How AI can help overcome gender bias in science

DO YOU play Go?

Oddly enough, the answer will almost certainly give away the gender of the person responding. The Go community, much like those of certain scientific disciplines and academic programmes, is very much male-dominated.

Social and cultural factors are contributing to this gender imbalance. There’s strong evidence to suggest that the way the game is passed down from generation to generation – just like the way science is taught in school – plays a big part in keeping this inequality alive.

In 2022, researchers from the University of Hong Kong compared two methods of teaching Go: by human teachers and by an artificial intelligence (AI)-powered character. The main outcomes measured were tournament-level winning probability, average move quality during gameplay and the number of errors made.

While the students started on an equal footing, the AI-trained group achieved better outcomes than the other.

Importantly, there was no gender difference in outcomes in the former group. However, the gender gap – with males faring better than females – persisted in the human-taught group.

Why this disparity?

According to the researchers, certain stereotypes held by these teachers, often unconsciously, could have influenced their teaching – for example, engaging with or encouraging boys more than girls – and sent signals that the students picked up and amplified.

When the girls were trained by an unbiased AI and had no interaction with their classmates, they were not influenced by such social cues.

While this study focuses on the game of Go, I’m firmly convinced that AI’s potential to address gender bias in the broader world of scientific education warrants further investigation.

Empowering women to pursue Stem fields

The instructional value of AI is a complex subject that sociologists and education experts will debate for many years to come. From outright rejection to a full embrace, opinions on the technology’s role in education differ tremendously.

While efforts continue in trialling and improving the use of AI to provide effective, personalised supervision to students, we need to ensure that the models themselves are not poisoned by gender bias.

A number of recent cases, in particular when AI was used in recruitment, have shown that the technology can reinforce or amplify discriminatory practices that persist in human-led hiring processes.

More work is urgently needed in this area.

According to statistics from the United Nations Educational, Scientific and Cultural Organization, young women make up just 35 per cent of the science, technology, engineering, and mathematics (Stem) student population worldwide.

At a time when huge scientific and engineering challenges are confronting humanity, the world continues to deprive itself of a significant proportion of talent.

Industrial and technological players cannot remain mere spectators in these debates. By designing AI systems that will be integrated into sensitive environments – such as education, training and recruitment – they bear a particular responsibility not to mechanically reproduce existing imbalances, but to systematically explore ways to reduce them.

To empower more female students to pursue careers in Stem, some countries, such as Singapore, are making headway.

A 2024 joint study by the Infocomm Media Development Authority and Boston Consulting Group found that women make up around 40 per cent of Singapore’s technology workforce, outperforming the global average of 28 per cent, and placing the country at the forefront of South-east Asia in promoting gender diversity and inclusion in the tech industry.

This progress did not happen by chance; it reflects years of sustained investment in education, skills and national initiatives such as SG Women in Tech, which aims to attract, retain and develop women tech talent in the country.

This commitment is reflected at Thales, too, where about 40 per cent of our employees in Singapore and 28 per cent globally are women. Leaders, like Tan Dunlin, research and technology director at Thales, normalise female representation in traditionally male-dominated fields such as Stem.

“Who we see shaping technology and society matters. Diverse and visible leadership helps challenge bias and build systems that reflect the societies they serve,” says Tan, who plays a leading role in advancing research, development and deployment of trusted AI for critical systems, as part of the firm’s AI accelerator.

But representation alone is insufficient. AI does not create inequality on its own; it often amplifies patterns that already exist.

The real question for organisations is not just who builds AI systems, but who gets to define the problems, participate in pilot projects, and receive visibility and credit. Intentional and inclusive participation determines whether opportunity expands or narrows.

The excitement surrounding AI is generating a multitude of potential use cases. Some of these may turn out inappropriate or unsustainable. But in the use cases that can yield real impact, we must do what we can to ensure human biases do not perpetuate through AI.

We must not deprive science of essential talent.

The writer is chairman and chief executive officer of Thales Group