The Future of Prediction: The Social Consequences of Algorithmic Forecast

Our research on the use of algorithmic forecasting in medicine, insurance, and policing has highlighted innovative uses of prediction that do not explain phenomena but nevertheless give insights to effectively act on a future event. This invites rethinking the idea that pure correlation-based forecasts do not enable action to change the outcome, unlike explanatory predictions based on underlying causal mechanisms, which can do so. My talk describes different social forms of using forecasting that combine statistical prognoses with algorithmic predictions to indicate how to act on a specific future outcome -- in medicine one speaks of actionability, in insurance of proactivity, in policing of targeted forms of prevention. When what matters is the specific case at hand, scientific replicability and generalizability of predictions are not the priority. Controlling the accuracy and effectiveness of these predictions involves different criteria, now operationalized in a multiplicity of recently developed performance metrics.

In many fields there has been a long-standing emphasis on inference (obtaining unbiased estimates of individual effects) over prediction (forecasting future outcomes), perhaps because the latter can be quite difficult, especially when compared with the former. Here we show that this focus on inference over prediction can mislead readers into thinking that the results of scientific studies are more definitive than they actually are. Through a series of randomized experiments, we demonstrate that this confusion arises for one of the most basic ways of presenting statistical findings and affects even experts whose jobs involve producing and interpreting such results. In contrast, we show that communicating both inferential and predictive information side by side provides a simple and effective alternative, leading to calibrated interpretations of scientific results. We conclude with a more general discussion about integrative modeling, where prediction and inference are combined to complement, rather than compete with, each other.

Automated, machine-learning prediction is central to the digital economy, via platforms prominent in everyday life and what the paper calls their data “icebergs.” Achieving direct observational/interviewing access to these platforms is notoriously difficult, so the research reported here examines platforms “obliquely,” via the experiences of those who advertise on them, and follows materially disruptive episodes that render platform practices more visible.

One such episode, Apple’s 2021 App Tracking Transparency changes to iPhones, has fed into, and makes visible, a major divide. On one side are material practices of data accumulation and prediction that rely on what participants call “identity resolution”: determining as far as possible which digital traces represent actions by the same human individual. On the other side are practices that materially block identity resolution (for instance seeking to preserve what Apple calls “crowd anonymity”), thus forcing prediction to be de-individualized. Fierce public controversy in 2020-21 has now been replaced by “messy,” implicit, largely subterranean, conflict between these two sets of material arrangements.

The paper draws upon 111 interviews with 88 practitioners of digital advertising and related technical specialists, along with extensive participation in sector meetings, and, e.g., a training course on the advertising of games and other apps.

Enabled by massive investments in 'datafication', military imaginaries of command and control now materialize the 17^thcentury dream of ‘a sign system that linked all knowledge to a language and sought to replace all languages with a system of artificial symbols and operations of a logical nature’ (Foucault Order of Things 1994: 63). Conjoining socio-technologies of surveillance, mapping, categorization, and enumeration, military ‘intelligence’ rests on a platform of ‘data foundations’ whose interfaces promise privileged access to warfighting’s worlds. In positioning data as the infrastructure of knowledge making, these imaginaries take for granted the translations through which worlds of interest are rendered as machine detectable signals that feed data models at the back end and data visualizations at the interface. Treating data as given enables a focus on optimizing processing and transmission, while placing outside the frame the necessary reductions, erasures, and betrayals of data-fication. This paper reports work in progress on tracking the material semiotic sleights of hand that sustain US militarism’s mythology of its own coherence and rationality, and the counter narratives that testify to war’s senseless and ungovernable injuries and offer openings to resistance.

Computational social science is more than just large repositories of digital data and the computational methods needed to construct and analyze them. It also represents a convergence of different fields with different ways of thinking about and doing science. The goal of this Perspective is to provide some clarity around how these approaches differ from one another and to propose how they might be productively integrated. Towards this end we make two contributions. The first is a schema for thinking about research activities along two dimensions—the extent to which work is explanatory, focusing on identifying and estimating causal effects, and the degree of consideration given to testing predictions of outcomes—and how these two priorities can complement, rather than compete with, one another. Our second contribution is to advocate that computational social scientists devote more attention to combining prediction and explanation, which we call integrative modeling, and to outline some practical suggestions for realizing this goal.