Special Intensive Revision: AI detects 6.41 lakh anomalies in Bengaluru

For the first time since the enumeration phase began, the Greater Bengaluru Authority (GBA) has disclosed that there have been at least 6.41 lakh “logical discrepancies” or anomalies within its jurisdiction. So far, only 38.81% of enumeration forms were digitised as of 4 p.m. on Saturday. The overall figure also includes 2.66 lakh anomalies flagged in Bengaluru Urban.

However, the nature of these discrepancies has not been disclosed in the official progress report. 

GBA officials confirmed to The Hindu that Artificial Intelligence (AI) is being used to identify these discrepancies. “An AI software tool is used to identify discrepancies. Once a BLO digitises and submits the enumeration forms, the system flags any anomalies,” a senior official said. 

The official added that the same software is also used to identify duplicate voters through the Demographic Similar Entries (DSE) tool, which flags voters sharing identical demographic details, such as names or fathers’ names, across electoral databases.

“The DSE tool also flags logical discrepancies arising from changes in names or other demographic details,” the official said. 

However, while the GBA’s daily progress report separately discloses the number of anomalies, the Statewide daily progress report makes no mention of this category even now. 

The disclosure comes even as Chief Electoral Officer (CEO) V. Anbu Kumar reiterated that anomalies will be known only after the publication of the draft electoral rolls. 

The GBA has a total electorate of over 1.03 crore. Apart from anomalies, more than 15.83 lakh voters have been categorised as “unmapped or no mapping done” — indicating that they submitted their enumeration forms without filling in the 2002 electoral rolls details because they were unable to trace them.

“Many voters have at least one discrepancy. Several have two or three discrepancies, such as spelling variations across documents or expansion of initials,” the official said. “However, the AI is also flagging anomalies that are impossible. For instance, it flags a discrepancy saying that the gap between two progeny linked to the same parent in the previous SIR is less than nine months.”

Officials said it is difficult to estimate how many discrepancies will ultimately be identified before the draft electoral rolls are published. Before the draft rolls are finalised, a “data scrubbing” exercise — as seen in West Bengal — will be carried out to remove discrepancies that may have been wrongly flagged by the AI. Post that, only a mechanical review of the remaining cases will be undertaken.