Andy Clark has written an impressive piece. Predictive processing ideas have been the hype in the neurocognitive community for some years, for all the reasons that the target article's review identifies. They propose to unify models of perception, action, and learning within a framework – which is elegant, aligned with neuroanatomical and functional findings, computationally plausible, and able to generate empirical research with relatively clear hypotheses.
So far the ideas have been a well-kept secret within the community. This BBS article is likely to change that. As one of the first, Clark brings the predictive processing framework in touch with more general views in cognition and philosophy of mind in a format available to a wider audience. Stripping it of the mathematical formality without losing out on the conceptual stringency, opens for a wider discussion of potential implications for how we think of the brain and of ourselves. Key terms like anticipation, expectancy, models of reality, attention, agency, and surprise appear to move seamlessly between the neuronal, the mathematical, the phenomenological, and the behavioral. The ambition to extend this to a general model of human cognition is impressive, but this is also where the proposal becomes very open-ended. For, ultimately, how human-specific is this predictive framework? In the current formulation, hardly at all. The underlying neural models are basically species-unspecific, and the empirical cases move back and forth between many different model systems. This is not a weakness of the framework; on the contrary, the ambition is to lay out a general theory of brain function, cortical responses, predictive coding, free energy, and so forth. However, it leaves a lot of work open when gauging how this relates to a specific understanding of human action and cognition.
To begin this, one may need to ask what is characteristic of humans as a life form? We don't know for sure, but there are a few candidates. One is an unusual ability for interaction – people coordinate, couple, take turns – at many different levels (Levinson Reference Levinson, Enfield and Levinson2006). Through interactions, they come to share a structuring of activities in time, and, perhaps, bring brain internal processes in sync too. Another, probably not unrelated, is an amazing ability to co-construct artefacts and build shared worlds that are at the same time material and symbolic (Clark Reference Clark2006b; Roepstorff Reference Roepstorff2008): worlds that exist outside the individual, and in time-windows, which extends beyond the here-and-now of interaction; worlds that, somehow, get internalized. Are these two principles uniquely human? Probably not: Other species also coordinate actions, and other species also modify their surroundings, building niches that are both material and cognitive, but the degree to which people do it is amazing, and we still need to figure out how this can come about, also at a cognitive level.
In sociology and anthropology, one influential attempt to relate interactions and the co-constructed shared worlds has been a focus on human practices (Bourdieu Reference Bourdieu and Nice1977; Roepstorff et al. Reference Roepstorff, Niewohner and Beck2010) as particular unfoldings of temporality set within specific materialities. Translated into predictive coding lingo, these practices may help establish priors or even hyperpriors, sets of expectations that shape perception and guide action (Roepstoff & Frith Reference Roepstoff and Frith2012). Following from this, human priors may not only be driven by statistical properties in the environment, picked up by individual experience, or hardwired into the developing cognitive system. They are also a result of shared expectations that are communicated in interactions, mediated by representations, solidified through materiality, and extended into an action space, going way beyond the physical body and into proximal and distal forms of technology.
This means that both the “predictive” and the “situated” in Clark's title may get a radical twist. It is not so much a matter of living inside a “socio-cultural cocoon,” as Clark puts it (sect. 5.2, para. 4). This metaphor suggest that we will at some point grow up and come out of the cocoon into the real world. It is also not just a matter of “man” as “an animal suspended in webs of significance he himself has spun,” as Clifford Geertz (Reference Geertz1966), following Max Weber, famously suggested. This formulation over-emphasizes the symbolic and the individualistic, and it fails to see that the webs “we” have spun are indeed also very material, and that the dimensions of materiality “we” can spin ourselves into seem to be constantly changing. Humans appear to live lives where both priors and possibilities for action – and perhaps also, increasingly, the world – are shaped by actions of others and constrained, stabilised, and afforded by those structures built in the process. But if “being human” in general is about living in unfolded practices, what, then, is it about our cognition that allows us to do that? We don't know. But something about how humans can bridge the material and the symbolic, and something about how they in and through interactions can share both external and internal time, may be critical.
The predictive framework, in “linking action, perception, and learning,” is highly relevant also to researchers outside of the neurosciences. But at this stage, there is much to fill in for it to function as a general model of human cognition and action. Certainly, the free energy principle, the predictive hierarchical stuff, the putative links between action, perception, and learning seem to be good candidates for the new “rough guide” to brain function. However, these guiding principles appear to work equally well in rats, in macaques, and in humans. For those of us who are particularly interested in what humans do to themselves, to each other, and to their world, there seem to be a lot of lacunae to be explored, and a lot of gaps to be filled. Getting these right may perhaps also teach something about what humans, as interactive agents, embedded in sociocultural worlds, may do to their brains. Will this throw new light on neuroscience too? Perhaps. There is certainly much work to be done by researchers from many disciplines.
Andy Clark has written an impressive piece. Predictive processing ideas have been the hype in the neurocognitive community for some years, for all the reasons that the target article's review identifies. They propose to unify models of perception, action, and learning within a framework – which is elegant, aligned with neuroanatomical and functional findings, computationally plausible, and able to generate empirical research with relatively clear hypotheses.
So far the ideas have been a well-kept secret within the community. This BBS article is likely to change that. As one of the first, Clark brings the predictive processing framework in touch with more general views in cognition and philosophy of mind in a format available to a wider audience. Stripping it of the mathematical formality without losing out on the conceptual stringency, opens for a wider discussion of potential implications for how we think of the brain and of ourselves. Key terms like anticipation, expectancy, models of reality, attention, agency, and surprise appear to move seamlessly between the neuronal, the mathematical, the phenomenological, and the behavioral. The ambition to extend this to a general model of human cognition is impressive, but this is also where the proposal becomes very open-ended. For, ultimately, how human-specific is this predictive framework? In the current formulation, hardly at all. The underlying neural models are basically species-unspecific, and the empirical cases move back and forth between many different model systems. This is not a weakness of the framework; on the contrary, the ambition is to lay out a general theory of brain function, cortical responses, predictive coding, free energy, and so forth. However, it leaves a lot of work open when gauging how this relates to a specific understanding of human action and cognition.
To begin this, one may need to ask what is characteristic of humans as a life form? We don't know for sure, but there are a few candidates. One is an unusual ability for interaction – people coordinate, couple, take turns – at many different levels (Levinson Reference Levinson, Enfield and Levinson2006). Through interactions, they come to share a structuring of activities in time, and, perhaps, bring brain internal processes in sync too. Another, probably not unrelated, is an amazing ability to co-construct artefacts and build shared worlds that are at the same time material and symbolic (Clark Reference Clark2006b; Roepstorff Reference Roepstorff2008): worlds that exist outside the individual, and in time-windows, which extends beyond the here-and-now of interaction; worlds that, somehow, get internalized. Are these two principles uniquely human? Probably not: Other species also coordinate actions, and other species also modify their surroundings, building niches that are both material and cognitive, but the degree to which people do it is amazing, and we still need to figure out how this can come about, also at a cognitive level.
In sociology and anthropology, one influential attempt to relate interactions and the co-constructed shared worlds has been a focus on human practices (Bourdieu Reference Bourdieu and Nice1977; Roepstorff et al. Reference Roepstorff, Niewohner and Beck2010) as particular unfoldings of temporality set within specific materialities. Translated into predictive coding lingo, these practices may help establish priors or even hyperpriors, sets of expectations that shape perception and guide action (Roepstoff & Frith Reference Roepstoff and Frith2012). Following from this, human priors may not only be driven by statistical properties in the environment, picked up by individual experience, or hardwired into the developing cognitive system. They are also a result of shared expectations that are communicated in interactions, mediated by representations, solidified through materiality, and extended into an action space, going way beyond the physical body and into proximal and distal forms of technology.
This means that both the “predictive” and the “situated” in Clark's title may get a radical twist. It is not so much a matter of living inside a “socio-cultural cocoon,” as Clark puts it (sect. 5.2, para. 4). This metaphor suggest that we will at some point grow up and come out of the cocoon into the real world. It is also not just a matter of “man” as “an animal suspended in webs of significance he himself has spun,” as Clifford Geertz (Reference Geertz1966), following Max Weber, famously suggested. This formulation over-emphasizes the symbolic and the individualistic, and it fails to see that the webs “we” have spun are indeed also very material, and that the dimensions of materiality “we” can spin ourselves into seem to be constantly changing. Humans appear to live lives where both priors and possibilities for action – and perhaps also, increasingly, the world – are shaped by actions of others and constrained, stabilised, and afforded by those structures built in the process. But if “being human” in general is about living in unfolded practices, what, then, is it about our cognition that allows us to do that? We don't know. But something about how humans can bridge the material and the symbolic, and something about how they in and through interactions can share both external and internal time, may be critical.
The predictive framework, in “linking action, perception, and learning,” is highly relevant also to researchers outside of the neurosciences. But at this stage, there is much to fill in for it to function as a general model of human cognition and action. Certainly, the free energy principle, the predictive hierarchical stuff, the putative links between action, perception, and learning seem to be good candidates for the new “rough guide” to brain function. However, these guiding principles appear to work equally well in rats, in macaques, and in humans. For those of us who are particularly interested in what humans do to themselves, to each other, and to their world, there seem to be a lot of lacunae to be explored, and a lot of gaps to be filled. Getting these right may perhaps also teach something about what humans, as interactive agents, embedded in sociocultural worlds, may do to their brains. Will this throw new light on neuroscience too? Perhaps. There is certainly much work to be done by researchers from many disciplines.