In an era of global sanitary, economic and ecological crisis, beliefs in the predictive power of artificial intelligence (AI) progressively penetrate the legal and political spheres, in search of new ways to anticipate and govern the future. In this context, it is critical to understand the idiosyncratic nature of the interplay between governance and algorithmic logics of prediction. This contribution discusses how the association between governance and AI makes the future knowable in the present and shapes a programmatic way of formalising, justifying and deploying action in the here and now. We focus on three principles of institutional mobilisation in the face of uncertainty and indeterminacy: precaution, pre-emption and preparedness, each of which is affected by the use of AI relying on so-called ‘real-time predictions’. Drawing from risk theory and Science and Technology Studies, we argue that the current convergence between AI and governance is shaping a new sociotechnical imaginary, promoting a distinctive conception of life and of the future in the age of the Anthropocene.

Electronic linking of public records and predictive analytics to identify families for preventive early intervention increasingly is promoted by governments. We use the concept of social license to address questions of social legitimacy, agreement, and trust in data linkage and analytics for parents of dependent children, who are the focus of early intervention initiatives in the UK. We review data-steered family policy and early intervention operational service practices. We draw on a consensus baseline analysis of data from a probability-based panel survey of parents, to show that informed consent to data linkage and use is important to all parents, but there are social divisions of knowledge, agreement, and trust. There is more social license for data linkage by services among parents in higher occupation, qualification, and income groups, than among Black parents, lone parents, younger parents, and parents in larger households. These marginalized groups of parents, collectively, are more likely to be the focus of identification for early intervention. We argue that government awareness-raising exercises about the merits of data linkage are likely to bolster existing social license among advantaged parents while running the risk of further disengagement among disadvantaged groups. This is especially where inequalities and forecasting inaccuracies are encoded into early intervention data gathering, linking, and predictive practices, with consequences for a cohesive and equal society.

Algorithmic decision tools (ADTs) are being introduced into public sector organizations to support more accurate and consistent decision-making. Whether they succeed turns, in large part, on how administrators use these tools. This is one of the first empirical studies to explore how ADTs are being used by Street Level Bureaucrats (SLBs). The author develops an original conceptual framework and uses in-depth interviews to explore whether SLBs are ignoring ADTs (algorithm aversion); deferring to ADTs (automation bias); or using ADTs together with their own judgment (an approach the author calls “artificing”). Interviews reveal that artificing is the most common use-type, followed by aversion, while deference is rare. Five conditions appear to influence how practitioners use ADTs: (a) understanding of the tool (b) perception of human judgment (c) seeing value in the tool (d) being offered opportunities to modify the tool (e) alignment of tool with expectations.

U.S., UK, and European municipalities are increasingly experimenting with data as an instrument for social policy. This movement pertains often to the design of municipal data warehouses, dashboards, and predictive analytics, the latter mostly to identify risk of fraud. This transition to data-driven social policy, captured by the term “digital welfare state,” almost completely takes place out of political and social view, and escapes democratic decision making. In this article, I zoom in on The Netherlands and show in detail how sound data governance is lacking at three levels: data experiments and practices take place in a so-called “institutional void” without any clear democratic mandate; moreover, they are often based on disputable quality of data and analytic models; and they tend to transgress the recent EU General Data Protection Regulation (GDPR) about privacy and data protection. I also assess that key stakeholders in this data transition, that is the citizens whose data are used, are not actively informed let alone invited to participate. As a result, a practice of top-down monitoring, containment and control is evolving despite the desire of civil servants in this domain to do “good” with data. I explore several data and policy alternatives in the conclusion to contribute to a higher quality and more democratic usage of data in the digital welfare state.

Predictive analytics in health is a complex, transdisciplinary field requiring collaboration across diverse scientific and stakeholder groups. Pilot implementation of participatory research to foster team science in predictive analytics through a partnered-symposium and funding competition. In total, 85 stakeholders were engaged across diverse translational domains, with a significant increase in perceived importance of early inclusion of patients and communities in research. Participatory research approaches may be an effective model for engaging broad stakeholders in predictive analytics.

Introduction: With hospital occupancy rates frequently approaching 100%, even small variations in daily admission numbers can have a large impact. The ability to predict variance in emergency admission rates would provide administrators with a significant advantage in managing hospital daily bed requirements. There is a growing interest in patterns of hospital admissions, and many EDs utilize historical admission patterns to attempt to predict daily bed requirements. Previous studies have utilized patient demographics and past medical history to develop an admission likelihood model. We wished to examine the predictive strength of individual CEDIS presenting complaints (PC) on admission likelihood Methods: Using a database analysis of over 285,000 ED presentations (2013-2017), we calculated visit frequencies and admission rates by PC. Using a logistic regression analysis PCs were ordered from high to medium predictive strength. Results: Of 285,155 presentations, there were 38,090 hospital admissions, a rate of 13.36%. Based on the number of visit frequencies and admission rates, the PCs demonstrating high predictive strength were Direct Referral (effect=0.36, binomial CI: 0.28 to 0.44); Shortness of Breath (0.32: 0.26 to 0.41); General Weakness; Weakness/Query CVA; & Chest Pain Cardiac Features (each 0.30: 0.25 to 0.42); Altered level of consciousness (0.24: 0.16 to 0.31); and Confusion (0.18: 0.08 to 0.26). With our sample size, all remaining CEDIS PCs had low predictive value (the effect is <0.1), or were not predictive at all. Conclusion: We have demonstrated that, for our population, certain PCs are associated with an increased likelihood of admission and have quantified this effect using logistic regression analysis. Variance from the average daily admission rate may be predicted, in our population, by identifying these PCs at registration.We plan to develop a tool, based on this data and implemented at registration, to predict cumulative likely daily admission requirements as patients present over a 24hr period.

Introduction/Innovation Concept: Background: Competency based medical education (CBME) is a method of assessing resident performance through standardized tasks and milestones. The Royal College of Physicians and Surgeons of Canada has started phasing in CBME as the preferred training method, but no tool support exists to process this data. Approximately 400 data points are collected per resident per year at McMaster’s Division of Emergency Medicine. This is an unwieldy amount of data to analyze. Objective: Recognizing that collection and analysis of resident data is an important facet to postgraduate medical education, McMaster University began developing a program to provide predictive automated data analysis of resident performance. Methods: To achieve the stated objective, we adapted a design thinking methodology, which emphasizes the importance of human-centered design. By interviewing stakeholders, we collected user requirements and “pain points” that allowed us to build and evaluate multiple prototypes addressing their problems, such as the ability to process data into reports, real-time reporting, and predictive analytics. We solicited feedback from our stakeholders to iteratively refine the prototypes, ensuring that it was user intuitive and met user needs. Curriculum, Tool, or Material: We developed a software platform that collects, aggregates, reports, and has the possibility of analyzing resident data in real time. It also can present performance data via a real-time dashboard. Having automated the report generating process, administrative workload is reduced to a monitoring capacity. Quantitative data on resident performance has been analysed using artificial Neural Network to identify patterns in resident performance. It performs with a sensitivity of 81% and a specificity of 43%, and accurately predict which residents require remedial support 43% of the time. When built into a learning management system, this allows for the provision of additional support to residents-at-risk. Conclusion: Combining machine learning with resident assessment data has allowed us to build a promising predictive model to predict resident outcomes. This gives us the potential to decrease administrative workload and improve data quality by providing real-time performance dashboards and eliminating the redundancies of manual data processing. If scaled, this innovation might assist program directors in determining competency of residents and human resource planning for the healthcare systems at large.