Infosys co-founder suggests India create own AI facial biometric training datasets to prevent bias

For India to successfully roll out facial recognition across the country, it needs to create its own databases for AI and machine learning technologies, Infosys co-founder Kris Gopalakrishnan says, according to The Economic Times.

Before rushing with wide implementation, India should consider test pilots to train algorithms with facial images from the local community, fearing the data used so far was from white Anglo-Saxon men.

“We also need to look at biases. One of the reasons why I believe India must do research in artificial intelligence (AI) and machine learning (ML) particularly is because most of the databases that are used to train these systems which we use today are being trained with data which is not from India,” he told PTI in an interview.

Gopalakrishnan is confident AI and machine learning can be used to deliver a number of new services, including facial recognition, text translation, and diagnostics.

The National Crime Records Bureau (NCRB) announced last year that it would deploy a facial recognition system to find missing children and apprehend criminals, but it has pushed the deadline back five times.

“We need to sit down and come out with ethics governing the facial recognition as it is an issue that the people all over the world are grappling with,” he said in a conversation about privacy risks and surveillance.

“We need to start building our own databases. Training databases must include sufficient data from India. It is a chicken and egg problem. Because when you have huge amount of data, if you want to leverage the computing power to analyze that data, and come out with patterns or inferences, machine learning is the tool to use to find the patterns for you,” he said.

Gopalakrishnan emphasizes the importance of India collecting its own biometric data including standardized DNA sequencing databases to train the system and develop new treatment.

“We should be doing our own trials before deciding whether it works in India or not. The human being is as important as these tools because we make the decision to either not use the data, or to fine tune the data,” Gopalakrishnan added.

New NIST research on demographic differentials of biometric facial recognition accuracy, or ‘bias,’ found that despite significant improvement over previous research, there is still a difference in the accuracy of some algorithms in matching women and people with darker skin.

Article Topics

artificial intelligence  |  biometric bias  |  biometric dataset  |  biometrics  |  facial recognition  |  India  |  Infosys  |  machine learning  |  training