computation, 2017). Active inference is the motor application of the free energy principle, and it unifies perception and action.
Andy Clark (University of Edinburgh, then University of Sussex) developed the philosophical side of Friston's framework. He did so in Whatever Next? Predictive Brains, Situated Agents, and the Future of Cognitive Science (Behavioral and Brain Sciences, 2013) and then in Surfing Uncertainty (Oxford University Press, 2016). Clark's reading made Friston's framework more accessible to clinical practitioners without advanced mathematical training.
The historical lineage of motor predictive coding goes back to Erich von Holst and Horst Mittelstaedt (Das Reafferenzprinzip, 1950, Naturwissenschaften). They formalized the efference copy mechanism as a prediction of the sensory feedback expected from a voluntary movement. See also: Efference copy (von Holst-Mittelstaedt 1950).
Several teams have recently developed clinical applications of active inference. Some work in affective neuroscience (Andy Clark; Jakob Hohwy, The Predictive Mind, 2013). Others work in the clinical neuroscience of chronic pain, with Bayesian interpretations of chronic and nociplastic pain. These applications were still being consolidated when this glossary was written.
Several academic texts document the paradigmatic controversy between active inference and Gibsonian ecological dynamics. Examples include Bruineberg, Kiverstein and Rietveld 2018, The anticipating brain is not a scientist, and Anthony Chemero, Radical Embodied Cognitive Science, 2009. RNP explicitly declines to settle this controversy. Instead, it proposes a sequential articulation of the two paradigms, set out in the dedicated article Ecological dynamics vs. predictive coding tension.
Now, in practice
Do you work in rehabilitation? Read Behavioral resistance to change to understand why your clients with chronic pain resist changing their postural strategy. It will help you calibrate an approach that respects internal predictive coherence before trying to modify it.
Do you coach or train athletes? Read Internal versus External Focus of Attention (Wulf) to calibrate your athletes' attentional focus to the functional level they have reached: internal focus at levels 1 and 2, external focus at levels 3 and 4.
Do you support learning and emotions? Read Somatic markers (Damasio) to link active inference with the neuro-affective reading of somatic markers in children struggling at school.
Scope: a functional, non-medical approach. It complements healthcare and does not replace it. Active inference refers to a scientific framework linked to NIT, both complementary to and in tension with Gibsonian ecological dynamics.