Bias by Design? The Quiet Power of Intentional Algorithmic Manipulation

When we think about algorithmic bias, we often picture it as accidental, the unfortunate result of incomplete data, skewed assumptions, or historical inequality baked into training sets. But what if bias isn’t always a mistake? What if, sometimes, it’s the point?

Take the airline industry. Visit a travel booking site operated by a particular airline and you're likely to see that airline's flights displayed at the top of the search results. This isn't coincidence or consumer preference at work, it’s a deliberate design choice. The algorithm has been engineered to prioritise the operator’s commercial interests over neutrality, user convenience, or even price. It’s a textbook case of intentional algorithmic bias, and it's more common than many realise.

This form of bias differs sharply from the inadvertent type most frequently discussed in ethical AI circles. It isn’t caused by oversight or lack of awareness. It’s strategic. And it raises a fundamentally different set of concerns, not just about fairness, but about power and accountability. These algorithms are not neutral tools; they are extensions of institutional self-interest, quietly shaping markets and consumer behaviour behind a veneer of objectivity.

Examples abound beyond aviation. E-commerce platforms may promote their own-brand products over competitors'. Credit-scoring algorithms can be tuned to reflect internal profit motives rather than public fairness standards. Even content moderation systems might prioritise advertiser-friendly narratives over genuinely harmful material. In each case, the bias is embedded not by accident, but by design.

Addressing this isn’t a matter of better data or improved machine learning techniques. It requires transparency mandates, regulatory scrutiny, and user rights to challenge and understand algorithmic decisions. As legislators around the world begin to crack down on self-preferencing and dark patterns, it’s critical that we expand the conversation beyond unintentional bias. We must be willing to ask not only who trained the model, but who benefits from how it behaves.

Algorithmic bias is not always a bug. Sometimes, it’s a feature. And when it is, our response needs to be not just technical, but political.