How the East got ahead of the West on facial recognition technology

I had written back then that while these Big Tech firms were responding to pressure in the wake of the Clearview AI scandal, work on facial recognition tech would continue unabated (bit.ly/4c82BQv). 

To jog your memory, Clearview AI, without specific permission, was accessing all platforms, like Facebook, Instagram and others, that may have had images of our faces.

At the time, Microsoft extended its moratorium on selling facial recognition tech to police indefinitely, emphasizing the need for government regulation before resuming these sales. Amazon initially placed a one-year moratorium on police use of its facial recognition technology, Rekognition, and has since extended this ban. 

Also read: International passengers can soon breeze past security and immigration with Digi Yatra

IBM took a clearer stance by exiting the general-purpose facial recognition market, shifting its focus to more specialized applications such as visual recognition for specific industrial purposes. These companies have largely adhered to their promises, but they continue to develop and deploy facial recognition tech in other sectors.

Meanwhile, other companies have filled the gap. International markets, particularly in countries with less stringent regulations, continue to see enthusiastic development and deployment of facial recognition tech. 

A wag once told me that due to facial recognition and advanced location technologies, at least three governments and possibly 5-6 Big Tech and probably several startups always know precisely where you are. Given the value and convenience that this technology offers ordinary people, this is a good thing.

However, when it comes to racial differentiation, the story is very different—and this is an issue that crops up more often in multi-racial societies, like the US, which explains Big Tech’s skittish stance. 

The issue is that studies have consistently shown that commercially available facial recognition systems are significantly more likely to make errors with faces with skin tones that are not Caucasian. This discrepancy can lead to severe consequences, such as wrongful arrests and prejudice reinforcement.

US government research indicates that facial recognition systems are 10-100 times more likely to misidentify individuals with darker skin tones (bit.ly/3WObwCn). In the hands of law enforcement, such error rates can be dangerous; this highlights the need for improved accuracy and bias mitigation in these technologies, which was why Big Tech firms pulled back. 

In Microsoft’s case in 2019, its software had almost ten times more false positives for women of colour than men of colour (bit.ly/3WNbJFM), while some years prior, Google’s technology was labelling African people as gorillas (bit.ly/3Wrf7Vz). It was wise to exit.

Also read: Businesses had better adapt quickly to India’s new privacy law

Despite the problems observed in the US, India and China have continued to implement this technology extensively, often with claims of high accuracy. This may be because there is less racial heterogeneity in these countries, but there are other reasons as well, primarily governmental support for its use. 

The accuracy rates claimed by facial recognition systems in India are generally high, supported as they often are by integration with national identification systems such as Aadhaar.

China has invested heavily in facial recognition technology, with claims of high accuracy rates supported by large-scale data collection and integration with various surveillance and identification systems. 

The Chinese government reports accuracy rates as high as 99%, though these figures are not always independently verified and may be influenced by Beijing’s investment in state control and surveillance programmes. In China, independent verification and the ethical implications of these technologies remain areas of concern.

Several factors might explain high accuracy claims in India and China that contrast with America’s experience with facial recognition. Government backing of projects, with significant funding combined with access to governmental resources, for example, could have made a difference. 

India and China both have much larger populations than other countries in the world, so their access to vast databases of images allows for more comprehensive training of facial recognition algorithms, improving their accuracy. Moreover, combining facial recognition with other identification methods (such as national ID systems) can enhance the overall accuracy of these systems. 

Regulatory frameworks in India and China also tend to be less stringent than in the US, allowing for more flexible and widespread deployment and iterative improvement of facial recognition technology. At least in India, this has offered citizens convenience, such as faster security checks during air and rail travel.

Also read: Puja Khedkar case fallout: Aadhaar, facial recognition, QR code to come into play as UPSC bats for fool-proof exams

Microsoft, Amazon and IBM’s commitments to stop the sale of facial recognition technology to police forces have been widely endorsed in the US, reflecting a response to ethical and societal concerns. Despite this, the development of facial recognition technology has continued unabated, as I had predicted. 

While racial bias remains a significant issue in the US, India and China continue to advance their facial recognition systems, claiming high accuracy rates. To my mind, this accuracy will increase. And in India, its use cases have so far been of great value to people at large.