A new algorithm trains AI to shake off its biases

Jan 30, 2019, 7:08 AM EST
(Source: Eddy Milfort/flickr)
(Source: Eddy Milfort/flickr)

A stubborn fact with artificial intelligence algorithms is that they are governed by the same ingrained prejudices that foul most human minds. Of late, a lot of research is being invested in ridding AI programs of their biases, and the latest tool by engineers from MIT and Harvard University is a step towards achieving this.

The tool, unveiled at the Conference on Artificial Intelligence, Ethics and Society, automates the process of debiasing an algorithm through a series of audits and corrections, writes Futurism.

While it cannot be ruled out that the training algorithm may itself carry traces of implicit biases, but the approach deserves an applause, as it marks an actionable step towards weeding out algorithmic biases.

The use of an algorithm to erase bias from AI systems is a simplistic approach though and the one that attempts to take the responsibility off the shoulders of human developers. Instead, there are steps that would fix this problem, which include a systemic change in the way AI programs are coded and developing AI systems to identify biased data, notes Forbes.