The Promise of a Smarter Force
Police forces across India are understaffed and overworked, facing the monumental task of ensuring public safety for over a billion people. Proponents of AI argue that it can be a powerful force multiplier. AI systems can analyse vast amounts of data—like
CCTV footage, call logs, and crime records—at speeds no human could match. This can help identify crime hotspots for more efficient patrols, a practice known as predictive policing. In cities like Delhi, police have used Automated Facial Recognition Systems (AFRS) to scan crowds and identify suspects. Other states are using AI to enhance blurry video evidence, reconstruct faces from unidentified remains, and even solve cold cases, offering a vision of a more efficient and effective police force.
The High Cost of Being Wrong
The problem is that these systems are not infallible. Facial recognition technology (FRT), in particular, can have significant accuracy issues. Studies have shown that these algorithms can be less accurate when identifying women and individuals with darker skin tones, a major concern in a diverse country like India. An incorrect match—a 'false positive'—is not just a technical error; it can have devastating human consequences. In one case following the 2020 Delhi riots, a man was held in custody for over four years based substantially on an 80% facial recognition match from a grainy CCTV image. Shockingly, affidavits submitted in court for Delhi Police's FRT showed an accuracy rate as low as 2% in 2018, which fell to under 1% the next year, yet the technology was still deployed.
The Bias Baked into the System
Beyond individual errors, there's a deeper problem: algorithmic bias. AI systems learn from data, and if the data reflects existing societal biases, the AI will learn and even amplify them. In India, historical crime data is often skewed, with marginalised communities, such as Dalits and Muslims, being disproportionately represented. When an AI is trained on this data to predict crime 'hotspots', it can lead to those same communities being over-policed, creating a feedback loop of suspicion and arrests. This isn't a deliberate act of prejudice by the machine, but a reflection of the data it was fed, turning technology into a tool that can entrench, rather than solve, discrimination.
A Legal and Ethical Void
Despite the rapid rollout of these technologies across at least 170 systems in various stages of deployment, India lacks a specific legal framework to govern their use in policing. Key questions about data privacy, transparency, and accountability remain unanswered. For instance, police are often not legally required to disclose if an identification was made using AI, denying the accused a chance to challenge the technology's reliability in court. The Digital Personal Data Protection Act of 2023, while a step forward, contains broad exemptions for state agencies, which could leave a massive gap in oversight for law enforcement tools. Without clear rules, there is a significant risk of these powerful surveillance tools being used without adequate checks and balances.
Finding a Path to Responsible Use
The solution may not be to completely abandon AI, but to govern it with extreme care. Experts suggest a number of safeguards are urgently needed. These include mandatory transparency about how algorithms work, independent audits to check for bias, and clear laws defining when and how AI evidence can be used. Critically, AI should be used to assist, not replace, human judgement. An AI-generated match should be treated as a lead for further investigation, not as conclusive proof of guilt. By prioritising low-risk administrative tasks and establishing robust oversight, AI can support police work without undermining the principles of justice and fairness.
