prediction error, there is a preferred level of surprisal that leads to a maximally affective response, plotted on the y-axis. However, this optimal surprisal level is not uniform over musical features (e.g., expressive timing, harmonic structure), but rather is closely coupled to the specific characteristics of that musical feature or behaviour. As Clark states, context sensitivity is funda- mental, and in the case of music, different levels of constraint will exist simultaneously across different systems of pitch space and time. For example: Singing often has high constraints in terms of pitch, tuning, and scale, while timing constraints may be more flexible; but drumming usually involves strict timing con- straints, with more flexibility in terms of pitch. Our perceptual systems are finely attuned to these constraints, to the point that rhythmic deviations that fit with certain aspects of perceived musical structure are less well detected (Repp 1999), and humanly produced deviations from a steady rhythm are preferred over randomly added noise (Hennig et al. 2011). This tuning of our perceptual system to specific deviations from an internal model is seen not only in performance aspects of music (such as expressive microtiming), but also in compo- sitional aspects found in the score (such as syncopation). Most musical styles require and indeed “play” with levels of surprisal in the temporal domain, from the musical rubato of Romantic piano performance, to the syncopated off-beat rhythms of jazz, to the complex polyrhythms of African percussion. Proficient musicians and composers are implicitly aware of these effects, and tailor their efforts to interact with the surprisal responses of the listener. This leads to what has been coined “communica- tive pressure” in creating music (Temperley 2004): an implicit knowledge of the musical dimension in which prediction can be manipulated stylistically, without leading to a lack of clarity of the musical ideas. While this complexity corresponds closely to what Clark refers to as a designed environment, it is important to note that different musical environments have different rules, that different listeners (due to their different exposure backgrounds, such as culture and training) seek different environ- ments, and that the desired outcome is a complex affective response. Indeed, exposure has been shown to influence liking for a completely new musical system after only 30 minutes of exposure (Loui et al. 2010). This finding supports the idea of a strong personalized configuration of one’s own preference for unpredictability, reflected in musical likes and dislikes, as well as one’s own prediction abilities, shown to be quite stable over time per individual, affecting interpersonal coordination (Pecenka & Keller 2011). An individual personality might be thrill-seeking and seek out highly unpredictable new musical experiences, or, more commonly, might seek out highly predict- able familiar, favorite musical experiences. Thus, different kinds of musical experience, different musical styles, and personal musical preferences lead to different predic- tions, error responses, arousal, and affect responses across a range of musical dimensions and hierarchical levels. The upshot is that the surprisal response is non-uniform for music: The positioning of a curve describing “optimal surprisal” for affective or aesthetic reward will be determined by culture, training, or musical style, and its precise shape (e.g., kurtosis) may be specific to the type and level of the prediction or mental model. And while the charac- teristics of the optimal surprisal for each aspect of music differs, the commonality remains affect, which, we propose, plays a major part in what makes prediction error in music (large or small) meaningful, and indeed determines its value. To the extent that prediction is established as a powerful mech- anism in conveying musical meaning, it seems clear then that it is the affective response to the prediction error that gives the initial prediction such power. We thus propose that the valence of the prediction error, leading to a range of affective responses, is a necessary component of the description of how predictive proces- sing can explain musical behaviour. The function of such affective predictability will require discussion elsewhere, but we postulate that this will include deep connections with social understanding and communication, from simple group clapping, a uniquely human behaviour requiring constant automatic adjustments of probabilistic representation (Molnar-Szakacs & Overy 2006; Overy & Molnar-Szakacs 2009), to more sophisticated rhythmic organization and self-expression (Nelson 2012) with an emphasis on “error” as positive, meaningful information. Extending predictive processing to the body: Emotion as interoceptive inference doi:10.1017/S0140525X12002270 Anil K. Seth a,b and Hugo D. Critchley a,c a Sackler Centre for Consciousness Science, University of Sussex, Brighton BN1 9QJ, United Kingdom; b Department of Informatics, University of Sussex, Brighton BN1 9QJ, United Kingdom; c Department of Psychiatry, Brighton and Sussex Medical School, Brighton BN1 9QJ, United Kingdom. a.k.seth@sussex.ac.uk H.Critchley@bsms.ac.uk www.anilseth.com www.sussex.ac.uk/sackler/ Abstract: The Bayesian brain hypothesis provides an attractive unifying framework for perception, cognition, and action. We argue that the framework can also usefully integrate interoception, the sense of the internal physiological condition of the body. Our model of “interoceptive predictive coding” entails a new view of emotion as interoceptive inference and may account for a range of psychiatric disorders of selfhood. In his compelling survey, Clark powerfully motivates predictive processing as a framework for neuroscience by considering the “view from inside the black box,” the notion that the brain must discover information about the world without any direct access to its source. The ensuing discussion, and the large majority of the literature surveyed, is focused on just these relations between brain and (external) world. Perhaps underemphasized in this view is the question of how perceptions of the body and self arise. However, the brain’s access to the facts of its embodi- ment and of its physiological milieu is arguably just as indirect as its access to the surrounding world. Here, we extend Clark’s integrative analysis by proposing that interoception – the sense of the physiological condition of the body (see Craig 2003) – can also be usefully considered from the perspective of predictive pro- cessing. Our model of “interoceptive predictive coding” (Critchley & Seth 2012; Seth et al. 2011) suggests a new view of emotional feelings as interoceptive inference, and sheds new light on disso- ciative disorders of self-consciousness. Interoceptive concepts of emotion were crystallized by James (1890) and Lange (1885/1912), who argued that emotions arise from perception of changes in the body. This basic idea remains influential more than a century later, underpinning frameworks for understanding emotion and its neural substrates, such as the “somatic marker hypothesis” (Damasio 2000) and the “sentient self” model (Craig 2009), both linked to the notion of “interocep- tive awareness” or “interoceptive sensitivity” (Critchley et al. 2004). Despite the neurobiological insights emerging from these frameworks, interoception has remained generally understood along “feedforward” lines, similar to classical feature-detection or evidence-accumulation theories of visual perception as sum- marized by Clark. However, it has long been recognised that expli- cit cognitions and beliefs about the causes of physiological changes influence subjective feeling states and emotional behaviour. Fifty years ago, Schachter and Singer (1962) famously demonstrated that injections of adrenaline, proximally causing a state of physio- logical arousal, would give rise to either anger or elation depend- ing on the concurrent context (an irritated or elated confederate). This observation was formalized in their “two factor” theory, in which emotional experience is determined by the combination Commentary/Andy Clark: Predictive brains, situated agents, and the future of cognitive science BEHAVIORAL AND BRAIN SCIENCES (2013) 36:3 47