Episode 46: Why Does AI Make Biased Decisions? Understanding How Bias Happens and What We Can Do About It

Does AI Have “Preconceptions”? Understanding the Issue of Bias

In our daily lives, we often make judgments based on unconscious assumptions. For example, when meeting someone in a suit for the first time, we might think they seem reliable. Or if we hear someone is from a certain region, we might assume they’re kind. But of course, these impressions don’t necessarily reflect who that person truly is. These kinds of assumptions are natural human reactions.

Now imagine if artificial intelligence (AI) had similar “preconceptions.” That’s what we call “bias.” When AI makes decisions, the results can sometimes disadvantage certain individuals or groups. However, this doesn’t mean AI is intentionally discriminating. Rather, it’s because the data and systems used during its learning process may contain hidden biases that get passed along without anyone realizing.

Why Does AI Make Biased Decisions?

Unlike humans, AI doesn’t learn through emotions or personal experiences. Instead, it identifies patterns and trends from large amounts of data to make decisions. But within that data may lie deeply rooted values or prejudices from human society.

For instance, if an AI is trained using past hiring records to determine what kind of candidates are likely to be hired, and those records contain gender or age biases, then the AI will adopt those same tendencies. It might mistakenly conclude that “the type of person most often hired in the past = ideal candidate.”

This issue ties closely to what’s known as “training data”—the material used for teaching AI. Another important concept here is “generalization,” which refers to an AI’s ability to apply learned rules to new situations. If the original data is biased, then the rules derived from it will also be skewed. And once bias enters the system, it can affect all future decisions made by that AI.

Where Bias Shows Up: Everyday Examples

Let’s look at a more relatable example. Suppose an image recognition AI is trained to identify photos of doctors. If most of the training images feature male doctors, then when shown a photo of a female doctor, the AI might fail to recognize her correctly—it could even mislabel her as a nurse.

This isn’t just a technical error; it reflects unconscious human biases being inherited by machines.

Such bias can lead to serious consequences—especially in areas where fairness matters greatly, like loan approvals or job screenings. As AI becomes more involved in these processes, we must avoid situations where some people are unfairly disadvantaged simply because of how an algorithm was trained.

That’s why it’s important not to blindly trust that “AI must be right,” but instead take time to understand how its decisions are made and what influences them.

What We Can Do: Small Steps Toward Fairness

Of course, completely eliminating all bias may not be realistic. But by becoming aware of what kinds of bias tend to appear and how we might detect them early on, we can gradually reduce unfair outcomes.

Recently there has been growing attention on concepts like “fairness” in technology and developments such as “explainable AI,” which aims to make decision-making processes more transparent. We’ll explore these topics further in future articles.

Ultimately, AI is a product of human society—it reflects our values and historical context. While “bias” may sound negative at first glance, it also serves as a mirror that prompts us to examine ourselves more closely.

Because this issue can’t be solved by technology alone, each of us has a role in understanding and addressing it step by step. That mindset may be our first step toward building safe and trustworthy relationships with AI.