Racial bias in predictive policing algorithms has been the focus of a number of recent news articles, statements of concern by several national organizations (e.g., the ACLU and NAACP), and simulat...
Does Predictive Policing Lead to Biased Arrests? Results From a Randomized Controlled Trial
Racial bias in predictive policing algorithms has been the focus of a number of recent news articles, statements of concern by several national organizations (e.g., the ACLU and NAACP), and simulat...
Predictive Policing Explained
Attempts to forecast crime with algorithmic techniques could reinforce existing racial biases in the criminal justice system.
Artificial Intelligence, Predictive Policing, and Risk Assessment for Law Enforcement
There are widespread concerns about the use of artificial intelligence in law enforcement. Predictive policing and risk assessment are salient examples. Worries include the accuracy of forecasts that guide both activities, the prospect of bias, and an apparent lack of operational transparency. Nearly breathless media coverage of artificial intelligence helps shape the narrative. In this review, we address these issues by first unpacking depictions of artificial intelligence. Its use in predictive policing to forecast crimes in time and space is largely an exercise in spatial statistics that in principle can make policing more effective and more surgical. Its use in criminal justice risk assessment to forecast who will commit crimes is largely an exercise in adaptive, nonparametric regression. It can in principle allow law enforcement agencies to better provide for public safety with the least restrictive means necessary, which can mean far less use of incarceration. None of this is mysterious. Nevertheless, concerns about accuracy, fairness, and transparency are real, and there are tradeoffs between them for which there can be no technical fix. You can't have it all. Solutions will be found through political and legislative processes achieving an acceptable balance between competing priorities.
Artificial Intelligence, Predictive Policing, and Risk Assessment for Law Enforcement
There are widespread concerns about the use of artificial intelligence in law enforcement. Predictive policing and risk assessment are salient examples. Worries include the accuracy of forecasts that guide both activities, the prospect of bias, and an apparent lack of operational transparency. Nearly breathless media coverage of artificial intelligence helps shape the narrative. In this review, we address these issues by first unpacking depictions of artificial intelligence. Its use in predictive policing to forecast crimes in time and space is largely an exercise in spatial statistics that in principle can make policing more effective and more surgical. Its use in criminal justice risk assessment to forecast who will commit crimes is largely an exercise in adaptive, nonparametric regression. It can in principle allow law enforcement agencies to better provide for public safety with the least restrictive means necessary, which can mean far less use of incarceration. None of this is mysterious. Nevertheless, concerns about accuracy, fairness, and transparency are real, and there are tradeoffs between them for which there can be no technical fix. You can't have it all. Solutions will be found through political and legislative processes achieving an acceptable balance between competing priorities.
Predictive policing AI is on the rise − making it accountable to the public could curb its harmful effects
AI that anticipates where crimes are likely to occur and who might commit them has a troubling track record. Democratic accountability could shine a light on the technology and how it’s used.
Predictive policing AI is on the rise − making it accountable to the public could curb its harmful effects
AI that anticipates where crimes are likely to occur and who might commit them has a troubling track record. Democratic accountability could shine a light on the technology and how it’s used.
Predictive Policing and the Politics of Patterns
Abstract. Patterns are the epistemological core of predictive policing. With the move towards digital prediction tools, the authority of the pattern is rea
Predictive Policing and the Politics of Patterns
Abstract Patterns are the epistemological core of predictive policing. With the move towards digital prediction tools, the authority of the pattern is rearticulated and reinforced in police work. Based on empirical research about predictive policing software and practices, this article puts the authority of patterns into perspective. Introducing four ideal-typical styles of pattern identification, we illustrate that patterns are not based on a singular logic, but on varying rationalities that give form to and formalize different understandings about crime. Yet, patterns render such different modes of reasoning about crime, and the way in which they feed back into policing cultures, opaque. Ultimately, this invites a stronger reflection about the political nature of patterns.
[122] Arresting Flexibility: A QJE field experiment on police behavior with about 40 outcome variables
A forthcoming paper in the Quarterly Journal of Economics (QJE), “A Cognitive View of Policing” (htm), reports results from a field experiment showing that teaching police officers to &…
How AI can lead to false arrests and wrongful convictions
Danger arises when law enforcement believes that AI models are retrieving certainties rather than generating likelihoods.
How AI can lead to false arrests and wrongful convictions
Danger arises when law enforcement believes that AI models are retrieving certainties rather than generating likelihoods.
How the LAPD and Palantir Use Data to Justify Racist Policing
In a new book, a sociologist who spent months embedded with the LAPD details how data-driven policing techwashes bias.
A Cognitive View of Policing
Abstract What causes adverse policing outcomes, such as excessive uses of force and unnecessary arrests? Prevailing explanations focus on problematic officers or deficient regulations and oversight. We introduce an overlooked perspective. We suggest that the cognitive demands inherent in policing can undermine officer decision making. Unless officers are prepared for these demands, they may jump to conclusions too quickly without fully considering alternative ways of seeing a situation. This can lead to adverse policing outcomes. To test this perspective, we created a new training that teaches officers to consider different ways of interpreting the situations they encounter. We evaluated this training using a randomized controlled trial with 2,070 officers from the Chicago Police Department. In a series of lab assessments, we find that treated officers were significantly more likely to consider a wider range of evidence and develop more explanations for subjects’ actions. Critically, we also find that training affected officer performance in the field, leading to reductions in uses of force, discretionary arrests, and arrests of Black civilians. Meanwhile, officer activity levels remained unchanged, and trained officers were less likely to be injured on duty. Our results highlight the value of considering the cognitive aspects of policing and demonstrate the power of using behaviorally informed approaches to improve officer decision making and policing outcomes.
Characterizing Fairness Over the Set of Good Models Under Selective Labels
Algorithmic risk assessments are used to inform decisions in a wide variety of high-stakes settings. Often multiple predictive models deliver similar overall performance but differ markedly in their predictions for individual cases, an empirical phenomenon known as the “Rashomon Effect.” These models may have different properties over various groups, and therefore have different predictive fairness properties. We develop a framework for characterizing predictive fairness properties over the set of models that deliver similar overall performance, or “the set of good models.” Our framework addresses the empirically relevant challenge of selectively labelled data in the setting where the selection decision and outcome are unconfounded given the observed data features. Our framework can be used to 1) audit for predictive bias; or 2) replace an existing model with one that has better fairness properties. We illustrate these use cases on a recidivism prediction task and a real-world credit-scoring task.
Certifying and Removing Disparate Impact
What does it mean for an algorithm to be biased? In U.S. law, unintentional bias is encoded via disparate impact, which occurs when a selection process has widely different outcomes for different groups, even as it appears to be neutral. This legal determination hinges on a definition of a protected class (ethnicity, gender) and an explicit description of the process.When computers are involved, determining disparate impact (and hence bias) is harder. It might not be possible to disclose the process. In addition, even if the process is open, it might be hard to elucidate in a legal setting how the algorithm makes its decisions. Instead of requiring access to the process, we propose making inferences based on the data it uses.We present four contributions. First, we link disparate impact to a measure of classification accuracy that while known, has received relatively little attention. Second, we propose a test for disparate impact based on how well the protected class can be predicted from the other attributes. Third, we describe methods by which data might be made unbiased. Finally, we present empirical evidence supporting the effectiveness of our test for disparate impact and our approach for both masking bias and preserving relevant information in the data. Interestingly, our approach resembles some actual selection practices that have recently received legal scrutiny.
