The guitar became a pillion rider: Bengaluru's AI traffic cop keeps fining its mistakes

In mid-September, Bengaluru resident Souvik Dutta rode his Ather 450X electric scooter from Hebbal towards Tin Factory on the Outer Ring Road with a guitar strapped to his back. An AI-powered traffic camera watched him pass at 12.19 pm on September 15 — and decided the guitar case was a human being. The system recorded the violation, and an online challan of Rs 500 was generated the next day and sent to the scooter's registered owner, Dutta's wife. The challan photograph told a different story: a helmeted rider with a black guitar case strapped to his back. Dutta posted the image online and appealed to the Bengaluru Traffic Police to revoke what he called a fine issued against his instrument. A senior traffic police officer told The Hindu that the incident was a judgement error by an AI camera that forms part of the Bengaluru Traffic Police's Intelligent Traffic Management System, or ITMS. Crucially, the officer admitted the error did not stop at the camera: the flagged violation went for manual review, and a challan was issued anyway through human oversight as well. That punctures the system's central promise. Joint Commissioner of Police (Traffic) Karthik Reddy has claimed the ITMS, with minimal human intervention, has nearly 99% accuracy and that wrong flagging is negligible. The Hindu's review of the police's own data tells a more complicated story. According to ASTraM data, the Bengaluru Traffic Police issues an average of nearly 26,500 challans through contactless means every day — and over 19,000 of those are generated after violations are identified by AI cameras. Of those 19,000, about 25 cases, or 0.13%, are challenged daily. But as the explainer notes, the challenge rate cannot be treated as a measure of accuracy: contesting a small fine costs time most motorists would rather spend elsewhere, so most wrongful fines simply get paid. Another senior officer told the paper that after filtering the data for cases rectified during manual review by the Traffic Management Centre workforce, the AI's real accuracy may be closer to 90%. By the end of 2023, the police themselves had recognised accuracy had dropped to 80% for certain violations — pillion riding, helmet and seatbelt non-use, and signal jumping. The accuracy has been scaled up over the past two years, the police say, but the guitar incident shows the gap between a press claim and the street. The ITMS, adopted in December 2022 to minimise human intervention in issuing challans, uses artificial intelligence and machine learning to detect violations automatically; the feed goes to the Traffic Management Centre, the BTP's head office. The deeper question the guitar has raised is about the burden of proof. In a manual system, an officer who makes a mistake owns it. In an automated one, the machine accuses, a rushed reviewer rubber-stamps, and the citizen must prove the camera wrong — with photographs and portals. The technology processes violations at a scale no police force could match. Accountability has not scaled with it.