Artificial intelligence: How to avoid racist algorithms

Diversity is increasingly becoming a concern among everyone and it has now extended, surprising as it may seem, algorithms. Developers claim that algorithms, while searching for things, evidently churn up things that are more easily visible and more often than not, such things are against diversity, reinforcing the dominant paradigm. In short, what you get is a biased decision, biased result and biased opinion on everything under the sun. Something as innocuous as hand, when searched on Google, only yields white hand. You can repeat the search with many other things and still, it will yield white bias in its results.

Hence, not only do content creators need to include alternative options to balance this strange difference, but also needs to be picked up by the algorithms that search the content in search engine. Google, however, claims that the results shown only show the majority of the content instead of the supposed values. So, basically Google pushes the responsibility back to the content developers, starting a never-ending loop. However, AI is trying to break this loop by creating more racially diverse algorithms. Sounds impossible? You will be surprised to know what AI can do.

How AI impacts the scene?

The formation of Algorithmic Justice League or AJL has been the mainstay of this project. It took shape when in a strange turn of event, an MIT student’s dark skin could not be recognized by facial recognition. Only when a white mask was used could the student be identified. Such a strangely reductive algorithm was immediately a point of interest for the student. The problem has been persistent and the student believes many black people have had the same experience. Her work, from that point of time, has been nothing sort of revolutionary as teachers are showing her work to students.

Her effort has opened the eye of quite a few researchers as they have realized their flaws and this new corrective measure seems to be the perfect solution. Recently, a researcher took the cue to create an algorithm for detecting skin cancer traits especially on coloured skin. However, the level of diversity does not limit itself to skin color but extends to age too as the color transforms with age. So, if you are using AI, it is important to ask, is the AI inclusive enough for you to be using it or believing its results?

Diversifying the data

Internet giants like Google, Facebook and IT giants like Microsoft are increasingly paying attention towards diversity at work place which is surely a big change. Of course, the more diverse the workplace, the more the options for testing the algorithm and the greater will be the inclusivity of it. In fact, people have biased idea of beauty where facial structure is secondary to skin colour. So, color bias is real and needs rectification in more than one ways. Beyond facial recognition, even criminal data shows color bias towards blacks and identifies them as riskier than the fellow whites.

In short, the skewness in data has manifold implications and unwanted manifestations. An effort is being made to curb the issue so that such problems are addressed content wise and then possible solutions are sought. Experts believe that developers, content managers and IT giants should work together to diversify data set and include certain metadata to ensure diversity in search results. Algorithms should know what they are setting out to do instead of random searching based on majority. Breaking the white myths and white dominance over the internet content may take time, but it will surely be a reality in the years to come.