Your Boss Is an Algorithm and It Judges You Differently

Your Boss Is an Algorithm and It Judges You Differently

The first time an algorithm rejected your application, you probably didn't feel much. Just a cold, automated email: We have decided to move forward with other candidates. No explanation. No human on the other end. You shrugged it off as the cost of efficiency.

But here is what that algorithm was doing while you were shrugging: It was categorizing you. Not by your resume alone, but by patterns invisible to any human recruiter. It was comparing your word choice, your response time, your LinkedIn activity, your facial expressions during a video interview, and the tone of your voice. And it was doing all of this without knowing what it was doing. That is the paradox at the heart of the new workplace.

Sarah Bankins, Anna Carmella Ocampo, Mauricio Marrone, and Simon Lloyd D. Restubog (2023) reviewed 516 studies on artificial intelligence in organizations and found something unsettling: we are being judged by systems that do not understand us, yet we are expected to trust them more than we trust our human bosses.

The authors call this the "algorithmic management" problem. And it is changing what it means to have a job.

The Algorithm Sees You Differently Than Your Boss Does

Bankins et al. (2023) organized their findings into five themes. The most provocative one is this: humans and AI do not perceive capability the same way. When a human boss evaluates your performance, they factor in context. They know you had a rough week. They remember that project you nailed last quarter. They can see you struggling and might cut you some slack.

An algorithm does not see any of that. It sees data points. It sees keystrokes per hour, customer satisfaction scores, response times, and error rates. It sees what you did, not who you are.

The authors found that workers consistently rated algorithmic evaluations as less fair than human evaluations, even when the algorithm was objectively more accurate. Why? Because fairness is not just about accuracy. It is about being seen as a whole person. And algorithms are terrible at seeing whole people.

The Collaboration That Isn't

One of the paper's five themes is "human AI collaboration." But Bankins et al. (2023) found that this collaboration is rarely equal. In most workplaces, AI is not a partner. It is a supervisor. It tells you what to do, when to do it, and how fast to do it. And it judges you for failing.

The authors reviewed studies of platform based work: Uber drivers, Amazon warehouse workers, delivery couriers. In every case, the algorithm controlled the pace, the route, the pay, and the termination. Human managers were largely absent. The algorithm was the boss.

This matters because algorithmic management creates a different psychological experience than human management. When a human boss gives you a bad shift, you can argue. You can explain. You can appeal. When an algorithm gives you a bad shift, there is no one to argue with. The system simply is what it is. And that powerlessness, the authors found, is a major driver of worker dissatisfaction and burnout.

The Trust Gap

Here is where it gets weird. Bankins et al. (2023) found that people actually trust algorithms more than humans in some situations. For example, when evaluating their own performance, workers preferred algorithmic feedback over human feedback. Why? Because the algorithm does not lie. It does not have a bad day. It does not play favorites.

But that trust evaporates when the algorithm evaluates them. Suddenly, the cold objectivity feels like a trap. The authors call this the "trust asymmetry." We trust algorithms to be fair when we are not the ones being judged. But when we are the subject, we want a human.

This creates a strange dynamic in organizations. Managers are increasingly using AI to evaluate workers, then delivering the results themselves. The human becomes the messenger for a machine's judgment. And workers know it. They know the real evaluation came from an algorithm. They just do not know how it worked.

What the Algorithm Actually Measures

Bankins et al. (2023) reviewed studies that measured what AI systems actually capture. The results are sobering. Most workplace AI systems are trained on historical data. That means they learn the patterns of the past, including the biases. If a company historically promoted men over women, the algorithm learns that men are more promotable. If the company paid white workers more than Black workers, the algorithm learns that white workers are more valuable.

The authors found that AI systems in hiring, promotion, and performance evaluation consistently reproduced existing inequalities. They did not create new biases. They just made the old ones faster and harder to see.

But there is a subtler problem. Even when the data is clean, the algorithm still measures the wrong things. It measures output, not effort. It measures speed, not quality. It measures compliance, not creativity. The authors found that workers quickly learned to game these metrics. They optimized for what the algorithm could see, not for what actually mattered.

The Group Effect

The paper is structured as a "multilevel review," meaning it looks at individual, group, and organizational factors. One of the most interesting findings is at the group level. Bankins et al. (2023) found that teams with high AI surveillance developed different norms than teams with low surveillance. They became less collaborative. They hoarded information. They stopped helping each other.

Why? Because the algorithm was watching. And it rewarded individual performance, not team performance. So workers adapted. They stopped sharing credit, stopped covering for each other, stopped mentoring new hires. The algorithm turned colleagues into competitors.

This is not a bug. It is a feature. Many organizations implement AI specifically to increase individual accountability. But the authors found that this comes at a cost. Teams under algorithmic management had lower trust, lower cohesion, and higher turnover.

What the Research Does Not Prove

The Bankins et al. (2023) review is comprehensive, but it has limits. The authors are careful to note that most of the studies they reviewed were conducted in specific contexts: platform work, call centers, warehouses. We do not know if the same effects hold for knowledge workers, creative professionals, or executives. The algorithm that manages a delivery driver is very different from the algorithm that recommends candidates for a CEO position.

The authors also note that AI is evolving fast. The systems studied in the papers they reviewed are already outdated. Newer AI, including generative models, may change the dynamics entirely. We do not know yet.

And there is a big open question: can workers learn to trust algorithmic management if the algorithm is transparent? Some studies suggest yes. If workers understand how the algorithm works and why it makes its decisions, they evaluate it more fairly. But most workplace AI is a black box. Companies do not want workers gaming the system. So they keep the logic hidden. And that erodes trust.

The Five Pathways Forward

Bankins et al. (2023) identified five pathways for future research. They are worth listing because they reveal how little we actually know:

• ▸How does AI change the meaning of work for different types of workers?
• ▸Can AI be designed to support worker well being, not just productivity?
• ▸How do workers resist algorithmic control, and what are the consequences?
• ▸What happens to career development when AI makes promotion decisions?
• ▸How do organizations balance efficiency with fairness when using AI?

These are not academic questions. They are the questions every manager, every worker, and every policymaker will face in the next decade.

What This Actually Means

• ▸If you are a worker, assume the algorithm is watching. But also assume it is measuring the wrong things. Optimize for what matters to you, not just what the system can see. And if you feel the system is unfair, document everything. The algorithm does not know your context, but a human might.

• ▸If you are a manager, do not delegate judgment to the algorithm. Use AI as a tool, not a boss. The authors found that organizations where humans retained final decision making had better outcomes than those where AI made the final call. Keep the human in the loop.

• ▸If you are designing these systems, build transparency in from the start. Workers do not need to know every detail of the algorithm, but they need to know what it measures and why. The authors found that perceived fairness depends more on understanding the process than on the outcome.

• ▸If you are a policymaker, the evidence is clear: algorithmic management creates new forms of inequality and burnout. Regulation should require transparency, auditability, and human oversight. The authors recommend mandatory impact assessments before deploying AI in the workplace.

• ▸If you are a human being, remember this: the algorithm does not know you. It knows your data. Those are not the same thing. When you feel judged by a machine, you are right to feel that way. The machine is judging you. But it is judging a version of you that does not fully exist. That is the scariest finding in the entire review. The algorithm thinks it knows you. It does not. And neither does your boss.