Pre-Emptive Justice: The Ethics of Using AI to Predict and Prevent Crime

Key Findings

Predictive Policing Data

Predictive policing cannot rely on past arrest data in places with strong privacy laws because legal limits on data retention and transparency block the feedback loops needed to train the systems.

Predictive policing systems rely on large amounts of past arrest data to find patterns over time. These systems assume that data collection will continue without limits. But in places like the European Union, strict privacy laws limit how long personal data can be kept. Laws such as the GDPR also give people the right to understand automated decisions. This means police cannot always use historical data to train algorithms. Without continuous access to data, the algorithms cannot learn from past patterns. When data use is restricted, the feedback loop from past arrests to future predictions breaks down. Legal rules that protect privacy and fairness reduce the risk that biased data will repeat old policing patterns. In countries that enforce these rights, the data needed for predictive policing is not always available. This weakens claims that biased data will always shape where police focus their efforts.

Predictive Policing Bias

Predictive policing does not always increase surveillance because legal rules can block biased data use and change how the systems operate.

Predictive policing algorithms use old crime data. These data reflect past police activity, not actual crime levels. In cities with deep social divides, police historically watched some neighborhoods more closely. This means the data count more arrests in those areas, not more crime. The algorithms learn from this flawed data. They target the same communities again. This increases police presence where people already face more scrutiny. The system repeats the past. But this cycle only continues if no rules block it. In places with strong privacy laws or court rulings, data use is limited. The European Union restricts how data can be reused. U.S. courts have upheld privacy rights. Legal challenges by groups like the ACLU have forced changes. Many predictive systems now face strict limits. Without these legal checks, the tools would deepen bias. With them, the outcome shifts. So the claim that these tools always increase police reach is not fully true. It ignores how laws can change the result. Legal rules can stop the cycle. They alter how the systems work in practice. The effect of predictive policing depends on oversight.

Predictive Policing Feedback Loop

Predictive policing reinforces surveillance in over-policed areas because it relies on historical crime data, creating a feedback loop that diminishes when legal oversight demands transparency and fairness.

Predictive policing systems guide police resources based on past crime data. These systems focus more surveillance on communities already heavily policed. More policing leads to more reported crimes in those areas. This creates a cycle where historical data shapes future enforcement. The cycle persists when no one reviews the system critically. It weakens when laws limit data use or require transparency. Courts in some regions have challenged unchecked automated decisions. This shifts the focus from pure prediction to fairness and accountability. When systems must provide auditable explanations or face legal limits, the logic shifts. The cycle breaks when rules demand justification over prediction. Oversight changes how these tools operate in practice.

Crime Prediction Tools

Crime prediction tools spread because tight budgets favor measurable tech fixes over social programs, making cost savings the main force behind their use.

Courts and police now use algorithms to assess crime risk. These tools spread because governments choose them over social programs. Funding for public safety often skips housing or job support. Instead it pays for tech solutions. This saves money in the short term. Algorithms turn complex lives into simple risk scores. These scores fit neatly into budget plans. But they ignore root causes of crime. Things like job loss or unstable housing. Governments track data from these tools more easily than social needs. Austerity makes this worse. Poorer cities adopt predictive systems faster. International reports confirm this trend. The real driver is not policing data or law changes. It is the need to show savings. Algorithms seem efficient. They produce numbers that justify tight budgets. So cities use them more. This reduces focus on prevention. The tools shape policy more than crime itself. Data feedback and legal checks matter less than cost savings.

Predictive Policing Feedback Loop

Predictive policing increases police presence in over-policed neighborhoods because the algorithms learn from biased data and create a cycle of surveillance that reinforces itself.

Predictive policing systems use data like 911 calls and housing problems to mark neighborhoods as high-risk. These systems often ignore actual crime reports. Instead they rely on statistics that reflect past policing patterns. In New York City the Domain Awareness System was built with Microsoft and the NYPD. It uses these flawed data to guide police activity. Areas already heavily watched get more police attention. This happens because algorithms learn from old arrest data. Those data come from over-policed communities. The models then predict more crime there. This leads to more patrols. A feedback loop forms. More patrols lead to more arrests. More arrests feed the algorithm. This cycle makes the model seem accurate. But it only reinforces bias. The system treats suspicion as normal for certain areas. It does not treat crime as something uncertain. This changes how cities govern. Surveillance becomes routine. It feels normal. But it targets the same neighborhoods. Civil rights oversight is weakest there. Most big U.S. cities use such systems. They have not lowered violent crime. But they have increased police presence. This creates a form of justice that acts before crimes occur. It assumes some people should always be watched.

Predictive Policing Gaps

Predictive policing fails to produce reliable crime forecasts because most cities lack the integrated, shared data systems needed to continuously update the models.

Predictive policing tools rely on data to guide police patrols. These tools assume that arrest records feed into centralized systems. The systems then update models to predict future crime. But most U.S. cities lack integrated data networks. Police departments use separate databases that do not share information well. Data sharing often happens in informal, temporary ways. This fragmentation breaks the feedback loop the models need. Arrest data does not flow reliably into the prediction systems. Without constant updates from real-world enforcement, models cannot improve. Most cities cannot sustain the data flow needed. As a result, the models fail to produce accurate crime forecasts over time.

Policing By Prediction

Predictive policing repeats old biases because it uses biased arrest data to decide where police go, making crime seem higher in already over-policed areas.

Predictive algorithms are used in law enforcement to decide where patrols should go. These systems rely on old arrest data. That data reflects past bias, not real crime patterns. Police are sent more often to neighborhoods already over-policed. More patrols lead to more arrests in those areas. This makes crime appear more common there. The data then shows more crime, justifying more patrols. It becomes a cycle. The algorithm treats this loop as evidence of risk. This feedback locks in historical bias. The result is not better predictions. It is more police attention where there was already too much. This happens even if actual crime is no higher. Systems like CompPol in New York show this pattern. Scholars from the Brennan Center have documented the effect. The outcome looks fair because it uses data. But it repeats old patterns of unequal policing.

Deeper Analysis

What happens when predictive policing systems are trained on non-personal or aggregated data that falls outside the scope of GDPR but still influences law enforcement decisions?

Predictive Policing Loophole

Predictive policing systems produce biased outcomes by using data that appear neutral but stand in for historically policed communities, thus bypassing privacy laws while maintaining discriminatory patterns.

In democratic countries with strong privacy laws, predictive policing systems avoid strict data rules. They do not use personal data. Instead, they rely on broad statistics like 911 call rates or anonymized movement patterns. These data types are not covered by GDPR in the same way personal data are. The systems use them to map crime risk. Because the data appear neutral, they bypass consent rules and the right to explanation. This is allowed under European guidelines. The key is reframing sensitive information as general context. The data reflect past policing patterns. Even without personal identifiers, they mirror where police have focused before. Using this data keeps predictive power. It also follows privacy rules on paper. But it can still target the same neighborhoods. The reason is that the data stand in for historically watched communities. So outcomes remain biased. The system stays legal while repeating old patterns. The mechanism is the reclassification of data, not better fairness. Predictive policing thus avoids regulation while reproducing bias. Formal compliance does not mean equal treatment.

Explore further:

If the predictive power of non-personal data proxies depends on their historical correlation with policed populations, would predictive policing lose effectiveness in a society that had no history of discriminatory policing?

What happens to the effectiveness of predictive policing algorithms when communities systematically alter their reporting behaviors to game the system or reduce police exposure?

Police Prediction Failure

Predictive policing fails when communities reduce crime reporting to avoid police, because the system relies on report data that becomes too scarce to predict accurately.

Some cities use computer programs to predict where police patrols are needed. These programs rely on crime reports to make predictions. When people in a community report fewer crimes, the system gets less data. This often happens because residents want to avoid more police presence. Without enough reliable reports, the prediction software cannot work well. The problem grows when there are strong legal limits on police surveillance. In these cases, community members may stop reporting crimes altogether. This reduces the accuracy of the predictions. The algorithms do not fail on their own. The data they depend on becomes flawed. The more legal controls exist around data use, the more reporting drops. When reporting drops, predictions become less effective. This effect is stronger in cities with strong oversight laws.

Explore further:

What happens to predictive policing's effectiveness when communities selectively underreport crimes not uniformly but only in neighborhoods undergoing rapid gentrification, altering both data patterns and political accountability simultaneously?

What happens to predictive policing systems when the communities they target reorganize their behavior to exploit the algorithm’s reliance on historical data patterns?

Predictive Policing Failure

Predictive policing fails when community behavior changes in ways that break the link between past data and future risk, undermining the model's core assumption of stable patterns.

Predictive policing systems use past crime data to decide where to focus police efforts. These systems assume that people will continue behaving as they have in the past. But when communities change their behavior to avoid detection, the data stops being accurate. For example, if people shift illegal activity to places or forms that the system does not track, predictions fail. This happens because the system relies on patterns that no longer reflect reality. It is not broken by poor design but by smart responses from the people it targets. Communities that learn how the system works can adapt in ways that undermine its logic. This kind of evasion is similar to how people beat credit scoring or border algorithms. The system’s failure is hidden until official reviews expose it. Rules like those in the U.S. and EU require fairness and transparency. But these rules do not stop the failure. They only help us see it after it happens. The real collapse occurs when the system can no longer use the past to predict the future. That moment comes when people stop acting like they did before. The system then becomes useless not because of legal rulings but because its core assumption fails.

Blame Avoidance Tools

Predictive policing persists because it creates documented risk assessments that shield officials from blame, making it valuable to risk-averse bureaucracies regardless of its accuracy or data limitations.

Police departments keep using predictive policing systems because these tools help protect officials from blame. They do not persist mainly because they save money or bypass data laws. Democratic governments face high pressure when crimes happen unexpectedly. A single serious incident can damage reputations and careers. This creates strong incentives to avoid risk at all costs. Agencies adopt algorithms not for their accuracy but for their ability to create documented risk assessments. These assessments turn uncertain futures into auditable forecasts. When crime occurs, leaders can point to the forecast as proof they took action. Even if communities change behavior or data is flawed, the system still serves its core purpose. The record shows officials followed procedure. This makes the technology hard to displace. Resistance to change comes less from technical flaws than from the system’s role in deflecting accountability. As long as avoiding blame remains a priority, these tools will stay in place. The survival of predictive policing rests on this protective function, not on whether it stops crime.

Explore further:

What happens to predictive policing systems when communities not only adapt their behavior but collectively fabricate misleading crime patterns to manipulate the algorithm’s predictions?

What would happen to the adoption of AI in crime prediction if public security funding were decoupled from austerity-driven performance metrics?

AI Crime Prediction

AI crime prediction tools spread because they meet budget audit needs, not because they cut crime, so changing financial incentives would weaken their appeal.

When public budgets focus more on saving money than on community well-being, police use of AI crime prediction tools grows. These tools are not mainly used to prevent crime. They are used because they generate reports that justify spending. In times of tight budgets, such as after the 2008 financial crisis, governments in wealthy countries faced pressure to show quick results. Measures promoted by bodies like the European Commission encouraged short-term, measurable outputs. Crime prediction systems fit this need perfectly. They turn uncertain legal grounds into clear numbers and charts. This makes them valuable for passing budget audits. The real benefit of these systems is not stopping crime. It is helping agencies meet financial rules. If funding were not tied to such narrow performance targets, these tools would lose much of their value. Changing how budgets are judged would weaken the hold of AI prediction in policing more effectively than better data or privacy rules would.

AI Crime Prediction

AI crime prediction spreads when budgets depend on measurable results because algorithms supply the data needed to justify spending, so removing that pressure reduces adoption.

Public safety spending often depends on showing quick results. After 2010, UK police faced deep budget cuts. At the same time, they had to prove they were cutting crime. This led them to adopt AI tools that promised measurable outcomes. Other wealthy nations have seen the same pattern. Tight budgets push police to use systems that produce numbers. These numbers justify spending in audits. Algorithms that predict crime fit this need perfectly. They generate data that looks like progress. But if funding were not tied to such metrics, this pressure would fade. Without the need to prove cost savings through data, police would have less reason to use AI. The main driver for adopting these tools is not bias or legality. It is the demand to show performance through simple numbers. Remove that demand and the appeal of AI weakens. The link between budgets and measurable results is what pushes adoption. Break that link and use of AI in policing would drop.

Explore further:

What happens to public safety innovation in jurisdictions where fiscal pressure is absent but political demand for visible crime reduction remains high?

What would happen to predictive policing outcomes if the historical arrest data used to train algorithms were adjusted to account for documented racial bias in past enforcement?

Predictive Policing Systems

Predictive policing still targets minority neighborhoods because risk scores rely on city service records shaped by segregation, not just crime data.

Predictive policing systems use historical arrest data to forecast crime risk. Some systems adjust this data to reduce racial bias. Yet these systems still target minority neighborhoods. The reason is not the arrest data alone. It is how police connect with other city services. Emergency medical calls and housing inspections are more common in poor areas. These services generate reports used by police models. Even if arrests are counted less, these reports remain. They reflect old patterns of urban inequality. In Chicago, the Strategic Subject List scores people based on their social networks. It uses health and housing records. This treats risk as shaped by environment, not just past crime. The city's racial segregation shapes where these services operate. So, the models still focus on the same communities. Using technical fixes does not change this. The system appears neutral but still reinforces surveillance in minority areas. Adjusting data does not change the outcome. The deeper structure remains.

Explore further:

If predictive systems were trained on decoupled municipal data that excludes non-criminal service records, would the persistence of racialized outcomes imply that the model's structure, rather than its inputs, determines the distribution of surveillance?

If predictive crime systems depend on continuous data updates but most jurisdictions lack integrated databases, what prevents agencies from sharing data despite the known limitations of fragmented systems?

Police Data Sharing

Predictive policing fails not due to technology but because fragmented authority blocks data sharing needed to update predictions.

Police departments across cities often fail to share data. This is not due to lack of technology. It happens because each department guards its own authority. Local control acts like a veto against joining systems. U.S. policing is legally fragmented. This structure stops the creation of unified data systems. Crime prediction tools need constant updated data from many sources. When agencies use different rules, they limit data sharing. This breaks the cycle of updating predictions with real events. Accurate crime risk maps cannot form without steady data flow. Only a few big cities have overcome this. They use coordinated command systems. These allow broader data access. Without rules forcing agencies to share, the full potential of crime prediction tools cannot be reached. The main barrier is not poor data or weak technology. It is the lack of required cooperation between independent agencies.

What if police departments used crime data not shaped by historical biases—would predictive algorithms still reinforce spatial disparities in surveillance?

Police Data Tracking

Predictive policing reinforces inequality because public service data, shaped by urban inequality, tracks people in marginalized neighborhoods regardless of actual crime.

Predictive policing uses records from health, housing, and child welfare systems to assess crime risk. This practice is supported by federal policies that encourage cities to share data. These records act as proxies for criminal risk, even though they come from non-criminal services. The data is used because of long-standing urban inequality. Poor and marginalized neighborhoods have more contact with public services. This means those areas appear in the data more often. Risk scores reflect where people live, not whether they commit crimes. Fixing the data by adjusting for past bias does not solve the problem. Even with race corrections, the same neighborhoods are flagged. The reason is that public service records are tightly clustered in certain areas. These communities are tracked simply because they depend on services. Routine reporting turns them into surveillance targets. The problem is not faulty algorithms. It is the reliance on data from unequal systems.

Explore further:

If predictive policing systems were required to exclude all non-criminal administrative data, would they still produce geographically concentrated risk patterns?

If the predictive power of non-personal data proxies depends on their historical correlation with policed populations, would predictive policing lose effectiveness in a society that had no history of discriminatory policing?

Predictive Policing Flaw

Predictive policing fails in fair societies because it relies on biased historical data patterns, not actual crime risk.

Predictive policing systems often use indirect data sources. These include anonymous 911 call volumes or aggregated movement patterns. Their accuracy depends on how closely these data sources matched past police activity. A French study of the ALTO system confirmed this link. The key issue is not data detail but how police have historically focused on certain areas. This focus makes emergency calls more common in over-policed neighborhoods. More calls lead to more police attention in the future. This cycle continues even if no one’s privacy is violated. The system learns to target places based on past policing, not crime levels. If past policing had been fair, these data patterns would not exist. Then the models would not work as well. Their predictions rely on biased patterns. Without those biases, the models lose their power.

Predictive Policing Failure

Predictive policing fails without a history of biased enforcement because its accuracy depends on persistent patterns from past policing, not on current behavior.

In countries with strong data privacy laws, predictive policing works not by tracking individuals but by relying on long-standing patterns of police activity. These systems use indirect data, like traffic flows or service calls, as stand-ins for past crime patterns. The data itself is not the key. What matters is that police practices have remained consistent over time. This consistency allows systems to forecast crime based on historical enforcement patterns in specific neighborhoods. The forecasts stay accurate because past policing concentrated in certain areas. When police activity reflected bias, those patterns got built into the data. The models depend on that history. If a society changed its policing practices and ended biased enforcement, the old patterns would no longer apply. Then, the indirect data would no longer predict crime well. This happens because the systems rely on echoes of past policing, not on real-time behavior. Studies of crime prediction tools in cities with long-standing patrol patterns confirm this. When models were tested on times before heavy policing, their accuracy dropped sharply. Predictive policing thus fails when past enforcement patterns are no longer present. The systems do not work in the absence of historical bias.

What happens to predictive policing's effectiveness when communities selectively underreport crimes not uniformly but only in neighborhoods undergoing rapid gentrification, altering both data patterns and political accountability simultaneously?

Crime Data Gap

Predictive policing loses accuracy in gentrifying neighborhoods because community distrust reduces crime reporting, and strong oversight rules amplify the resulting data gaps.

Predictive policing systems rely on consistent reporting of crimes and police activity. In cities, these systems need reliable data to forecast crime hotspots. When neighborhoods change quickly due to gentrification, trust in police often declines. Residents who fear displacement avoid contacting authorities. This leads to fewer reported crimes, not because crime dropped but due to fear of attention. The missing reports create gaps in data. Algorithms mistake this silence for safety. This misreading happens most where oversight rules protect privacy. Strong data laws limit how much information police can keep. But these same rules make systems more sensitive to reporting drops. When communities withdraw from reporting, the data grows less accurate. Predictive models then fail to spot real risks. Accuracy declines in areas most protected from surveillance. The problem is not broken technology. It is that accountability changes the data. Fewer reports undermine the system’s foundation. So predictions become less reliable. This occurs exactly where civil safeguards are strongest. Therefore, in changing urban areas, the systems lose effectiveness.

What happens to predictive policing systems when communities not only adapt their behavior but collectively fabricate misleading crime patterns to manipulate the algorithm’s predictions?

Mutual Aid Changes Data

Mutual aid networks reduce formal service use, which breaks the link between government data and actual risk, causing predictive models to fail.

After 2010, U.S. cities began using everyday government service records to assess community risk. They relied on data like welfare use or housing aid to map danger and need. The method assumed disadvantaged groups always appear in government data due to lack of other options. But during the 2020–2023 housing crisis, many high-risk neighborhoods turned to mutual aid groups. These community networks offered health and safety support outside government systems. As more people used these alternatives, their interactions with official services dropped. This shift broke the link between government data and real-world risk. Algorithms still treated low service use as low need. But low data now meant avoidance, not safety. The models failed because they misread disengagement as low risk. The issue was not missing data alone. It was that trust in institutions had shifted. Community action changed how data reflected reality.

Algorithmic Policing Tools

Algorithmic policing tools persist because they fit audit and reporting systems, not because they reduce crime.

When governments focus more on budget accountability than on actual results, public safety algorithms are kept not because they reduce crime. They are used because their reports meet requirements set by higher-level authorities. Austerity has made budget justification more important than crime reduction. This has led agencies to adopt predictive systems that produce easily audited outputs. These outputs help justify spending, even if they do not prevent crime. The European Semester and UK cost benchmarks show how performance is measured by financial compliance. In many OECD countries, predictive policing continued after 2008, even when reviews found no drop in repeat offenses. Despite poor results, these systems were retained. Their main advantage is fitting into audit processes. Police departments keep using them because they help pass funding reviews. The systems become embedded in routine certifications, as seen in France and Germany. Even after austerity eased, the tools stayed because they serve reporting needs. The real reason these systems persist is not crime control. It is their usefulness in financial and administrative reporting. Therefore, simply removing budget pressure will not end their use. The deeper reason is how reporting systems are built.

What happens to public safety innovation in jurisdictions where fiscal pressure is absent but political demand for visible crime reduction remains high?

AI Crime Prediction

AI crime prediction spreads in wealthy democracies because it produces visible, auditable data that meets political demands for measurable action, not because it reduces crime more effectively.

In wealthy democracies, police use AI to predict crime even when crime is already low. This is not because AI reduces crime better than other methods. It is because governments need to show clear results. People expect fast, visible progress on safety. Electoral cycles reward short-term appearances of success. Traditional methods do not produce constant, detailed data. AI systems generate real-time risk numbers that appear precise and objective. These numbers are easy to report and audit. Public funders like the European Union reward projects that produce measurable outputs. Even if crime is falling, officials must prove they are acting. AI fills this need by supplying data that looks rigorous. The pressure to display action drives AI adoption. Effect forces. The key factor is the demand for visible performance metrics.

If predictive systems were trained on decoupled municipal data that excludes non-criminal service records, would the persistence of racialized outcomes imply that the model's structure, rather than its inputs, determines the distribution of surveillance?

Neighborhood Surveillance

Neighborhood surveillance targets areas with rising property values because city policies tie safety to real estate, making data density follow money instead of crime.

City governments often link public safety to property values. This shapes how they use surveillance tools. Areas with rising real estate values get more cameras and monitoring. This happens even if crime is not increasing there. City budgets and zoning policies favor developers over community safety programs. As a result, data collection focuses on neighborhoods seeing new investment. More sensors mean more digital records in these areas. Surveillance density follows money, not crime. Federal reports confirm that police data clusters where property values rise fastest. Predictive systems then see these areas as high priority. This pattern repeats racial bias. It does so not because of flawed data or broken communication between agencies. The root cause is deeper. Safety efforts serve real estate goals more than public needs. So race-based prediction persists. It does so because data coverage reflects economic interests first. Urban policy makes some areas more visible to machines. That visibility depends on capital, not crime.

Predicting Crime Using City Data

Crime prediction systems spread surveillance unevenly by treating frequent city service use as risk, which echoes past racist urban policies.

Crime prediction tools often use records from city services like 311 calls and housing inspections. These records link public health issues to crime risk. This happens not because race is directly used but because city services have historically focused on certain neighborhoods. Past policies like redlining and urban renewal shaped where services are used today. In New York City police used a system called Patternizr. It relied on data from multiple city agencies. Even without arrest records the system still targeted the same neighborhoods. This is because service calls happen more often in areas where poverty and racial segregation overlap. More calls create more data. The models treat this as a sign of higher risk. So people in these areas face more police scrutiny. This pattern continues even when criminal records are left out. The reason is not bias in one data source. It is how the model treats everyday service use as a risk signal. This design choice spreads surveillance unevenly. It leads to more monitoring in the same over-policed areas. The result is racialized outcomes without using race as a variable.

If predictive policing systems were required to exclude all non-criminal administrative data, would they still produce geographically concentrated risk patterns?

Risk Maps In Cities

Predictive policing targets poor neighborhoods because frequent public service use creates data patterns that algorithms treat as risk signals, not because of crime.

Predictive policing systems focus on certain city neighborhoods. They do this not because of crime data but because of non-criminal records. These records include visits to public health clinics, housing complaints, and child services referrals. Such data pile up in poor neighborhoods due to long-standing inequality. Federal urban policy encourages cities to share this data. Programs like Byrne JAG grants reward the use of service records as markers of risk. This turns routine help-seeking into grounds for surveillance. The result is that neighborhoods with more state contact appear riskier. This happens even if arrest records are removed. The reason is that poor communities interact more with public services. These interactions create a pattern of visibility. Algorithms use this pattern to assign risk. So the system keeps targeting the same areas. It does so even if no criminal data are used. The concentration of state presence in specific neighborhoods drives the result. As long as people in poor areas rely more on public services, those areas will be flagged. The location of state contact shapes risk scores. This contact is tied to place, not crime. So risk maps stay the CDC same.

Predictive Policing Data

Predictive policing fails when past bias in enforcement is removed because its accuracy depends on skewed data from unequal police presence.

Predictive policing systems use data like 911 calls or movement patterns to forecast crime. These data seem neutral but reflect past policing behavior. Police have long focused on specific neighborhoods. This focus shaped how often calls were made and incidents recorded. Over time, certain areas appeared riskier due to higher reporting rates. But this was not because crime was more common. It was because police were more present. More presence led to more reports. This pattern became embedded in the data. Algorithms learn from this data and repeat the same patterns. They treat high reporting as if it means high crime. But this only remains true where policing has been uneven. In places with fair oversight and equal monitoring, reporting rates are similar across areas. There, crime risk does not cluster by neighborhood. So the data no longer show strong patterns. Predictive models lose accuracy. Their power relies on past bias. Without bias, the patterns vanish. Therefore, these systems only predict well in unequal environments.

Library

Pre-Emptive Justice: The Ethics of Using AI to Predict and Prevent Crime

Key Findings

Predictive Policing Data

Predictive Policing Bias

Predictive Policing Feedback Loop

Crime Prediction Tools

Predictive Policing Feedback Loop

Predictive Policing Gaps

Policing By Prediction

Deeper Analysis

What happens when predictive policing systems are trained on non-personal or aggregated data that falls outside the scope of GDPR but still influences law enforcement decisions?

Predictive Policing Loophole

What happens to the effectiveness of predictive policing algorithms when communities systematically alter their reporting behaviors to game the system or reduce police exposure?

Police Prediction Failure

What happens to predictive policing systems when the communities they target reorganize their behavior to exploit the algorithm’s reliance on historical data patterns?

Predictive Policing Failure

Blame Avoidance Tools

What would happen to the adoption of AI in crime prediction if public security funding were decoupled from austerity-driven performance metrics?

AI Crime Prediction

AI Crime Prediction

What would happen to predictive policing outcomes if the historical arrest data used to train algorithms were adjusted to account for documented racial bias in past enforcement?

Predictive Policing Systems

If predictive crime systems depend on continuous data updates but most jurisdictions lack integrated databases, what prevents agencies from sharing data despite the known limitations of fragmented systems?

Police Data Sharing

What if police departments used crime data not shaped by historical biases—would predictive algorithms still reinforce spatial disparities in surveillance?

Police Data Tracking

If the predictive power of non-personal data proxies depends on their historical correlation with policed populations, would predictive policing lose effectiveness in a society that had no history of discriminatory policing?

Predictive Policing Flaw

Predictive Policing Failure

What happens to predictive policing's effectiveness when communities selectively underreport crimes not uniformly but only in neighborhoods undergoing rapid gentrification, altering both data patterns and political accountability simultaneously?

Crime Data Gap

What happens to predictive policing systems when communities not only adapt their behavior but collectively fabricate misleading crime patterns to manipulate the algorithm’s predictions?

Mutual Aid Changes Data

Algorithmic Policing Tools

What happens to public safety innovation in jurisdictions where fiscal pressure is absent but political demand for visible crime reduction remains high?

AI Crime Prediction

If predictive systems were trained on decoupled municipal data that excludes non-criminal service records, would the persistence of racialized outcomes imply that the model's structure, rather than its inputs, determines the distribution of surveillance?

Neighborhood Surveillance

Predicting Crime Using City Data

If predictive policing systems were required to exclude all non-criminal administrative data, would they still produce geographically concentrated risk patterns?

Risk Maps In Cities

Predictive Policing Data