AI, as we know, is being trained by feeding real images, but the danger is that it could turn biased. To make it less discriminatory, AI could be trained on synthetic images. Gartner estimates 60 per cent of all data used to train AI will be synthetic by 2024, and it will overshadow completely the real data for AI training by 2030.

Synthetic images market has two types of companies — those who use GANs and those who design 3D graphics from scratch.

Fake data is not used only in computer vision systems but it is also used for predictive software. Banks use the fake data to decide the loan eligibility. To illustrate, to design algos to distribute loans to minority groups. Here database is made of artificial people from minority groups with average credit rating. Thus they are closer to other groups. It is akin to manipulating data that algos are trained on. It facilitates positive discrimination.

The use of fake data is a step in the right direction. However, synthetic data would not eliminate the bias 100 per cent, since such bias is innate in the people who develop these tools.