Actionable predictions: How designers of algorithmic systems calibrate criminal futures
Hopes and fears about algorithmic predictions are often rooted in the assumption that they represent a particularly actionable form of knowledge. Algorithmic predictions, the story goes, turn historical data into anticipatory actions instantly and on a large scale. Recent empirical evidence, however...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
SAGE Publishing
2025-06-01
|
| Series: | Big Data & Society |
| Online Access: | https://doi.org/10.1177/20539517251340636 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Hopes and fears about algorithmic predictions are often rooted in the assumption that they represent a particularly actionable form of knowledge. Algorithmic predictions, the story goes, turn historical data into anticipatory actions instantly and on a large scale. Recent empirical evidence, however, casts doubt on whether algorithmic predictions are best understood as being inherently actionable. In this study, we draw on adjacent debates about actionable knowledge and reconceptualize the actionability of predictions as an active and deliberate construction accomplished by the designers of algorithmic systems. We demonstrate the value of this new perspective using material from an ethnographic research project on predictive policing in Germany. Over a 12-month period, we followed the work of designers in a police research unit responsible for developing an algorithmic system that generates crime predictions. We found that an important part of the designers’ work is calibrating the predictions. When engaging in calibration work, the designers adjust the form of the statistically calculated predictions—their volume, time, and space—to reflect their assumptions of what is most actionable knowledge for frontline police officers. Our study highlights a type of work on algorithmic systems that has received little attention, but which can have a significant impact on how the intended effects of algorithmic predictions are translated into their users’ actions. |
|---|---|
| ISSN: | 2053-9517 |