Marginalized Americans are highly skeptical of artificial intelligence, research finds

Artificial intelligence can be marketed as the great equalizer of society – transforming businesses, rationalizing work and facilitating life for all, but for many marginalized Americans, AI does not feel like a promise.

It looks like a threat.

A new study by the University of Michigan reveals that gender minorities, women and people with disabilities – in particular those who are neurodivergente or who live with mental health problems – have much more negative attitudes towards AI than their peers of majority groups.

Using data from a national representative survey with 742 people, researchers asked participants to report their attitudes concerning the impact of AI on their lives and work. The responses reveal a strong positivity fracture towards AI through the identity lines.

“The AI ​​may be everywhere, but it’s not for everyone – at least not yet,” said the main author Oliver Haimson, assistant professor at the Institut d’Um School of Information and Digital Studies. “If we continue to ignore the prospects of marginalized people, we risk building a future propelled by AI which deepens inequalities rather than reducing them.”

Key results:

• Non -binary and transgender participants reported the most negative views of AI overall.
• Women were significantly less likely than men to see AI positively.
• Handicapped participants, especially those who are neurodivergent or who have mental health problems, also expressed negative AI attitudes.
• Surprisingly, black participants had more positive views of AI than white participants, suggesting a more nuanced relationship with technology between people of color.

The study highlights real damage which can fuel the negative perceptions of AI:

• Facial recognition software that hurts mines in sight or in trouble with trans and non -binary people – often during their surveillance.
• Predictive police algorithms that strengthen racial prejudices and lead to unfair arrests.
• Health care systems based on unable IA models thinking of people with disabilities.

“These are not abstract concerns,” said Haimson. “People who are wary of AI have often experienced experiences with systems that identify them, exclude them or hurt them because of their sex or their handicap.”

Research questions the dominant account of AI as neutral or universally beneficial tool. He urges technologists, businesses and decision -makers to take a break and ask for: who works this technology and who is left behind?

The study co-authors are doctoral student Samuel Reiji Mayworm, researcher Alexis Shore Ingber and Deputy Professor Nazanin Andalibi.

Research was presented during the recent ACM conference in 2025 on equity, responsibility and transparency (FACCT ’25).