Building the case for restricted use of predictive policing tools in India
Bibliographic record
Abstract
The use of predictive policing by law enforcement agencies has lately proliferated across several states in India. Predictive policing uses machine learning models to analyse substantial amounts of crime data to map and predict crimes, offenders, identities of the perpetrators and victims of crime. The aim is to enable the efficient allocation of limited resources at the disposal of law enforcement agencies for crime prevention. However, evidence suggests that predictive policing, in its current form, suffers from serious limitations. Inferior quality datasets are being used to train algorithms and citizens are unable to contest inaccurate algorithmic outcomes that could lead to their preventive detentions, often to the detriment of criminal justice norms and constitutional fundamental rights of citizens. More importantly, in the absence of oversight mechanisms restricting its use, the use of predictive policing has ended up reinforcing and amplifying police biases in law enforcement. This paper focuses, in particular, on the many ways in which predictive policing increases the risk of preventive detentions under section 151 of the Code of Criminal Procedure (CrPC), 1973 and proposes recommendations to preclude possibilities of unlawful preventive detentions using predictive policing tools. In part 1, the paper reviews the practice of preventive detention of potential offenders by the police in India—under section 151 of the CrPC. In part 2, I build the case for restricted use of predictive policing tools to only predict the location, type and time of the crime and not potential offenders or victims. In the last part, I recommend substantive measures and procedural safeguards to secure the reliable and accountable use of place-based predictive policing tools in India.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.012 | 0.041 |
| Meta-epidemiology (narrow) | 0.000 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.003 | 0.003 |
| Science and technology studies | 0.004 | 0.011 |
| Scholarly communication | 0.010 | 0.009 |
| Open science | 0.004 | 0.008 |
| Research integrity | 0.004 | 0.012 |
| Insufficient payload (model declined to judge) | 0.005 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".