One of the first cities around the world to embrace predictive policing was Santa Cruz, California, in 2012. Newspapers at the time were flooded with stories about how predictive policing is assisting law enforcement, and the technology was frequently touted as a much-needed police tool. It was then followed by many cities globally to curb the criminal activities in their neighbourhood. 

One of the first law enforcement agencies in India to use this technology was the Jharkhand Police. Currently, the state police have a Geographic Information System (GIS) that provides specialised maps that are essential to operations against Maoist organisations, a Naxal Information System, a Crime Criminal Information System (to be merged with the CCTNS), and a Naxal Information System.

Jharkhand police, one of the most technologically advanced forces in the nation, uses a Crime Analytics Dashboard to display the frequency of crimes by type and location on an easily accessible website. This gives officers the most recent information and undoubtedly increases their situational awareness.

The Delhi Government has also been working with ISRO (Indian Space Research Organization) to launch a tech-dubbed Crime Mapping Analytics and Predictive System (CMAPS), according to a recent RajyaSabha report. This programme will make use of space tech to prevent crime and uphold the rule of law.

The report stated that a personal digital assistant (PDA) will be given to the police officers, and this assistant would be connected to a storage processor that houses criminals’ data. Furthermore, a digital message will be created from each distress call. Additionally, real-time data collection at the crime site can also be done using this module.

Additionally, Himachal Pradesh has set up over 19,000 CCTVs to create a “CCTV Surveillance Matrix.” Aiming for one camera per 100 residents, the state would need to deploy nearly 68,000 cameras. In essence, it is creating a foundation for predictive policing.

In order to map, visualise, and generate inquiries and reports regarding crimes and criminal occurrences, UP police and ISRO also collaborated in 2018. The MoU between the state police and ISRO was to last for next three years, i.e., till 2021. 

According to the NCRB newsletter, after an onsite visit to Hyderabad from March 21–25, 2022, the team recently installed the Crime Mapping Analytics and Predictive System (CMAPS) in Telangana successfully. 

However, in 2020, Santa Cruz became the first city in the USA to ban predictive policing. 

Bias in predictive policing

The reason why Santa Cruz and many cities overseas, including Chicago, New York, and Los Angeles, stopped using predictive policing is the inconsistent data it was being fed. Predictive policing initiatives employ incomplete data to make inaccurate predictions if they are prejudiced towards particular communities or groups of individuals. According to some experts, predictive policing encourages over-policing in minority communities, which serves to perpetuate the country’s endemic structural racism.

For instance, the LASER programme (from LA) was terminated in 2019 as a result of an internal audit that found “serious flaws with the programme, including inconsistent practises in the selection and maintenance of persons in the system.”

It was also found that the officers who were using the data to patrol particular neighbourhoods may not have had a valid justification to make arrests since the algorithm was running on inconsistent data. 

Back Home

Caste- and community-based discrimination has been levelled against Indian police. In a 2019 study of 11,834 police officers from 21 states, around 50% of participants said that immigrants and Muslims were more likely to commit crimes.

In 2018, a video of a Maharashtra-based IPS officer admitting to fabricating many cases against Dalits became viral. Police brutality against particular groups in Indian society, such as the Pardhi community in Madhya Pradesh, has also occurred routinely.

During the recent protests in Delhi against the CAA and NRC, there is verified evidence of the use of facial recognition technology and predictive policing, and as per reports, 44% of the questioned police officers preferred police punishment over legal proceedings.

Bias in Data 

India’s homegrown Crime Mapping Analytics and Predictive System (CMAPS) may have some kind of bias in itself, primarily because of the system by which data is being fed into the system. As per some experts, majorly the data received from Dial 100 and FIR is fed into the module, which, itself, is assumed to contain bias to a certain degree. 

•  Historical Bias

Although gathering information is one of the main tasks of law enforcement, many experts think the procedure is not always unbiased. Since colonial times, information has always been gathered in a discriminatory manner—with caste, gender, class, and religious minorities being subject to more monitoring.

Compared to those who live in Delhi’s affluent neighbourhoods, those who reside in slums frequently encounter law enforcement officials who are indifferent to their needs. Conversations within the call centre, according to a research article, support the assertion.

This, together with the contentious history of the police in regard to discrimination and brutality, raises the possibility that historical bias is not just ingrained but also purposefully formalised and injected into the data.

• Representation Bias

As input data for CMAPS majorly consists of calls to the Dial 100 call centre and a national database used to track crime and criminals, there might be a significant underrepresentation of individuals from privileged socio-economic backgrounds and also of upscale areas in the data. 

The rationale is that some regions of the city and some social groups are more amenable to the sampling techniques, such as calls to an emergency hotline or pre-existing criminal records in a criminal database (not a database of convictions).

The DMD receives over 20,000 calls every day, according to a study, which also states that some staff claimed that individuals from upscale neighbourhoods “hardly contacted” and that the vast bulk of these calls was from slums. This might indicate that the module will be inclined to think that crime is more likely to occur in high-engagement regions, creating a vicious cycle of increased attention for the most vulnerable—which would eventually result in more arrests and reports from these places.

• Measurement Bias

Rich neighbourhood residents occasionally struggle to identify the area from which they are calling the police; for instance, when a victim calls with a complaint, they typically don’t even recognise the closest police station, which results in incorrect information being entered into the system.

Because of the more nuance in data coming from Delhi’s affluent neighbourhoods and the less accurate physical distribution of the city, information clusters tend to be less numerically overwhelming and attract less attention in the future. This bias grows as a result of institutional blind spots and the inability of those who are vulnerable to engage with the system and other people.