CoChat
Sign Up
Cited by 361Embodied and Extended CognitionAction Observation and SynchronizationChild and Animal Learning Development

The DOI matches its Crossref record; 81 of the 82 references it cites were found in a registry or an index.

A duet for one.

Open access
Friston, Karl, Frith, Christopher·2015·Consciousness and Cognition
DOI 10.1016/j.concog.2014.12.003PMC PMC4553904
Open this paper in CoChat — free accountOpen PDF

Abstract

This paper considers communication in terms of inference about the behaviour of others (and our own behaviour). It is based on the premise that our sensations are largely generated by other agents like ourselves. This means, we are trying to infer how our sensations are caused by others, while they are trying to infer our behaviour: for example, in the dialogue between two speakers. We suggest that the infinite regress induced by modelling another agent - who is modelling you - can be finessed if you both possess the same model. In other words, the sensations caused by others and oneself are generated by the same process. This leads to a view of communication based upon a narrative that is shared by agents who are exchanging sensory signals. Crucially, this narrative transcends agency - and simply involves intermittently attending to and attenuating sensory input. Attending to sensations enables the shared narrative to predict the sensations generated by another (i.e. to listen), while attenuating sensory input enables one to articulate the narrative (i.e. to speak). This produces a reciprocal exchange of sensory signals that, formally, induces a generalised synchrony between internal (neuronal) brain states generating predictions in both agents. We develop the arguments behind this perspective, using an active (Bayesian) inference framework and offer some simulations (of birdsong) as proof of principle.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1016/j.concog.2014.12.003

Cites

81 of 81 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 This figure summarizes hierarchical neuronal message passing in predictive coding using the (simplified) neuroanatomy of a songbird. Neuronal activity encodes expectations about the causes of sensory input, where these expectations minimize prediction error. Prediction error is the difference between (ascending) sensory input and (descending) predictions of that input. This minimization rests upon recurrent neuronal interactions among different levels of the cortical hierarchy. The available evidence suggests that superficial pyramidal cells (red triangles) compare the expectations (at each level) with top–down predictions from deep pyramidal cells (black triangles) of higher levels. Left panel: these equations represent the neuronal dynamics implicit in predictive coding. Prediction errors at the ith level of the hierarchy are simply the difference between the expectations encoded at that level and top–down predictions of those expectations. The expectations per se are driven by prediction errors so that they perform a gradient ascent on the sum of squared (precision weighted) prediction error. See the appendix for a detailed explanation of these (simplified) equations. Right panel: this provides a schematic example in the auditory system of a songbird: it shows the putative cells of origin of ascending or forward connections that convey (precision weighted) prediction errors (red arrows) and descending or backward connections (black arrows) that construct predictions. In this example, area X sends predictions to the higher vocal centre, which projects to the auditory thalamus. However, the higher vocal centre also sends proprioceptive predictions to the hypoglossal nucleus, which are passed to the syrinx to generate vocalisation through classical reflexes. These predictions can be regarded as motor commands, while the descending predictions of auditory input correspond to corollary discharge. Note that every top–down prediction is reciprocated with a bottom–up prediction error to ensure predictions are constrained by sensory information. Please see the Appendix for a description of the equations and a definition of the variables. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 1 This figure summarizes hierarchical neuronal message passing in predictive coding using the (simplified) neuroanatomy of a songbird. Neuronal activity encodes expectations about the causes of sensory input, where these expectations minimize prediction error. Prediction error is the difference between (ascending) sensory input and (descending) predictions of that input. This minimization rests upon recurrent neuronal interactions among different levels of the cortical hierarchy. The available evidence suggests that superficial pyramidal cells (red triangles) compare the expectations (at each level) with top–down predictions from deep pyramidal cells (black triangles) of higher levels. Left panel: these equations represent the neuronal dynamics implicit in predictive coding. Prediction errors at the ith level of the hierarchy are simply the difference between the expectations encoded at that level and top–down predictions of those expectations. The expectations per se are driven by prediction errors so that they perform a gradient ascent on the sum of squared (precision weighted) prediction error. See the appendix for a detailed explanation of these (simplified) equations. Right panel: this provides a schematic example in the auditory system of a songbird: it shows the putative cells of origin of ascending or forward connections that convey (precision weighted) prediction errors (red arrows) and descending or backward connections (black arrows) that construct predictions. In this example, area X sends predictions to the higher vocal centre, which projects to the auditory thalamus. However, the higher vocal centre also sends proprioceptive predictions to the hypoglossal nucleus, which are passed to the syrinx to generate vocalisation through classical reflexes. These predictions can be regarded as motor commands, while the descending predictions of auditory input correspond to corollary discharge. Note that every top–down prediction is reciprocated with a bottom–up prediction error to ensure predictions are constrained by sensory information. Please see the Appendix for a description of the equations and a definition of the variables. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)CC BY · PubMed Central
Fig. 2 Schematic showing the construction of the generative model for birdsongs. The upper panel illustrates the generative model that comprises two Lorenz attractors, where the higher attractor delivers two control parameters (cyan circles) to a lower level attractor, which, in turn, delivers two control parameters to a synthetic syrinx to produce amplitude and frequency modulated stimuli. This stimulus is represented as a sonogram in the right panel. The equations represent the hierarchical dynamic model in the form described in the appendix. The lower panel illustrates the setup that we will be using, where sensory signals are generated using the scheme in the upper panel and are subsequently used to produce predictions and prediction errors (of the sort that can be measured empirically).
Fig. 2 Schematic showing the construction of the generative model for birdsongs. The upper panel illustrates the generative model that comprises two Lorenz attractors, where the higher attractor delivers two control parameters (cyan circles) to a lower level attractor, which, in turn, delivers two control parameters to a synthetic syrinx to produce amplitude and frequency modulated stimuli. This stimulus is represented as a sonogram in the right panel. The equations represent the hierarchical dynamic model in the form described in the appendix. The lower panel illustrates the setup that we will be using, where sensory signals are generated using the scheme in the upper panel and are subsequently used to produce predictions and prediction errors (of the sort that can be measured empirically).CC BY · PubMed Central
Fig. 3 Omission-related responses: The left panels show the original song and responses evoked. The right panels show the equivalent dynamics on omission of the last chirps. The top panels show the stimulus and the middle panels the corresponding percept in sonogram format. The interesting thing to note here is the occurrence of an anomalous percept after termination of the song on the lower right. This corresponds roughly to the chirp that would have been perceived in the absence of omission. The lower panels show the corresponding (precision weighted) prediction error under the two stimuli at both levels. A comparison of the two reveals a burst of prediction error when a stimulus is missed and at the point that the stimulus terminates – even when there is no stimulus present at this time. The darker lines correspond to prediction error at the first level and the lighter lines correspond to prediction error at the second level.
Fig. 3 Omission-related responses: The left panels show the original song and responses evoked. The right panels show the equivalent dynamics on omission of the last chirps. The top panels show the stimulus and the middle panels the corresponding percept in sonogram format. The interesting thing to note here is the occurrence of an anomalous percept after termination of the song on the lower right. This corresponds roughly to the chirp that would have been perceived in the absence of omission. The lower panels show the corresponding (precision weighted) prediction error under the two stimuli at both levels. A comparison of the two reveals a burst of prediction error when a stimulus is missed and at the point that the stimulus terminates – even when there is no stimulus present at this time. The darker lines correspond to prediction error at the first level and the lighter lines correspond to prediction error at the second level.CC BY · PubMed Central
Fig. 4 creating one’s own sensations. This figure shows the results of a simulation in which the precision (gain) of proprioceptive prediction errors was increased from a log precision of −8 to 8 (highlighted by the insert on the right). This effectively switches on reflex arcs that respond to descending proprioceptive predictions from the higher vocal centre, causing the previously perceived song to be articulated. However, when we run the simulations (ensuring that the bird can hear itself) the ensuing song is unrecognisable. This is shown in the upper panel in terms of a bizarre looking sonogram that has lost the deep hierarchical structure and tempo of the sonograms in the previous figures. The middle (resp. lower) panel shows posterior expectations about hidden states in the first (resp. second) hierarchical levels. The dynamics characteristic of the Lorentz attractor have been destroyed and low amplitude high-frequency fluctuations are evident in these traces. These reflect the impact of precise prediction errors due to sensorimotor delays in the enactment of predictions.
Fig. 4 creating one’s own sensations. This figure shows the results of a simulation in which the precision (gain) of proprioceptive prediction errors was increased from a log precision of −8 to 8 (highlighted by the insert on the right). This effectively switches on reflex arcs that respond to descending proprioceptive predictions from the higher vocal centre, causing the previously perceived song to be articulated. However, when we run the simulations (ensuring that the bird can hear itself) the ensuing song is unrecognisable. This is shown in the upper panel in terms of a bizarre looking sonogram that has lost the deep hierarchical structure and tempo of the sonograms in the previous figures. The middle (resp. lower) panel shows posterior expectations about hidden states in the first (resp. second) hierarchical levels. The dynamics characteristic of the Lorentz attractor have been destroyed and low amplitude high-frequency fluctuations are evident in these traces. These reflect the impact of precise prediction errors due to sensorimotor delays in the enactment of predictions.CC BY · PubMed Central
Fig. 5 sensory attenuation during self-made acts. This figure uses the same format as Fig. 4 – and reports the same simulation. However, here, we have reduced the log precision of sensory (auditory) prediction errors from 2 to 0. This effectively prevents ascending – and irreducible – prediction errors due to sensorimotor delays from confounding the dynamics in the higher vocal centre. This means the higher vocal centre can provide veridical top–down proprioceptive predictions and elicit and uncorrupted birdsong. This example illustrates the permissive nature of sensory attenuation during the enactment of descending predictions through open loop control.
Fig. 5 sensory attenuation during self-made acts. This figure uses the same format as Fig. 4 – and reports the same simulation. However, here, we have reduced the log precision of sensory (auditory) prediction errors from 2 to 0. This effectively prevents ascending – and irreducible – prediction errors due to sensorimotor delays from confounding the dynamics in the higher vocal centre. This means the higher vocal centre can provide veridical top–down proprioceptive predictions and elicit and uncorrupted birdsong. This example illustrates the permissive nature of sensory attenuation during the enactment of descending predictions through open loop control.CC BY · PubMed Central
Fig. 6 a soliloquy for two. In this simulation, two birds with the same generative models – but different initial expectations – sing for two seconds and then listen for any response. However, here, the birds cannot hear each other (e.g., they are too far apart) and the successive epochs of songs diverge due to the sensitivity to initial conditions implicit in these (chaotic) generative models. The upper panel shows the sonogram heard by the first (red) bird. Because this bird can only hear itself, the sonogram reflects the proprioceptive predictions based upon posterior expectations in the higher vocal centre (middle panel) and area X (lower panel). The posterior expectations for the first bird are shown in red as a function of time – and the equivalent expectations for the second bird are shown in blue. Note that when both birds are listening, their expectations at the first level fall to zero – because they do not hear anything and auditory input is attended to (i.e., has a relatively high precision). This does not destroy the slower dynamics in area X, which is able to generate the song again after the end of each listening period. Note also that the second (blue) bird takes a few hundred milliseconds before it starts singing. This is because it takes a little time for the posterior expectations to find the attractor manifold prescribed by the higher level control parameters. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 6 a soliloquy for two. In this simulation, two birds with the same generative models – but different initial expectations – sing for two seconds and then listen for any response. However, here, the birds cannot hear each other (e.g., they are too far apart) and the successive epochs of songs diverge due to the sensitivity to initial conditions implicit in these (chaotic) generative models. The upper panel shows the sonogram heard by the first (red) bird. Because this bird can only hear itself, the sonogram reflects the proprioceptive predictions based upon posterior expectations in the higher vocal centre (middle panel) and area X (lower panel). The posterior expectations for the first bird are shown in red as a function of time – and the equivalent expectations for the second bird are shown in blue. Note that when both birds are listening, their expectations at the first level fall to zero – because they do not hear anything and auditory input is attended to (i.e., has a relatively high precision). This does not destroy the slower dynamics in area X, which is able to generate the song again after the end of each listening period. Note also that the second (blue) bird takes a few hundred milliseconds before it starts singing. This is because it takes a little time for the posterior expectations to find the attractor manifold prescribed by the higher level control parameters. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)CC BY · PubMed Central
Fig. 7 communication and generalised synchrony. This figure uses the same format as Fig. 6; however, here, we have juxtaposed the two birds so that they can hear each other. In this instance, the posterior expectations show identical synchrony at both the first and second hierarchical levels – as shown in the middle and lower panels respectively. Note that the sonogram is continuous over successive one second epochs – being generated alternately by the first and second bird.
Fig. 7 communication and generalised synchrony. This figure uses the same format as Fig. 6; however, here, we have juxtaposed the two birds so that they can hear each other. In this instance, the posterior expectations show identical synchrony at both the first and second hierarchical levels – as shown in the middle and lower panels respectively. Note that the sonogram is continuous over successive one second epochs – being generated alternately by the first and second bird.CC BY · PubMed Central
Fig. 8 the synchronisation manifold. These graphs plot the second level (area X) expectations in the second bird against the equivalent expectations in the first. The left panel shows chaotic and uncoupled dynamics when the birds cannot hear each other, while the right panel shows the generalised (identical) synchrony that emerges when the birds exchange sensory signals. The different colours correspond to the three hidden states for each bird. The synchronisation manifold for identical synchronisation would correspond to the X equals Y line.
Fig. 8 the synchronisation manifold. These graphs plot the second level (area X) expectations in the second bird against the equivalent expectations in the first. The left panel shows chaotic and uncoupled dynamics when the birds cannot hear each other, while the right panel shows the generalised (identical) synchrony that emerges when the birds exchange sensory signals. The different colours correspond to the three hidden states for each bird. The synchronisation manifold for identical synchronisation would correspond to the X equals Y line.CC BY · PubMed Central
Fig. 9 A predictive coding formulation of the Hermeneutic Circle.
Fig. 9 A predictive coding formulation of the Hermeneutic Circle.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

The self-check needs the paper’s full text.

Citations

Cites

Show all 82 references

Research works printed but not linked: 1

  1. Clark, 2013, The many faces of precision, Frontiers in Psychology, 4, 270

Cited by

Sorted by citations

Items 301–361 of 361

Related

See all

Metadata

Authors
Friston, K., & Frith, C.
Affiliations
Wellcome Centre for Human Neuroimaging (GB)
Corresponding author
Friston, Karl
ORCID
Friston, Karl
Venue
Consciousness and Cognition
Publisher
Elsevier BV
Published
2015-11
Volume
36
Pages
390-405
ISSN
1053-8100 (ISSN-L)
DOI
10.1016/j.concog.2014.12.003
PMC
PMC4553904
OpenAlex
W2024149692
Cited by
359
References
82
Citations by year
2026: 44 · 2025: 31 · 2024: 33 · 2023: 43 · 2022: 41 · 2021: 41 · 2020: 21 · 2019: 22 · 2018: 30 · 2017: 26 · 2016: 22 · 2015: 5
FWCI
16.63
Type
journal article
Language
en
Open access
Licence
Locations
  • Consciousness and Cognition (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • UCL Discovery (University College London) (submittedVersion)
Funders
Wellcome Trust (091593, 088130)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Embodied and Extended Cognition
  • Social Sciences › Psychology › Social Psychology › Action Observation and Synchronization
  • Social Sciences › Psychology › Developmental and Educational Psychology › Child and Animal Learning Development
Keywords
communication, joint action, shared narratives, Bayesian inference, infinite regress, shared models, sensory prediction, attention, attenuation, bird song, social cognition
MeSH terms
Animals, Attention, Communication, Humans, Models, Theoretical, Perception, Sensation, Thinking, Theory of Mind

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review