When AI gets it wrong: How people are bypassing algorithmic bias in everyday life

In today’s artificial intelligence-driven world, where algorithms influence everything from the content we see online to the jobs we apply for and the credit we receive, concerns over bias in AI systems are becoming increasingly urgent. A new study published in AI & Society reveals how users actively adapt to perceived unfairness in algorithm-driven systems.

The research titled "Understanding How Users May Work Around Algorithmic Bias" uncovers how people detect, respond to, and sometimes exploit bias, offering fresh insight into human-AI interaction and raising important questions for system designers and policymakers.

How do users detect and interpret algorithmic bias?

The study defines algorithmic bias as unfair or skewed outputs caused by data flaws, design decisions, or systemic inequalities embedded in AI models. However, user perception plays an equally critical role. According to the authors, bias exists in different configurations: sometimes real but unnoticed, sometimes imagined, and sometimes both perceived and present.

Detection often starts with cues users pick up during interactions, unexpected outcomes, repetitive patterns, or inconsistencies that suggest the system is favoring certain content, users, or decisions. These cues can be subtle, prompting users to develop theories about how the algorithm works. The research applies the Human–AI Interaction Theory of Interactive Media Effects (HAII-TIME) to explain this process. Under this framework, people use interface signals, patterns, and feedback loops to make sense of AI behavior, forming judgments about fairness and functionality.

Importantly, the study notes that misperceptions of bias are just as influential as real bias. Users may see unfairness where none exists, triggering unnecessary or counterproductive responses. This finding underlines the complexity of managing user trust in algorithm-driven environments. System designers must account not only for actual bias but also for how users interpret system decisions.

What strategies do users employ to work around algorithmic bias?

Once bias, real or perceived, is detected, users often employ workarounds to regain a sense of control. The study identifies two primary forms: internal workarounds, where users manipulate the system from within, and external workarounds, where they rely on outside tools or platforms.

Internal workarounds include actions such as adjusting settings, altering inputs, or gaming the algorithm by modifying behavior to achieve desired outcomes. For example, content creators may tweak keywords or posting patterns to optimize visibility on social media platforms they believe to be biased. Job seekers may reformat résumés to bypass automated screening filters.

External workarounds involve turning to alternative solutions outside the original system. Users may switch platforms, use third-party tools, or create parallel processes to counteract perceived bias. While these strategies can restore functionality, they also introduce inefficiencies, erode trust, and sometimes generate unintended consequences that complicate both user experience and system performance.

The researchers emphasize that these behaviors demonstrate user agency in digital environments. People are not passive recipients of algorithmic decisions; they adapt, resist, and innovate to reclaim fairness and effectiveness. However, these adaptations can also distort data, leading to feedback loops that reinforce bias or degrade system quality.

What are the implications for AI system design and regulation?

Understanding how users interact with bias provides valuable clues for designing systems that minimize the need for workarounds. If users are forced to bypass or manipulate algorithms to achieve fair treatment, it signals a breakdown in the system’s transparency and accountability.

The authors argue for greater algorithmic transparency, allowing users to understand how decisions are made without revealing proprietary secrets or exposing systems to exploitation. Clearer cues about system logic can help users differentiate between real bias and misperceptions, reducing unnecessary interventions.

Additionally, the study highlights the role of user education. Equipping users with better knowledge about how algorithms operate can improve their ability to respond effectively, fostering a healthier relationship between humans and AI. The research suggests that providing accessible explanations, interactive feedback, and bias detection indicators can empower users while maintaining system integrity.

For regulators, the findings underscore the need to address both technical and behavioral aspects of algorithmic fairness. Policies that enforce audits and bias mitigation must also consider how users perceive and react to AI systems. User-driven workarounds, if left unexamined, can create new risks and undermine efforts to build equitable technologies.

Furthermore, the study calls on system designers to develop user-centered algorithms that anticipate and integrate user responses. Rather than treating workarounds as problems to suppress, developers can use them as diagnostic tools to identify weaknesses and improve system design.