Bias in Non-ML Systems
What you'll learn: How bias emerges in deterministic, rule-based software—not just in machine learning models—and why traditional algorithms require ethical scrutiny.
The Hidden Bias in "Objective" Code
When we think about bias in software, machine learning often comes to mind. But bias exists in any system that makes decisions about people—even if it's just if/else statements and hardcoded rules.
Rule-based systems, search rankings, recommendation engines, and scheduling algorithms all encode assumptions. Those assumptions reflect the worldviews, priorities, and blind spots of their creators. Unlike statistical bias in ML, this is structural bias: it's baked into the logic itself.
Common Sources of Non-ML Bias
Rule-based filters might screen job applications by requiring 10+ years of experience for a technology that's only 8 years old—accidentally filtering out self-taught developers or career changers.
Search ranking algorithms might prioritize results based on click-through rates, which can amplify existing prejudices (e.g., showing male-dominated professions more prominently to men).
Recommendation systems using collaborative filtering ("users like you also liked...") can create echo chambers, reinforcing narrow perspectives.
Resource allocation algorithms (scheduling hospital beds, routing delivery drivers) might optimize for efficiency while ignoring equity—serving wealthy neighborhoods faster simply because they're closer together.
Think Like an Auditor
Ask: Who benefits from this rule? Who might be disadvantaged? A resume parser that rejects non-standard formats might exclude candidates with disabilities using assistive technology. A credit scoring rule requiring long credit histories disadvantages immigrants and young people.
Key Takeaway: Bias doesn't require AI—any deterministic system that automates decisions about people embeds values and assumptions that can systematically advantage or disadvantage groups, demanding the same ethical scrutiny you'd give machine learning models.