New Delhi: It was Mayank Tripathi’s second attempt at the Uttar Pradesh Police Constable Recruitment examination. He had been preparing for government jobs for the last seven years, but, from paper leaks to cheating scandals, something always got in his way. This time, however, the UP government had a new trick up its sleeve: artificial intelligence (AI).
On 31 August, the state government conducted the Uttar Pradesh Police Constable Recruitment examination under strict security measures. This exam, overseen by Rajeev Krishna, Director General of Vigilance in Uttar Pradesh, was one of the largest recruitment exams in India with 48 lakh applicants vying for 60,244 posts—though only 34 lakh candidates actually appeared for the test. Among other security measures, it marked the first time a state used AI to ensure the integrity of an exam.
“I have given many government job exams but this was the first time entry was conducted through biometric and facial recognition. Every thread, locket, and ring was removed at the gate. They even made an easy access page for aspiring candidates so they can smoothly check the information,” said Tripathi. He appeared for the exam at the Sultanpur centre.
Authorities believe the experiment was a success. They are now being approached by other states seeking to improve the government recruitment examination process.
“With this, we have laid a very good foundation. Now the major part is over but in the next steps, we will also use the AI for document verification and physical efficiency tests. We are trying to create a framework with zero human involvement,” Krishna told ThePrint.
Also Read: Who is the mysterious UPSC whistleblower? His truth bombs shake up coaching institutes
Behavioural monitoring & papers set with AI help
For the UP Police Constable Recruitment exam, at the 1,174 centres in 67 districts, CCTV cameras were equipped with artificial intelligence to carry out facial recognition across 16,440 rooms.
The technology not only matches faces but also monitors facial movements, so the CCTV cameras tracked the candidates’ movements as they completed the examination process. By identifying data points associated with the exam takers, AI was able to help prevent incidents of impersonation, which has been a major issue.
“The exam was conducted in 1,174 centres and we chose the government schools for it where CCTV cameras were available. It was a challenge as 34 lakh candidates appeared for the exam. And by using the AI technology we caught 530 cases in which people were impersonating or cheating the system so we can say it was a successful attempt,” Rajeev Krishna told ThePrint.
The UP Public Service Commission, which conducted the exam, worked with a number of agencies to create the system for this examination. Each of the companies takes responsibility for carrying out different tasks: spotting fake identities, making sure the photograph isn’t 10 days older, and minutely observing the candidate’s activity before they enter the exam room.
“We have worked with four AI agencies to go through with it and got huge support from the government too,” added Krishna. The administration held several meetings, led by Chief Minister Yogi Adityanath, to establish the framework for this exam.
“There was behavioural monitoring as well to observe if the candidate was suspicious. The AI agencies use different algorithms to address issues. Even though the question papers were made with AI’s help, there were multiple sets so the possibility of paper leaks can be neutralised. The companies use a randomisation network for this,” said Deepanshu Singh, a public policy expert working in AI with the Government of India.
To achieve this, normally, agencies employ different algorithms—such as the random forest which creates decision trees and then averages them out to make predictions, or the convolutional neural network, which helps computers recognise patterns in images, like identifying objects or faces.
In the case of the random forest classifier, once assigned a dataset, it divides the same into subsets and distributes them to tree-like hierarchical models (or decision trees). Each decision tree then comes up with a prediction; whenever a new data point enters the picture, the classifier, based on the majority of the results, predicts the final decision.
Convolutional neural network, meanwhile, is how visual imagery is used as input to identify the object in an image. For instance, when pixels of an image of a cat are fed in the form of arrays to the neural network’s input layer, which are then mined for hidden layers that extract features from the image. They are then put together in a fully connected layer that identifies the object in the image.
The government worked with the different agencies to come up with a a custom algorithm to help conduct the exams.
The government also took other steps to maintain the exam’s integrity. For instance, the question papers were generated just 30 minutes before the start of the examination process, where earlier they used to be prepared at least one day in advance.
The exam was conducted over five days in two shifts. More than 2,300 District Magistrates and 1,97,859 police personnel were deployed to ensure that the exam was conducted smoothly.
In the past, 29-year-old Sonu Yadav has participated in many protests over the Uttar Pradesh government’s failure to conduct a fair examination.
“I have never seen police personnel at the exam centre in such huge numbers. I reached the centre two hours before the exam and did my identification through biometrics followed by facial recognition,” Yadav, who lives in Lucknow and also took the exam at the Sultanpur centre, told The Print. This new identification process, he said, takes longer, but all aspirants are willing to reach the exam centre ahead of time if it works out well.
Transforming system & expanding use of AI
For years, candidates across the country—from Haryana and Jharkhand to Maharashtra and Rajasthan—have suffered because of issues that plague the government examination system, namely paper leaks, exam cancellations, and operations of paper-solving gangs.
However, with the introduction of AI, UP is hoping to transform the system and other states are also seeking help.
“This exam process was a success as 4-5 other state public service commissions have reached out to us. They want to use the same method to conduct the exams smoothly and hassle-free,” said Krishna.
The UP government is also looking to expand the use of AI to other parts of the examination, including handwriting detection and checking copied content.
“In the next step, we will install software (on the systems used at the exam centres) for the online examination that can measure the time candidates have taken. So if one candidate completed the exam in the last 10 minutes, we will get to know that the person cheated,” said Deepanshu Singh.
“There will be checks on copied content, text similarity, and plagiarism. The software will keep getting modified and new technologies will be joining the framework too,” he added.
Monty Shukla, 23, who appeared for the constable recruitment exam for the first time, is hoping the measures will be adopted for other state examinations.
Before him, his elder brothers spent years preparing for government recruitment exams. One of them has now joined a coaching centre to make ends meet. His father, Shukla added, has struggled every month to send money for his expenses.
“My father is a farmer and he has taken a Rs 65,000 loan for me. It is not just my father but my entire village that prays for me to get a government job, that is the only way I can change the fate of my family. But the system is so unfair and corrupt that every exam lands in trouble. I hope this method will get adopted in the other exams too,” said Shukla.
(Edited by Sanya Mathur)