Diversity amongst AI community: a key to eliminating AI bias


The AI world is in a state of flux. Its revolution has been underway for decades, but recent breakthroughs have led to unforgiven mistakes- perpetuating bias and inequality. Many may be wondering about how we can prevent this? A key to this is diversity in the AI community which has been historically lacking.

AI is a powerful tool, but it has its limitations

It is a matter of fact that AI is not an independent technology. It enables tech devices to translate and adapt to certain environments. And AI does this by using historical data for training that reflects human actions in our society. Oftentimes, this data is coded with historic prejudice. So, as we improve and progress, AI needs to as well, otherwise, it repeats historic injustices.

Gender gap analysis in the AI industry by sectors

Speaking of which, the most acclaimed case is that the AI industry is dominated by males, as more than 80% of AI professors are men. Or, that women currently makeup 24.4% of the AI workforce, and receive median salaries that are only 66% of the salaries of their male counterparts. The numbers and analysis go on and on, see picture on the left for more.

These inequalities grow even more if we take into the picture other under-represented communities and/or attributes such as race, sexual orientation, background, education, ethnicity, social status, etc.
Different initiatives and organizations are being pressured to work on exposing these limitations that pose serious risks to society and are done by irresponsible AI development.

For instance, ProPublica, an independent, non-profit newsroom that produces investigative journalism in the public interest, uncovers the case of software that predicts future criminals and it’s biased against black people. Or, the initiative called “Advance AI” provides resources and support to help create a more diverse community within the AI field. It was formed as a response to some high-profile incidents where algorithms appeared biased against certain groups based on their data sets or training processes — such as when Microsoft’s Tay chatbot went rogue after being exposed to racist tweets on Twitter.

There are many uprising examples that work on true change and provide in-depth solutions to the lack of representation across the AI industry, and one worth mentioning is the Responsible AI Institute that uncovers facts and figures such as “LGBTQ couples were 73 percent more likely to be denied a mortgage than heterosexual couples with comparable financial credentials.”

One could think that the creators of AI need to think critically about the biases they are introducing into their algorithms. But this is not easy as you read through it, right? However, a good starting point is that all the above-mentioned cases have one thing in common that could be the possible solution- diversity among the AI community.

Diversity matters, but why?


The first reason why diversity matters is that studies have shown again and again that lack of diversity leads to a limited perspective and can result in bias that may be difficult to detect and correct. It’s been shown that when you introduce a diverse group of people into an environment with biased systems, they can often identify those biases and remove them earlier in the process before they become problematic. Or worse, before they become public. Another reason is that diversity is and should be multidimensional- there needs to be diversity among hiring managers who make decisions about which new employees get hired or promoted, among those who are closely working on building AI products for the future, between the researchers, policy decision-makers, executives. You know the drill. Only like this, we will reach the full potential of transforming the narrative where different groups are being negatively impacted only because they don’t fit into the “unwritten society’s mold”.

Diversity matters as it provides holistic approaches in making AI technologies more responsible. It helps address challenges faster and clearer as local knowledge and front-line experience will be embedded in the core of every decision-making or working process. Also, with diversity, unexpected ally ships can be built to accelerate learning and research that will do magic for the better world we want to create tomorrow.

Diversity is not a goal, but rather a way of doing and working

Lately, we have seen a lot of PR statements where companies promise to be more gender-inclusive or support movements that fight for social justice and human rights. But, they fail to hold on to and continue doing ‘business as usual’ as soon as the storm is over. Achieving true diversity is an ongoing process that is initially started with the organization’s true values and goals and soon it grows into its DNA. Diversity, inclusion, and responsibility must not be only performative actions, but rather deliberate work towards understanding different points of experience and creating better practices.

There are many ways to combat AI bias, but diversity in the tech community is one of the most important. This article highlighted the aspect of why building responsible AI systems that will detect, mitigate and handle AI biases and risks is a team effort. We know that for any given task or problem, there will be a range of possible solutions and approaches. The more diverse perspectives we have on these problems, the better our chance at finding efficient approaches and quality solutions.