Paper short abstract:

This paper examines a series of epistemic blinders in current BD analytics for security: 1) the empiricist conflation of data, information, and knowledge; 2) the commitment of serious inference errors by algorithmic processing, and 3) the distraction of BD R&D from 2nd-order non-intended effects.

Paper long abstract:

This paper examines from an epistemic perspective the current hype around Big Data (BD) in S&T and policy contexts. Increasing access to and processing of data about /by citizens seems to have given a novel twist to the post-normal science (Ravetrz & Funtowicz), and the Mode-2 of knowledge production (Gibbons, Nowotny et al.) debates of the 1990s. While the potential practical benefits from drawing upon so many, diverse, real-time, action-centred data are obvious, most R&D endeavours set to exploit that new richness of data sources stress "what", but often blend out "what for" and "how" questions. This paper consequently examines three aspects: 1) The empiricist conflation of data into information into (actionable) knowledge, where algorithmic automatisms claim to produce correlational patterns, avoiding biased causal mechanisms ("new objectivity"); 2) The radical deprivation of macro-pattern analytics from micro-contextualisation, and the risks for validity, reliability and relevance of inference and evaluation (Type I, Type II, and Type III errors); and 3) the emergence of organisational ecologies on the interface of policy, research, and the market. These promote high-tech solutionist understandings of innovation in BD applications, suppressing thereby the reflection about non-intended/non-anticipated 2nd order impacts. The paper will use illustrative cases from current BD research in public security-relevant fields, and touch upon the question of what should count as evidence, but also upon the questions of accountability diffusion in producing and instrumentalizing knowledge.