Tech's diversity crisis is baking bias into AI systems

• Women represent 26% of AI workforce
• Blacks, Hispanics also underrepresented in AI workforce
• Built-in viewpoints can make use of digital tool risky

CHICAGO - As an Afro-Latina woman with degrees in computer and electrical engineering, Maya De Los Santos hopes to buck a trend by forging a career in AI, a field dominated by white men.

AI needs her, experts and observers say.

Built-in viewpoints and bias, unintentionally imbued by its creators, can make the fast-growing digital tool risky as it is used to make significant decisions in areas such as hiring processes, health care, finance and law enforcement, they warn.

"I'm interested in a career in AI because I want to ensure that marginalised communities are protected from and informed on the dangers and risks of AI and also understand how they can benefit from it," said De Los Santos, a first-generation U.S. college student.

"This unfairness and prejudice that exists in society is being replicated in the AI brought into very high stakes scenarios and environment, and it's being trusted, without more critical thinking."

Women represent 26% of the AI workforce, according to a UNESCO report, and men hold 80% of tenured faculty positions at university AI departments globally.

Blacks and Hispanics also are underrepresented in the AI workforce, a 2022 census data analysis by Georgetown University showed.

Among AI technical occupations, Hispanics held about 9% of jobs, compared with holding more than 18% of U.S. jobs overall, it said. Black workers held about 8% of the technical AI jobs, compared with holding nearly 12% of U.S. jobs overall, it said.

AI bias

De Los Santos will soon begin a PhD programme in human computer interaction at Brown University in Providence, Rhode Island.

She said she wants to learn not only how to educate marginalised communities on AI technology but to understand privacy issues and AI bias, also called algorithm or machine learning bias, that produces results that reflect and perpetuate societal biases.

Bias has unintentionally seeped into some AI systems as software engineers, for example, who are creating problem-solving techniques integrate their own perspectives and often-limited data sets.

Amazon.com scrapped an AI recruiting tool when it found it was selecting resumes favouring men over women. The system had been trained to vet applicants by observing patterns in resumes submitted to the company over a 10-year period.

Most came from men, a reflection of a preponderance of men across the industry, and the system in effect taught itself that male candidates were preferable.

“When people from a broader range of life experiences, identities and backgrounds help shape AI, they're more likely to identify different needs, ask different questions and apply AI in new ways," said Tess Posner, founding CEO of AI4ALL, a non-profit working to develop an inclusive pipeline of AI professionals.

“Inclusion makes the solutions created by AI more relevant to more people," said Posner.

Promoting diversity

AI4ALL counts De Los Santos as one of the 7,500 students it has helped navigate the barriers to getting a job in AI since 2015.

By targeting historically underrepresented groups, the non-profit aims to diversify the AI workforce.

AI engineer jobs are one of the fastest growing positions globally and the fastest growing overall in the U.S. and the United Kingdom, according to LinkedIn.

Posner said promoting diversity means starting early in education by expanding access to computer science classes for children.

About 60% of public high schools offer such classes with Blacks, Hispanics and Native Americans less likely to have access.

Ensuring that students from underrepresented groups know about AI as a potential career, creating internships and aligning them with mentors is critical, she said,

Efforts to make AI more representative of American society are colliding with President Donald Trump's backlash against Diversity, Equity & Inclusion (DEI) programs in the federal government, higher education and corporate levels.

DEI offices and programmes in the U.S. government have been terminated and federal contractors banned from using affirmative action in hiring. Companies from Goldman Sachs to PepsiCo have halted or cut back diversity programmes.

Safiya Noble, a professor at the University of California Los Angeles and founder of the Center on Resilience & Digital Justice, said she worries the government's attack on DEI will undermine efforts to create opportunities in AI for marginalised groups.

"One of the ways to repress any type of progress on civil rights is to make the allegation that tech and social media companies have been too available to the messages of civil rights and human rights," said Noble.

"You see the evidence with their backlash against movements like Blacks Lives Matter and allegations of anti-conservative bias," she said.

Globally, from 2021 to 2024, UNESCO says the number of women working in AI increased by just 4 percent.

While progress may be slow, Posner said she is optimistic.

"There's been a lot of commitment to these values of inclusion,” she said.

“I don't think that's changed, even if as a society, we are wrestling with what inclusion really means and how to do that across the board."

(Reporting by Natasha Ghoneim; Editing by Anastasia Moloney and Ellen Wulfhorst.)