How AI reflects and amplifies human bias

On International Women’s Day, we fight for a world of equality

On International Women’s Day, we reflect on the cultural, social and social achievements of women. And believe me, there is a lot of girl power in the world today. But at the same time it feels double. Why does International Women’s Day get so much attention?

The answer is simple, the solution complex. To date, there is still a huge gap between men and women. On a professional level alone, the numbers are painful: On average, women still earn less than men, hold fewer senior management positions than men, and receive fewer educational and career opportunities than men. And in the world of technology, they are clearly in the minority. The list is endless, the figures are very moving and the concrete examples are too frequent in our -what I thought- civilized society.

The dream of many, and certainly mine, is to evolve towards a world where there is no distinction between men, women or any gender identity. Let me be clear: being offered a job because you are a woman, and not because you have the best roles, is not the solution!

Equality means that gender in no way affects the decisions we make

I have been active in the field of Artificial Intelligence (or data mining as we called it ten years ago) for about ten years. A topic that has been receiving a lot of attention lately is the development and implementation of responsible AI, the responsible use of AI. With the proposal of the new European AI Law, it becomes even more important to develop an AI that is not only accurate and effective, but also fair and impartial towards minority groups.

An important aspect of Artificial Intelligence is that the intelligence that teaches itself is based on data. Data that we all contribute to and that we all create and produce.

The often unspoken social bias can be translated into explicit bias in AI models and systems.

This often leads to controversial AI systems. Think of the AI-based recruiting system that Amazon created in 2014 to select the right candidate for a job faster and more efficiently. It soon became clear that the AI ​​system favored male candidates and that female candidates had almost no chance of being hired. A 2015 study found that women were less likely to get ads on Google for high-paying jobs. That same tech giant was the subject of an investigation for discrimination in its search results. Google image searches for terms like CEO or professor returned markedly fewer results with women than with men. Further investigation of the search results revealed that the first image of a female CEO was an image of the Barbie toy doll. In 2019, Apple admitted that the AI ​​system typically raised its male customers’ credit card limits higher than those of its female customers despite having the same credit scores.

Although the results are often impressive, we must be grateful to artificial intelligence systems for making the real situation tangible. Controversial AI systems are based on mapping patterns in data. And again, this data is data that we build and produce ourselves. Amazon’s hiring system was based on the analysis of ten years of hiring processes carried out by human recruiters. Google’s ads and search results are based on the clicking behavior of you and others. Learn and classify material published online. If no female CEO photos are published, the algorithm will not be able to return female CEO photos. Apple’s credit card management algorithm learns its knowledge from historical data and customer behavior.

So there is still a lot of work to be done before all prejudice completely disappears from our world. Before condemning AI systems as discriminating demons, we must first look at ourselves and reflect on how such AI systems came to be. Only through good cooperation between human and artificial intelligence can we make our world a little better. And a world of equality and without prejudice, at all levels, that’s why we fight on International Women’s Day.

This is a contribution from: Véronique Van Vlasselaer, AI and Analytics Leader at SAS.

Leave a Comment