Source: Deep Learning on Medium
The terms like AI and Machine Learning has been around for decades, but they have never been seen so much in mainstream of technical world. The advancement of AI has overtaken every industry and every industry needs it, one way or another.
So basically, we provide data to AI for its functionality and based on that it learns and takes decisions. But wait, doesn’t that mean if we provide biased data it will take biased decision? Of course it will, and there are a few philanthropist that are doing research to study algorithms that are swayed by biased data. Not to mentions but be informed that there is a difference between wrong data and biased data.
MIT technology review editor, Will Knight, wrote in an article, that biased algorithms will have more impact on poor communities and minorities. The reason behind this could partially be understood by the fact that poor people are more dependent on country and government itself, and that same government is not willing to inspect or even care about those discrepancies happening in the world of Artificial Intelligence, wrote Will Knights (Will Knights, 2017).
For more clarity, we can take the example of Silicon Valley where people see men as more talented than women, in fact — strange as it may sounds — so does AI. Apparently, datasets seem to give more weightage to men if a word like “programmer” appears and gives weightage to women if a word like “homemaker” appears and the process of doing so is called word embedding (Will Knights, 2016). However, curing those data sets is quite possible and researchers are doing that in order to not let AI take biased decision.
But the irony is, changing the data in order to not let it take biased decision is also a kind of biasing, because then it will not depict the real picture of the world claims Arvind Narayanan, an assistant professor of computer science at Princeton. If you have read about the concept of overfitting, this is not very different from it, you just want your model to fit with total exactness and not the way it supposed to fit. So, the question boils down to — should we change data or let the machine take biased decisions? Well, I think AI will still be a hot topic for some more years to come and we hope to see the answer unfold.
Knight, W. (July 12, 2016). Biased Algorithms Are Everywhere, and No One Seems to Care.MITTECHNOLOGYREVIEW. https://www.technologyreview.com/s/608248/biased-algorithms-are-everywhere-and-no-one-seems-to-care/.
Knight, W. (November 23, 2016). How to Fix Silicon Valley’s Sexist Algorithms. MITTECHNOLOGYREVIEW.https://www.technologyreview.com/s/602950/how-to-fix-silicon-valleys-sexist-algorithms/.