Editorial
Artificial intelligence systems are trained to do incredible things like translate languages, recommend movies, summarize documents and even simulate conversations. But here is the catch: everything AI knows, it learned from us. That means if our history is biased, our data is biased and so is the AI. Many people believe that AI makes “objective” or “rational” decisions, but it reflects the patterns, preferences and problems of the people and systems that built it.
Before exploring how bias enters AI, it helps to understand where AI came from. Artificial Intelligence is not a single technology, but a broad concept that refers to machines acting in ways that resemble human intelligence.
Machine learning (ML), a key part of AI, began in the 1950s. ML algorithms learn from data to perform tasks without explicit instructions. In the 2010s, deep learning emerged, using neural networks that simulate how the human brain processes information. Shortly after came generative AI, which generates new content from what it has learned. Today, we are entering the age of explainable AI, where models attempt to clarify their reasoning to human users. On the horizon is agentic AI, which is capable of decision-making with limited human input. It can adapt and learn from its own outputs, something current systems do not do autonomously.
Milestones That Made Today’s AI Possible
Why does this matter?
It matters because the Western, Eurocentric and male-dominated perspectives that shaped machine learning models in the 1950s, 60s and 70s are still embedded in today’s tools. Imagine if that one family member or coworker — everyone knows the type — oversaw selecting, cleaning and validating all the data used to guide your daily choices. You can probably guess how that would turn out. In many ways, that is exactly who shaped the foundation of early AI.
Related Article: Colonialism in Code: Why AI Models Speak the Language of Empire
The First Domino: How Initial Choices Tilt the System
Consider a simple machine learning model (see graphic). Every point where a human is involved, from selecting data to writing the algorithm to adjusting its settings, is a place where bias can enter.
If a university builds an AI tool using 20 years of admissions data from a time when its student body was 85% white and male, the AI will likely reflect that skew. Similarly, if the same algorithm is trained using data from a predominantly Muslim company, it may misrepresent or fail to accommodate other populations. AI does not create new norms or values, but it recycles what it is given. It keeps perpetuating the same issue, again and again.
What AI Eats Determines What It Thinks
AI systems, especially those that process language, are trained on vast amounts of internet data. This includes books, websites, social media and news articles. Most of this content comes from English-speaking, Western countries, meaning AI is learning from a narrow band of perspectives.
AI must be taught. Humans choose the training data, decide which problems AI should solve and define what “success” looks like. If a hiring algorithm is trained on 20 years of resumes, and most were from men, it may automatically prioritize male candidates, even if unintentionally.
Once the AI learns patterns, it repeats them. A now-retired hiring algorithm used by Amazon offers a telling example: it began downgrading resumes that included the word "women's," as in "women’s chess club" or "women’s college." This was not because the algorithm was designed to be discriminatory, but because it was trained on ten years of hiring data that skewed heavily male. The model learned that resumes with male-oriented language were historically favored and, in turn, encoded that pattern into its decision-making logic.
Everyday Bias in the Tools We Use
Bias is everywhere AI is used: in home devices, healthcare, education, travel, defense, finance and more. Consider AI assistants like Alexa and Siri: most use female voices and are designed to manage household reminders, calendars and lists. They are “helpful” in roles traditionally assigned to women, but without the ability to express emotion or push back no matter how many times a child asks them to sing “Happy Birthday.”
IBM’s Watson for Oncology performed poorly in non-Western hospitals because it was trained on American data. Women of color have experienced higher rates of forced cesarean sections due to models trained on majority-white populations. ChatGPT frequently generates images of white men in suits when asked to visualize a professor or IT professional. Ask it for “women’s professional attire,” and it often returns pencil skirts and blouses, rarely including culturally grounded alternatives like sarees or hijabs.
These examples are not random. Standards of “professionalism” have long been defined by Eurocentric norms, and AI reflects those defaults. Hairstyles, clothing, accents and cultural expressions that fall outside those norms are often labeled “unprofessional” or excluded from the training data altogether.
AI Bias Is No Mistake – It's Built In
Many people assume that AI bias is a glitch, something that can be fixed with better code. But the deeper issue lies in the data and in the human choices behind it: who selected it, who cleaned it, who labeled it and who decided it was “ready.” Bias in AI is not simply an error; it is an inheritance, like a belief system passed quietly from generation to generation. These systems reflect historical power structures: who had a voice, who set the standards and who built the institutions that still shape our world today.
AI is trained on culture, but only a portion of it. The models amplify the perspectives that have the most data and digital visibility. When non-Western languages are excluded, misrepresented or flattened into Western frameworks, AI becomes a tool that reinforces inequality rather than dismantling it. And yes, every developer, data scientist, annotator and tester have the power to either introduce bias or actively work to reduce it in the technologies we rely on every day.
Related Article: MIT Researchers Develop New Method to Reduce Bias in AI Models
What Can We Do About AI Bias?
To fix AI bias, we must ask and act on the following:
- Who is represented in the training data? Training datasets should reflect the diversity of the populations the system is meant to serve, not just those who historically used it. We need global, multilingual and culturally diverse data sources. Not just more data, but better data.
- Whose language, culture and values shape the model? AI must be trained to respect multiple worldviews, not just align with dominant ideologies. We need to identify and challenge embedded cultural assumptions and ensure marginalized perspectives are treated as valid, not as fringe cases.
- Who is involved in designing, testing and refining these tools? If the teams behind AI lack gender, racial, linguistic and disciplinary diversity, bias is inevitable. Inclusion at every stage.
We need broader participation, more thoughtful definitions of “accuracy” and the willingness to admit that AI has never been neutral.
Webinar
Sep
25
CMS Briefing: A Live Look at What’s Next in AI-Driven Platforms
Learn how leading organizations are using AI‑driven tools to publish faster, personalize smarter and stay secure.
Register
Webinar
On demand
Ready or Not: How Data-First Organizations Are Unlocking Agentforce Potential
Learn how to cut through the noise, activate Agentforce and build a Salesforce AI strategy that actually delivers.
Watch Now
Webinar
On demand
AI in Customer Service: Faster Resolutions, Happier Customers
Don’t let rising demand burn out your team. See how to build a smarter, more resilient support org.
Watch Now
Webinar
On demand
From Hype to High-Impact CX Strategies That Actually Scale
Turn buzzworthy AI and outsourcing trends into measurable CX wins with fresh 2025 data.
Watch Now
Webinar
On demand
Insights to Action Rethinking the Contact Center for Real Business Impact
Join our exclusive webinar to hear CX executives share their innovative strategies for transforming service delivery.
Watch Now
Webinar
On demand
Accelerating Healthcare Ops with AI
Sign up to see how your peers are putting AI to work — and making it stick.
Watch Now
Make the Invisible Visible
Bias in AI is not just a glitch; it is a mirror. It reflects what society values, what it overlooks and who is centered in our systems of knowledge. But once we understand where bias comes from, we can begin to build better models. We can teach AI to reflect the full range of humanity, not just the loudest voices. Because AI is learning from us. And it is up to us to make sure it learns something worth repeating.
Learn how you can join our contributor community.
