Complex Data: Notes on the complexities of simplification in data-driven HRI experiments

TY - ABST

T1 - Complex Data

T2 - 6th NordicSTS conference

AU - Vejlin, Frederik

PY - 2023/6/7

Y1 - 2023/6/7

N2 - Common critiques of the present ‘data moment’ accuse algorithmic systems of abusing sophisticated technical simplifications to transform, and thereby distort, real-world complexity into machine-legible data through processes of classification, clustering, ranking, and prediction (Birhane 2021; Douglas-Jones, Walford, and Seaver 2021; Mackenzie 2015; Maguire et al. 2020). In STS, similar concerns have been raised in relation to the production of scientific knowledge (e.g., Mol and Law 2002), where observations must be purified of confounding artefacts – like materiality, sociality, and subjectivity (Law 2004) – to become ‘proper data’. In this talk, I bring these two perspectives together to explore the relations between complexity and datafication in so-called data-driven human-robot interaction (HRI). Data-driven HRI was developed at the Hiroshi Ishiguro Laboratories (HIL) in Japan as a hybrid algorithmic-experimental method for automatically generating robotic behaviour by training algorithms on human interaction data. The roboticists recognized that the dynamics of sociality are far too complex to be modelled predictably and require the use of ‘scalable’ methods that somehow both replicate and disregard social complexity (Liu et al. 2016). Based on material from fieldwork at the HIL, I unpack practices of datafication and experimentation in data-driven HRI to show how the interlacing of multiple simplification devices – algorithmic, experimental, technological, social – weaves equivocal relations between allegedly ‘simple’ data and ‘complex’ sociality. Inspired by Talia Dan-Cohen (2017, 2020; also, Mol and Law 2002), I emphasise the importance of probing the ontological politics of complexity enacted both in sociotechnical data studies and processes of datafication in algorithmic and experimental systems.

AB - Common critiques of the present ‘data moment’ accuse algorithmic systems of abusing sophisticated technical simplifications to transform, and thereby distort, real-world complexity into machine-legible data through processes of classification, clustering, ranking, and prediction (Birhane 2021; Douglas-Jones, Walford, and Seaver 2021; Mackenzie 2015; Maguire et al. 2020). In STS, similar concerns have been raised in relation to the production of scientific knowledge (e.g., Mol and Law 2002), where observations must be purified of confounding artefacts – like materiality, sociality, and subjectivity (Law 2004) – to become ‘proper data’. In this talk, I bring these two perspectives together to explore the relations between complexity and datafication in so-called data-driven human-robot interaction (HRI). Data-driven HRI was developed at the Hiroshi Ishiguro Laboratories (HIL) in Japan as a hybrid algorithmic-experimental method for automatically generating robotic behaviour by training algorithms on human interaction data. The roboticists recognized that the dynamics of sociality are far too complex to be modelled predictably and require the use of ‘scalable’ methods that somehow both replicate and disregard social complexity (Liu et al. 2016). Based on material from fieldwork at the HIL, I unpack practices of datafication and experimentation in data-driven HRI to show how the interlacing of multiple simplification devices – algorithmic, experimental, technological, social – weaves equivocal relations between allegedly ‘simple’ data and ‘complex’ sociality. Inspired by Talia Dan-Cohen (2017, 2020; also, Mol and Law 2002), I emphasise the importance of probing the ontological politics of complexity enacted both in sociotechnical data studies and processes of datafication in algorithmic and experimental systems.

M3 - Conference abstract for conference

Y2 - 7 June 2023 through 9 June 2023

ER -