Lack of diversity in AI development: ‘Often male, white and highly educated’ | Tech

Virtual assistants such as Siri and Alexa often have the appearance or voice of a woman by default. Researchers wrote in 2021 that many people find it easier or more pleasant to give commands to a female character than to a male version. But in doing so, AI assistants subscribe to the stereotype that women are obliging and submissive.

That is not the only (unintended) consequence of the development of responsible AI, where ethical risks must be removed as much as possible. For example, facial recognition sometimes works less well for women than for men and algorithms are less likely to match women to well-paid jobs.

That is why there must be more awareness of diversity when developing AI, concludes Maaike Harbers. As a lecturer in AI & Society at Rotterdam University of Applied Sciences, she completed an exploratory study on this subject.

Much research has already been done into possible discriminatory factors in artificial intelligence and the datasets with which the technology is trained. Much less is known about the relationship between (gender) diversity of people developing AI and how responsible that AI is. This information is certainly lacking in the Netherlands.

Research by Harbers shows that we often look at the issue of responsible AI through a technical lens. “The people developing artificial intelligence are often male, white and highly educated,” she says. “We look at AI from that angle and that is also necessary, but we also have to pay attention to other things.”

Little attention is paid to the female perspective. For example, women in the field indicate that they are often not taken seriously. Harbers thinks it is worrying: “It could lead to women dropping out and leaving the sector, while there are already major staff shortages in the IT world.”

Mariëlle Stoelinga, professor of risk management of high-tech systems at the University of Twente and Radboud University, calls the findings problematic. And she endorses them too. “It is a typical man’s world. And if you belong to a minority, you are quickly taken less seriously. That is unfortunately the case in many fields.”

The Central Bureau of Statistics has already calculated that the vast majority of ICT workers in the Netherlands are male. Only 16.2 percent of IT professionals are women. That is even below the European average, which at 19 percent is not very high.

Women face hurdles and prejudices in the technical world from a young age. This is what Vivian de Corti of VHTO, the expertise center for hair technical development, says. “If students receive study advice for a technical or exact profile, there is a chance that unconscious prejudices play a role. In the same situation, a boy has twice as much chance of receiving positive advice than a girl.”

And that is a shame, because women can enrich that man’s world. For example, they deal with ethical AI issues in a different way, says Harbers. They are often more reluctant to use new applications than men. And they often pay more attention to risks, such as the negative impact on the environment and the consequences for specific groups in society. “It may be more noticeable if you belong to the group that could potentially be disadvantaged by AI.”

More awareness is needed to ensure that not only more women, but also different ethnicities are involved in AI development. That change does take time.

“We are actually already too late with that,” says Harbers. Because many AI datasets are already large and are used by many companies. But she looks at it optimistically. “With this first study we show where the bottlenecks are. The risk of discrimination is high, so it is extra important that more attention is paid to diversity.”

The report therefore includes a number of recommendations. For example, Harbers and her colleagues advise companies and organizations to comply with ethical guidelines for responsible AI.

According to Stoelinga, it is just as important that companies first consider what they want to use AI for. “AI is a revolution that will determine our future,” she says. But that is why it is also important that not only different genders, but also different ethnicities, for example, work on it. “Otherwise you get solutions that work for a certain group, but not for the whole of society.”

This involves some awareness, De Corti agrees. “The women want to, but they continue to experience that they have to prove themselves twice as hard.”