The Need for Diversity in AI Development

Is AI teaching machines to think, or is there more to it? AI is a powerful advancement in technology, as it has begun to automate tasks that until now have been done by a human workforce. From performing certain functions without memory, which is the most basic form of AI, to self-aware machines, a step even further than decision-making AI.

However, the idea of machines deciding or supporting us in our decisions has triggered a discussion based on fear. What makes AI a threat to human decision-making if we are the ones creating it?

Steps to designing an AI system

Designing AI starts by identifying the problem. Every application of AI has the intention to solve a certain need. A human need. And people are at the core of the “why”.

After identifying the problem, data is gathered and prepared, selected algorithms are trained and a particular programming language is chosen to run the AI system on a certain platform. In this process, various decisions have to be made that impact how AI will teach each machine to think. For instance, the base data can include biased human decisions and reflect inequalities. It’s therefore not sufficient to simply start off with readily available data without reflecting on how it was compiled. Ethics lie at the core of any AI related system.

We have to dig deeper and analyse where these data come from, who or what they represent, or more specifically, who they don’t.

What if AI is biased?

This takes us back to our first question: why do we fear AI, if it’s merely based on data we gather? Because we know these data are already biased.

Diversity in the workforce, in venture capital investments, and in our education all play an important role. If we don’t pay attention to where our AI data is coming from, bias creeps in. A study by digitalundivided found that since 2009, black women have raised less than 1% of the total $424.7B in venture capital for tech companies. Amazon, FedEx, Target, and Capital One have already tested or used AI hiring software, and the results were quite unfair to particular populations. Amazon even had to abandon their AI recruitment project because the system showed a bias against women. Not surprisingly, it was built upon data of male-dominated CVs.

Even schools are sometimes guilty of digital discrimination. As long ago as 1988, a British medical school designed a computer program to determine which applicants would be invited for interviews, and this early form of AI was already biased against women and applicants with non-European names. Basing ourselves on these and other examples, should we fear AI or should we re-examine our way of gathering the data for machines to use?

Eliminating bias in AI

There’s a lot we can do to avoid bias in AI.

AI needs to be designed so it can be audited. When bias is revealed, it should be possible to remove it. That’s why educating ourselves on new insights in fairness, transparency and privacy open up our view on how to collect more diverse data. And last but not least, we need to take a closer look at the AI industry: a diverse AI talent pipeline will uncover the many benefits of a fair AI structure for business, economy, education and other societal challenges.

If AI is not a reflection of our diverse society, it might be something to fear. But more importantly, it’s a missed opportunity. Women alone control over $20 trillion in consumer spending globally. Imagine the massive benefits that could be gained by investing in women’s business ideas. Moreover, a 2019 Green Park report showed that only ten black, Asian and minority ethnic people are working in leadership roles across companies in the FTSE 100 (Financial Times Stock Exchange Index). Increasing diversity in these leading business roles not only has the potential to help tackle the AI bias, but, as a recent Boston Consulting Group study found, a diverse management team also increases a company’s revenue.

A huge opportunity for diversity and application of AI to enhance innovation arises once we step away from the discussion about whether or not we have to fear machines. I believe diversity in new technology development will enrich the solutions for our societal challenges.