Ezequiel Di Paolo


Escape from pervasive individualism: Why should embodied cognition seriously study the collective dynamics of social interaction?

When studying social cognition, adaptive behaviour and robotics research has tended to adopt without much questioning the methodological individualism that is prevalent in computational cognitive science, neuroscience and psychology. Accordingly, the challenges of social cognition reside in the cognitive architecture of the individual who must confront a social situation as a particularly complex sort of problem-solving. There are good reasons to question this point of departure. Firstly, dynamical and embodied alternative frameworks propose a view of the cognizer as an autonomous actor, an exploring being who does things in the world as opposed to a detached problem-solver. In no other realm is this aspect more striking than in social behaviour. Secondly, systemic perspectives allow for a non-mysterious description of interaction between dynamical levels from neural and bodily to the collective process of interaction. There is no a priori natural order between these levels and exploring how they influence each other during behaviour and development becomes the new challenge for an enactive theory of social cognition. And finally, artificial life and embodied AI are particularly well suited for exploring and clarifying such complex relations between individual and collective dynamics. In short, for the first time in history we have not just good motivations and good (if young) alternative frameworks, but also - and crucially - good tools for escaping methodological individualism. This talk will argue for a view on social behaviour that is not totally new, but is hardly orthodox in modern approaches to social cognition (although everyone pays lip service to it). A view based on the interaction process and its influence on the sense-making activity of the interactors. Several modelling studies of collective dynamics using evolutionary robotics methods will be discussed. In particular, it will be shown how the recognition of social contingency can happen at the collective level even when the individual actors do not have the cognitive capacity to achieve such performance. This has implications for understanding social behaviour in newborns in terms of more parsimonious hypotheses. Some of these models have now successfully predicted human social performance and reveal in detail how social coordination can structure not only the timing aspects of an interaction but the perceptual processes of the individual interactors. There are several implications for a new wave of synthetic modelling studies of social cognition. Interaction as a process must be much better understood. Our modelling assumptions about interaction are not innocent; they deeply affect the significance of the results we obtain. If we fully prescribe the structure of an interaction in our model, we freeze up the most fundamental layer of meaning-making, and complex behaviours such as communication cannot be fully understood (let alone language). The challenges facing artificial life modelling of social cognition are not how to obtain more complex communication per se, but real communication, proper social generation of meaning, and the abandoning of methodological individualism.