Predictive Policing Fails: Cities Reject Biased Algorithms

In 2020, Chicago quietly decommissioned its Strategic Subject List—a predictive policing algorithm that had assigned risk scores to more than 400,000 residents. A staggering 56% of Black men in the city aged 20 to 29 found themselves on this list, flagged by an algorithm that claimed to predict who would commit crimes or become victims. The system increased arrest rates for targeted individuals but did nothing to reduce actual crime. One year later, Los Angeles pulled the plug on PredPol, its crime-forecasting software, after an internal audit found insufficient evidence it worked. By 2022, Pasco County, Florida, paid a $105,000 settlement and shut down its intelligence-led policing program after residents sued for constitutional violations. Across America, the promise of data-driven crime prevention is colliding with a harsh reality: predictive policing doesn't just fail to deliver results—it amplifies the very biases it was supposed to eliminate.

The Seductive Promise of Prediction

Predictive policing emerged in the early 2010s with a tantalizing proposition: what if police could prevent crime before it happened? Using machine learning algorithms to analyze historical crime data, demographic information, and geographic patterns, systems like PredPol, HunchLab, and the Strategic Subject List promised to identify crime hotspots and high-risk individuals with mathematical precision. Cities rushed to adopt these technologies, drawn by claims of reduced crime rates and more efficient resource allocation.

The pitch was irresistible. Santa Cruz, California, reported a nearly 20% decline in burglaries within six months of implementing PredPol. Chicago's police department claimed a 23% reduction in homicides during the first year of its predictive program. Dubai Police credited AI tools with a 25% drop in serious crime. The algorithms seemed to work, and they carried the aura of objectivity—cold, hard data replacing gut instincts and potentially biased human judgment.

But beneath the surface, a different story was unfolding. The RAND Corporation, in a comprehensive 2015 study, found no statistical evidence that crime was reduced when predictive policing was implemented. As one analysis in Significance magazine noted, "The algorithms were behaving exactly as expected—they reproduced the patterns in the data used to train them." The problem? Those patterns were riddled with decades of discriminatory policing practices.

The Bias Embedded in the Data

Predictive policing algorithms learn from historical crime data—arrest records, reported crimes, and police dispatch logs. This creates an insurmountable problem: if past policing was biased, the algorithm will be too. A 2018 study by Joy Buolamwini and Timnit Gebru found that commercial facial recognition technologies exhibited error rates of up to 35% when identifying darker-skinned women, compared to less than 1% for lighter-skinned men. When similar bias exists in the crime data that trains predictive models, the results are devastating.

Consider how this works in practice. Historical arrest data shows higher crime rates in low-income neighborhoods and communities of color—not necessarily because more crime occurs there, but because those areas have historically been more heavily policed. When an algorithm is trained on this data, it predicts higher crime risk in those same neighborhoods. Police departments then allocate more patrols to those areas, leading to more arrests, which feed back into the algorithm as confirmation of its accuracy. The Human Rights Data Analysis Group's simulation of PredPol revealed exactly this feedback loop: the system interpreted police car sightings as crime indicators, leading to increased police presence in Black neighborhoods, which generated more crime reports, which justified even more policing.

In Bogotá, Colombia, researchers built a predictive policing algorithm using victim report data and found it predicted 20% more high-crime locations than actually existed. The bias stemmed from differential reporting rates—Black residents were more likely to report crimes than white residents, skewing the data. In Oakland, California, a predictive policing algorithm trained on historical arrest data replicated the marginalization of African Americans that was embedded in decades of policing records. As IBM researchers noted, "If this data is used to train a current predictive policing algorithm, the decisions made by the PPA are likely to reflect and reinforce those past racial biases."

The Black Box Problem

Beyond biased data, predictive policing systems suffer from a fundamental transparency problem. Most algorithms are proprietary software developed by private companies that guard their methods as trade secrets. This creates what experts call a "black box"—a system whose internal logic is hidden from public scrutiny.

The Brookings Institution analyzed predictive policing implementations across multiple U.S. cities and found that "in many cities, local governments had no public documentation on how predictive policing software functioned, what data was used, or how outcomes were evaluated." Police departments couldn't explain how the algorithms made decisions. Community members had no way to challenge predictions. Even judges and defense attorneys found it nearly impossible to audit the systems that were being used to justify stops, searches, and surveillance.

This opacity has profound legal implications. The Michigan Law Review published a detailed analysis arguing that trade-secret protections applied to law enforcement algorithms create constitutional blindness, violating due process safeguards. When Chicago's Strategic Subject List flagged someone as high-risk, that person had no way to know why, no ability to contest the designation, and no mechanism to appeal. The algorithm became an invisible witness that couldn't be cross-examined.

In San Diego, a vendor misconfiguration led to 12,960 unauthorized data queries by outside agencies over just two weeks—a massive breach that went undetected because audit mechanisms were inadequate. The incident revealed how easily surveillance technology can be misused when transparency and accountability are absent.

The Moment of Reckoning

The turning point came when the promises of predictive policing crashed into measurable reality. In October 2023, The Markup and Wired conducted a joint investigation analyzing 23,631 predictions generated by Geolitica (formerly PredPol) for the Plainfield, New Jersey, Police Department over ten months. The success rate? Less than 0.5%. Fewer than 100 predictions matched actual crimes. Captain David Guarino of the Plainfield Police was blunt: "Why did we get PredPol? I guess we wanted to be more effective when it came to reducing crime...I don't believe we really used it that often, if at all. That's why we ended up getting rid of it."

Other studies confirmed the dismal performance. Geolitica showed just 0.6% accuracy when predicting aggravated assaults and a mere 0.1% for burglary in some areas. Meanwhile, the costs were substantial—Mountain View, California, spent more than $60,000 between 2013 and 2018; Hagerstown, Maryland, paid $15,000 annually until 2018. Cities were spending tens of thousands of dollars on systems that barely worked.

Worse, these systems were causing real harm. In Pasco County, Florida, the sheriff's intelligence-led policing program compiled lists of people deemed likely to commit crimes, then sent deputies to their homes for repeated, unannounced visits. More than 1,000 residents—including minors—were cited for trivial violations like missing mailbox numbers and overgrown grass. Four residents sued in 2021, and the following year, the county reached a settlement admitting it had violated constitutional rights to privacy and equal treatment. The program was discontinued.

Los Angeles faced similar scrutiny. The city's Operation LASER (Los Angeles Strategic Extraction and Restoration) program was shut down in 2019 after the inspector general found "inconsistencies when labeling people" and raised concerns about racial bias. An internal audit in March 2021 concluded there was insufficient evidence that PredPol reduced crime, prompting the LAPD to end its contracts. As one law professor observed, "After ten years, with enough public pressure and community concern, it was an easy decision to just pull the plug."

The Role of Community Advocacy

Behind every policy reversal stood years of advocacy by civil rights organizations, community activists, and researchers who refused to accept the narrative of algorithmic objectivity. The NAACP, ACLU, and Electronic Frontier Foundation led campaigns calling for stringent oversight, bias audits, and outright bans on predictive policing. In a letter to the Department of Justice, U.S. Senators stated bluntly: "Mounting evidence indicates that predictive policing technologies do not reduce crime...Instead, they worsen the unequal treatment of Americans of color by law enforcement."

In Santa Cruz, California, advocacy bore fruit when the city council enacted the first municipal ban on predictive policing and facial recognition software on June 23, 2020. Police Chief Andy Mills supported the move, saying, "Predictive policing has been shown over time to put officers in conflict with communities rather than working with the communities." Mayor Justin Cummings added, "If policing itself is biased, then the data that's informing those models will be biased." Remarkably, even PredPol's CEO, Brian MacDonald, praised the ban: "Given the racial inequalities pervasive throughout American history and society, we as a company support this language. In fact, we would even go so far as to recommend that this standard be applied to all technologies adopted by the city of Santa Cruz."

Oakland followed with its own restrictions. Brian Hofer, chair of Oakland's privacy commission, captured the frustration many felt: "We fall for the marketing hype, go release this stuff out into the wild without understanding it, and then never really demand a cost-benefit analysis."

What Went Wrong: A Systems Analysis

Predictive policing failed for reasons that go deeper than technical glitches. At its core, the approach treats complex human behavior as predictable patterns that can be extracted from historical data. But crime isn't like an earthquake—the analogy PredPol's creators used when adapting seismology algorithms to policing. Crime emerges from social, economic, and political conditions that algorithms can't capture.

First, data quality issues plague every system. Historical crime data reflects policing priorities, not objective crime rates. If police focus on low-level drug offenses in certain neighborhoods, those neighborhoods will appear to have higher crime rates, even if more serious crimes occur elsewhere. Researchers building predictive models in Bogotá discovered this when their algorithm over-predicted crime in districts with high reporting rates, not high actual crime.

Second, proxy variables sneak bias into models even when developers try to exclude race. As scholars note, "It is extremely difficult to eliminate all proxies for such variables due to correlations between them and much of the other data available to law enforcement." Zip codes, prior arrests, and even the time of day can serve as stand-ins for race, allowing discrimination to persist under a veneer of neutrality.

Third, feedback loops amplify initial biases. When a model predicts crime in Area A, police patrol Area A more heavily, leading to more arrests in Area A, which the algorithm interprets as confirmation that Area A is high-crime. This creates a self-fulfilling prophecy that's difficult to break. A New York University study examining 13 U.S. jurisdictions found that predictive policing systems "exacerbated existing discriminatory law enforcement practices" precisely because of these loops.

Fourth, the proprietary nature of algorithms prevents external validation. Companies like Geolitica, Palantir, and others shield their methods as trade secrets, making independent audits nearly impossible. Smithsonian magazine noted in 2018 that "no independent published research had ever confirmed PredPol's claims of its software's accuracy." Without transparency, there's no way to identify and correct errors—or even to know when systems are failing.

The Path Forward: Accountability Over Algorithms

The collapse of predictive policing has sparked a search for alternatives rooted in transparency and community engagement. San Jose, California, has emerged as a leader with its adoption of AI principles requiring that any AI tool used by city government be effective, transparent to the public, and equitable in its effects. Departments must conduct risk assessments before deploying AI systems, and the public has access to information about how these tools work.

This represents a fundamental shift from black-box algorithms to democratic accountability. As researchers note, "Transparency and accountability are both tools to promote fair algorithmic decisions by providing the foundations for obtaining recourse to meaningful explanation, correction, or ways to ascertain faults that could bring about compensatory processes."

Some jurisdictions are exploring alternative analytical tools that avoid the pitfalls of traditional predictive policing. ResourceRouter, for example, analyzes crime incidents alongside non-crime data like weather patterns and public events to identify high-risk areas each shift. It excludes misdemeanor and nuisance crimes that reflect enforcement bias, focuses on documented serious crimes, and provides patrol metering with visible 15-minute timers so command staff can track officer deployment. These features address transparency and bias concerns while still using data to inform decisions.

Others are turning to community-based crime prevention strategies that don't rely on algorithms at all. Violence interrupter programs, which employ community members to mediate conflicts before they escalate, have shown promising results. Investment in social services—mental health care, substance abuse treatment, youth programs—addresses root causes rather than just reacting to symptoms. These approaches build trust rather than erode it.

At the federal level, the White House Office of Management and Budget issued a policy in March 2023 requiring independent testing, cost-benefit analysis, and public feedback for AI systems that impact rights, including predictive policing. The Department of Justice has called for predictive policing implementations to prioritize community partnership, technical oversight, and operational requirements that mitigate bias.

Legal and Ethical Frameworks Emerging

The backlash against predictive policing is also driving legal innovation. Courts are beginning to consider disparate-outcome standards similar to those in employment discrimination law. Under Title VII, plaintiffs can prove discrimination by showing that a policy has a disproportionate impact on a protected group, without needing to prove discriminatory intent. Applying this standard to algorithms could allow bias claims to succeed based on statistical disparities alone.

Scholars have proposed a "missing algorithm" remedy for Brady violations. The landmark Brady v. Maryland case requires prosecutors to disclose exculpatory evidence to defendants. When police use algorithmic tools to identify suspects or determine patrol routes, and those algorithms are hidden as trade secrets, defendants may be denied access to evidence that could exonerate them. The proposed remedy would require disclosure of algorithmic methods when they're used in criminal investigations.

Some researchers advocate viewing police technology deployments as experiments on human subjects, subject to the same ethical oversight as medical research. Just as institutional review boards approve clinical trials to protect participants, a similar framework could govern the introduction of surveillance and predictive technologies, ensuring informed consent, minimizing harm, and allowing for independent evaluation.

What This Means for the Future of Policing

The abandonment of predictive policing by major cities marks more than the failure of specific technologies—it signals a broader reckoning with the limits of algorithmic solutions to social problems. Crime is a social phenomenon, deeply tied to inequality, opportunity, mental health, and community cohesion. No amount of data analysis can substitute for addressing these underlying conditions.

Yet technology will continue to play a role in policing. The question is how. Will departments embrace transparency, community input, and rigorous bias testing? Or will they chase the next black-box solution that promises easy answers?

The University of Chicago developed a predictive model in 2022 that achieved 90% accuracy a week in advance using a tile-based system. The researchers emphasized that the system was "less prone to bias than older systems because it uses different data." Dubai Police's AI tools reportedly reduced serious crime by 25%. These examples suggest that well-designed, carefully monitored systems might offer value—if, and only if, they're developed with accountability from the start.

The stakes are enormous. Brandon del Pozo, a policing expert, put it this way: "When we're looking at how to bring AI into criminal justice, we have to foreground those values, and all our theories should be evaluated accordingly." The values he's referring to are fairness, transparency, and community trust—principles that were casualties of the first wave of predictive policing.

Lessons for Other Cities and Technologies

Cities still using or considering predictive policing should heed several clear lessons:

First, demand evidence. Before deploying any algorithmic system, require independent, peer-reviewed studies demonstrating effectiveness. Vendor claims are not sufficient. The failures of PredPol, Geolitica, and Chicago's Strategic Subject List all occurred because cities accepted marketing promises without rigorous evaluation.

Second, ensure transparency. Proprietary algorithms that can't be audited should be disqualified from consideration. If a vendor won't disclose how their system works, the public can't trust it, and courts can't assess its fairness.

Third, conduct bias testing. Before and after deployment, analyze whether the system produces disparate outcomes for different racial, ethnic, or socioeconomic groups. Regular audits by independent parties should be mandatory, not optional.

Fourth, involve the community. Those most affected by policing decisions should have a voice in technology adoption. Public forums, advisory boards, and community oversight mechanisms can prevent the kind of erosion of trust that occurred in Pasco County and elsewhere.

Fifth, consider alternatives. Technology isn't always the answer. Community-based violence prevention, restorative justice programs, and investments in social services may be more effective and less harmful than any algorithm.

Finally, build accountability mechanisms. When systems fail or cause harm, there must be pathways for recourse—complaints, appeals, and if necessary, discontinuation. Technology should serve people, not the other way around.

The Broader Implications for AI in Society

The predictive policing story offers lessons that extend far beyond law enforcement. As artificial intelligence spreads into hiring, lending, healthcare, and education, the same dynamics—biased data, opaque algorithms, feedback loops, and lack of accountability—threaten to replicate and amplify existing inequalities across society.

Algorithmic decision-making often carries an aura of objectivity that human judgment lacks. But as the failures of predictive policing demonstrate, algorithms are only as good as the data they're trained on and the values of the people who design them. When IBM researchers tested a predictive policing algorithm, they found it "reflected and reinforced past racial biases" because the training data did. The lesson is universal: bias in, bias out.

The push for algorithmic accountability frameworks is gaining momentum. The AI Now Institute has proposed governance options including awareness-raising, accountability in public sector use, regulatory oversight with legal liability, and global coordination for shared standards. These proposals recognize that transparency alone isn't enough—there must be mechanisms to identify harm, provide remedies, and impose consequences when systems fail.

Meanwhile, the concentration of algorithmic power in the hands of a few tech companies raises concerns about corporate influence over public policy. In August 2023, SoundThinking announced it was absorbing Geolitica's engineering team, patents, and customers, planning to cease Geolitica operations by year's end. This consolidation could reduce competition and innovation, making it harder for cities to find alternatives or negotiate better terms.

A Choice About the Kind of Society We Want

Ultimately, the debate over predictive policing is a debate about values. Do we accept the claim that mathematical models can identify future criminals, or do we insist that people have the capacity to change and shouldn't be judged by an algorithm's assessment of their past? Do we prioritize efficiency and resource optimization, or fairness and civil liberties? Do we trust opaque systems controlled by private companies, or demand transparency and public accountability?

Chicago, Los Angeles, New York, Santa Cruz, and other cities have made their choice: they've rejected the false promise of algorithmic objectivity in favor of approaches that center community trust and human dignity. As Mayor Justin Cummings of Santa Cruz said when announcing the ban, "Understanding how predictive policing and facial recognition can be disproportionately biased against people of color, we officially banned the use of these technologies in the city of Santa Cruz."

The path forward requires humility about what technology can and cannot do. It requires acknowledging that historical data reflects historical injustice, and that reproducing patterns from the past will only perpetuate inequality. It requires transparency, so that those affected by algorithmic decisions can understand and challenge them. And it requires a commitment to equity, ensuring that the benefits and burdens of new technologies are distributed fairly.

The collapse of predictive policing is not a failure of innovation—it's a victory for accountability. The question now is whether other sectors deploying AI will learn these lessons before repeating the same mistakes, or whether each domain will have to discover the limits of algorithmic justice the hard way. For those who believe technology should serve humanity rather than the other way around, the answer is clear: we must demand that every algorithm, in every application, meets the standards of transparency, fairness, and accountability that predictive policing failed to achieve. Only then can we build a future where technology truly enhances justice rather than undermining it.

Latest from Each Category

Space

Fusion Rockets Could Reach 10% Light Speed: The Breakthrough

Recent breakthroughs in fusion technology—including 351,000-gauss magnetic fields, AI-driven plasma diagnostics, and net energy gain at the National Ignition Facility—are transforming fusion propulsion from science fiction to engineering frontier. Scientists now have a realistic pathway to accelerate spacecraft to 10% of light speed, enabling a 43-year journey to Alpha Centauri. While challenges remain in miniaturization, neutron management, and sustained operation, the physics barriers have ...

Health

Epigenetic Clocks Predict Disease 30 Years Early

Epigenetic clocks measure DNA methylation patterns to calculate biological age, which predicts disease risk up to 30 years before symptoms appear. Landmark studies show that accelerated epigenetic aging forecasts cardiovascular disease, diabetes, and neurodegeneration with remarkable accuracy. Lifestyle interventions—Mediterranean diet, structured exercise, quality sleep, stress management—can measurably reverse biological aging, reducing epigenetic age by 1-2 years within months. Commercial ...

Environment

Digital Pollution Tax: Can It Save Data Centers?

Data centers consumed 415 terawatt-hours of electricity in 2024 and will nearly double that by 2030, driven by AI's insatiable energy appetite. Despite tech giants' renewable pledges, actual emissions are up to 662% higher than reported due to accounting loopholes. A digital pollution tax—similar to Europe's carbon border tariff—could finally force the industry to invest in efficiency technologies like liquid cooling, waste heat recovery, and time-matched renewable power, transforming volunta...

Humans

Why Your Brain Sees Gods and Ghosts in Random Events

Humans are hardwired to see invisible agents—gods, ghosts, conspiracies—thanks to the Hyperactive Agency Detection Device (HADD), an evolutionary survival mechanism that favored false alarms over fatal misses. This cognitive bias, rooted in brain regions like the temporoparietal junction and medial prefrontal cortex, generates religious beliefs, animistic worldviews, and conspiracy theories across all cultures. Understanding HADD doesn't eliminate belief, but it helps us recognize when our pa...

Nature

Bombardier Beetle Chemical Defense: Nature's Micro Engine

The bombardier beetle has perfected a chemical defense system that human engineers are still trying to replicate: a two-chamber micro-combustion engine that mixes hydroquinone and hydrogen peroxide to create explosive 100°C sprays at up to 500 pulses per second, aimed with 270-degree precision. This tiny insect's biochemical marvel is inspiring revolutionary technologies in aerospace propulsion, pharmaceutical delivery, and fire suppression. By 2030, beetle-inspired systems could position sat...

Society

Care Worker Crisis: Low Pay & Burnout Threaten Healthcare

The U.S. faces a catastrophic care worker shortage driven by poverty-level wages, overwhelming burnout, and systemic undervaluation. With 99% of nursing homes hiring and 9.7 million openings projected by 2034, the crisis threatens patient safety, family stability, and economic productivity. Evidence-based solutions—wage reforms, streamlined training, technology integration, and policy enforcement—exist and work, but require sustained political will and cultural recognition that caregiving is ...

Computers

AI Training Data Copyright Crisis: Lawsuits & Solutions

Every major AI model was trained on copyrighted text scraped without permission, triggering billion-dollar lawsuits and forcing a reckoning between innovation and creator rights. The future depends on finding balance between transformative AI development and fair compensation for the people whose work fuels it.