The Unseen Biases Lurking in Generative AI and How They Could Affect You

Author(s): Piyoosh Rai

Originally published on Towards AI.

Picture this: you ask an AI to create an image of a CEO. It gives you a clean-cut white man in a suit. Ask for a “businesswoman,” and you get a smiling, conventionally attractive young woman, often white.

This is not a coincidence. It’s a reflection of something deeper. Generative AI systems, while powerful, often carry forward the same biases society has struggled with for decades. And they can quietly shape how we see the world, and how the world sees us.

What Is Generative AI Bias?

Generative AI systems, which include tools like chatbots, image creators, and text generators, are trained on massive datasets pulled from the internet. These datasets contain all the content we, as a society, have created, complete with stereotypes, inequality, and social imbalances.

When these systems learn from biased data, they do not just replicate those biases. They amplify them, often invisibly and at scale.

How Does Bias Enter the System?

Bias can creep into AI models in several ways, even without direct intent:

• Skewed training data: If certain groups are overrepresented, like white men in leadership roles, the AI begins to assume this is the default.
• Algorithm design flaws: Developers may unknowingly build in assumptions that influence how the model processes information.
• Feedback loops: Once biased outputs are released, they can be reused as training data, reinforcing the same stereotypes again and again.

Real Examples of AI Bias

This issue is not abstract. It shows up in concrete and measurable ways.

• Gender stereotypes: UNESCO reported in 2024 that AI systems associate women with home and family four times more often than men. Meanwhile, men are linked to career and leadership roles.
• Racial representation: When asked to generate images of CEOs or financial analysts, AI overwhelmingly depicts white men, despite the real workforce being far more diverse.
• Subtle image patterns: AI-generated visuals often show women as younger and more cheerful, while men appear older or more serious. These subtle portrayals can reinforce assumptions about authority, professionalism, and competence.
• Sector-specific consequences: In hiring, healthcare, law enforcement, and financial services, biased algorithms have been shown to produce unfair outcomes. Examples include higher loan rejection rates, misdiagnoses, and discriminatory hiring practices.

A Closer Look: AI Bias in Hiring

Applicant Tracking Systems (ATS) are a key area where AI bias can have life-altering consequences. Many companies now use AI-powered video interviews to screen applicants. These tools assess not only what candidates say but also how they say it, evaluating tone, facial expressions, and accents.

Research shows that these systems often misjudge women, people with non-standard accents, and individuals from underrepresented backgrounds. A Harvard Business Review study found that voice recognition in ATS platforms tended to undervalue responses from women and non-native English speakers.

In 2025, a formal complaint was filed against Intuit and HireVue. Deaf and Indigenous candidates were reportedly disadvantaged because the AI failed to understand their speech or communication styles. Despite strong performance on the job, many of these candidates received poor video interview scores, leading to missed promotions and lower-visibility roles.

These systems are not just making suggestions. They are filtering people out before a human ever reviews their applications. Studies suggest that candidates from marginalized backgrounds are 1.5 times more likely to be unfairly excluded when AI tools are used for hiring.

When AI Bias Influences Human Behavior

One of the most alarming effects of biased AI is its influence on people. It turns out that interacting with biased AI can actually make users more biased themselves.

Professor Tali Sharot at University College London has shown that exposure to AI-generated stereotypes increases the likelihood of people adopting those same stereotypes. The AI does not just reflect society — it can change it, often in the wrong direction.

This creates a dangerous loop. AI systems reflect societal bias, influence users, and those users feed new data back into the system. The result is a slow but steady amplification of inequality.

Why Everyone Should Care

Even if you are not building AI, you are likely being affected by it. AI is involved in decisions that shape your life, whether you realize it or not.

• It screens resumes.
• It determines which ads you see.
• It influences what news is shown in your feed.
• It plays a role in loan approvals, insurance claims, and job interviews.

In short, biased AI systems can influence your opportunities, your reputation, and your access to resources. Ignoring the problem is no longer an option.

What You Can Do

Solving bias in AI is not simple, but progress is possible. Here are a few ways to help:

• Ask for transparency: Companies should be clear about how their AI systems are trained, what data they use, and how often they are audited for fairness.
• Support diverse teams: When AI is built by teams with a range of backgrounds and perspectives, it is more likely to be fair and inclusive.
• Stay critical: Be skeptical of AI-generated content, especially when it reinforces familiar stereotypes. Teach others to do the same.
• Push for policy: Advocate for regulations that require bias audits, accountability mechanisms, and ethical AI design.

A Shared Responsibility

Generative AI is a powerful tool, but it is not neutral. The biases hidden inside these systems reflect real-world inequalities that can be automated and scaled if left unchecked.

This is not just a technical problem. It is a social one. Addressing it requires input from developers, regulators, business leaders, and everyday users. Everyone has a role to play in building systems that are accurate, fair, and inclusive.

The next time you interact with AI, ask yourself: who benefits from this output, and who might be left out?

If this article sparked a thought or hit close to home, leave a comment. Have you experienced or noticed bias in AI-generated content or hiring systems? How do you think we can build a better future?

Let’s make this a conversation worth having.

Sources: Harvard Business Review, UNESCO, MIT Technology Review, ACLU, TestGorilla, Stryve Online

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI