AI Failed Mahakumbh Mela

In a tragic incident, more than 30 people died, and 60 others sustained injuries after a stampede broke out at the Mahakumbh Mela in Uttar Pradesh’s Prayagraj on Wednesday. According to authorities, several reasons led to the accident, including the continuous movement of VIPs and the closure of bridges for three days due to ‘Amrit Snan’. 

Moreover, the mishap took place during night-time, and even the officers who were monitoring the event with AI cameras couldn’t foresee the situation around Sangam Nose going out of control. 

To understand why the management failed, we contacted the CEO of a company, who has previously worked on events offering AI tools and cameras for security surveillance.

In an exclusive conversation with AIM, he anonymously said that the bid to offer security for the Mahakumbh Mela was open to all. However, certain consultants made it difficult for some companies to participate by imposing stringent requirements, such as high turnover thresholds. 

The key objective of this bid was to prevent stampedes. Multiple pathways lead to key locations, such as ghats designated for bathing. Each Ghat has multiple routes, and the idea was to install cameras at entry points, along the transit, and at exit points. These cameras are supposed to continuously count the number of people on each path.

If the number of people exceeds a predefined limit, an alert would be triggered. For example, if two out of eight paths were open, and each had a maximum capacity of 1,000 people, exceeding this threshold would result in rerouting.

The bid’s primary requirement was a real-time crowd-monitoring system to detect congestion and prevent stampedes. However, an internal audit took place around 14-15 days after the event began, revealing serious flaws in execution. On the first day alone, the cameras recorded 40,000 people, while the actual crowd was estimated at 3 crore (30 million). This discrepancy was a major issue from the very beginning.

The problem lies in the way the bid was structured. The CEO alleged that companies offering top-notch products and services couldn’t secure the bid. The contract was awarded to the lowest bidder, and it appears to have been tailored in a way that restricted participation. Ultimately, an AI-centric company did not win the bid. Instead, it was awarded to a combination of firms.

According to a source, resistance from the police further complicated matters. The police conducted their own headcounts and submitted them to the Integrated Command and Control Centre (ICCC), but discrepancies remained. 

Senior officials intervened and raised concerns, but no substantial improvements were made. Despite efforts, stampedes still occurred, and paths were blocked without clear technical or scientific justification.

VIP movements played a role, but the larger issue was poor planning. Authorities knew in advance that 40 crore (400 million) people would attend, yet AI-based predictive models failed to function as promised. The tender explicitly stated that AI would be used for crowd flow management, raising a critical question: why did it fail? Was the failure due to a poorly designed bid by the consultant, inherent biases in the model, or nepotism in the selection process?

The implemented system couldn’t even perform a basic YOLO (You Only Look Once) model-based people count. If implemented correctly, the headcount would have been accurate. However, ₹40 crore was spent over 90 days just to count people, and it still failed. When mismanagement leads to the deaths of 30 people, it is nothing short of a complete failure.

AI-Powered Mahakumbh Mela

Authorities are relying on AI-enabled cameras, RFID (radio frequency identification) wristbands, drone surveillance, and mobile app tracking to monitor pilgrim movement throughout the 45-day event. This was believed to allow for efficient crowd regulation by tracking entry and exit times with participants’ consent.

Further, the Uttar Pradesh government launched ‘Kumbh Sah‘AI’yak’, an AI-powered chatbot available in 11 languages, to assist pilgrims. This chatbot provides visitors with all the necessary information regarding key dates and events at the Mahakumbh.

However, despite these efforts, the situation still demands more advanced technology to effectively handle disasters.