Study: Robots can learn bias from each other

Sep 07, 2018, 8:07 AM EDT
(Source: Jiuguang Wang/flickr)
(Source: Jiuguang Wang/flickr)

You are up for disappointment had you been fancying AI robots to be saintly in their social attitudes and behavior. A research at the MIT and Cardiff University suggests that artificial intelligence robots can pass on their racial and gender prejudice to one another.

In a simulated game, AI robots developed their own biased new strategies by “identifying, copying and learning this behavior from one another,” reports Engadget. The researchers observed an irreversible trend, as shielding a small population from prejudicial groups led to the formation of further small prejudicial groups within.

The observations of the study bring us to a crucial question – will robots of the future be ingrained with human biases or is there a way to rid them of the flaw. The solution lies in the rigorous testing and training of machines on constant streams of unbiased data, unbiased classification and unbiased samples and whittling down the aberrations, notes Forbes.