AI Can Make Ethical HR Decisions Better Than Humans

The Problem With Your Boss’s Gut Feeling

Imagine you are up for a promotion. You have the numbers. You have the tenure. You have the track record. Your manager, however, has a hunch. She likes the other candidate better. They play golf together. They laugh at the same memes in Slack. Your boss is a good person, but her decision is about to be shaped by something she cannot even name: a cognitive bias that favors similarity, comfort, and familiarity over performance.

Now imagine that instead of your boss, a machine makes the call. Not a simple algorithm that scrapes resumes for keywords, but a system designed to reason through ethical principles. It does not care about golf. It does not care about memes. It cares about fairness, transparency, and consistency. It is not just an automated hiring tool. It is a machine that can think through moral tradeoffs the way a philosopher would, only faster, colder, and without the need for coffee.

That is the promise of a new framework from Waymond Rodgers, James M. Murray, Abraham Stefanidis, and William Y. Degbey, published in Human Resource Management Review in 2022. Their paper, "An artificial intelligence algorithmic approach to ethical decision-making in human resource management processes," argues that AI can make ethical HR decisions better than humans can. Not just faster. Not just cheaper. Better in the moral sense.

That is a bold claim. And the authors have the architecture to back it up.

What Makes a Decision Ethical, Anyway?

Before you can decide whether a machine can be ethical, you have to define what ethical means in the context of HR. Most people assume it means "following the rules." Do not discriminate. Do not retaliate. Pay people fairly. But ethical HR is more complicated than compliance. It involves tradeoffs. Should you hire the candidate with the better technical skills, or the one who will add diversity to a homogenous team? Should you promote the employee who has been loyal for years, or the one who just delivered a breakthrough project? Should you fire someone for a single mistake, or give them a second chance?

Humans handle these tradeoffs badly. We are inconsistent. We are influenced by mood, fatigue, and the last thing we ate. We have blind spots. We favor people who remind us of ourselves. We punish people we do not like. We make decisions based on emotions that we later rationalize with logic.

Rodgers and his colleagues argue that AI can do better. Not because machines are moral paragons, but because they can be designed to follow a structured ethical framework that humans cannot maintain consistently. The key is the Throughput model, a decision-making framework that the authors adapted for algorithmic HR.

The Throughput Model: How a Machine Thinks About Ethics

The Throughput model breaks down decision-making into four stages: perception, information, judgment, and choice. In plain English, it asks: What do you see? What do you know? How do you interpret it? And what do you do about it?

For an AI system, each stage can be programmed with ethical guardrails. At the perception stage, the system can be trained to recognize relevant data and ignore irrelevant data. It will not notice whether a candidate is attractive or speaks with an accent. It will only notice the factors that matter for the decision. At the information stage, the system can access a wider range of data than any human could process, including performance metrics, peer reviews, and historical outcomes. At the judgment stage, the system applies a predefined ethical framework, such as utilitarianism (maximize overall happiness) or deontology (follow universal rules). At the choice stage, the system selects the option that best aligns with the framework.

This is not a black box. The authors emphasize that the system is transparent. Every decision can be traced back to the data and principles that produced it. That is something humans cannot offer. Ask your boss why she promoted her golf buddy, and she will give you a story about "cultural fit." Ask the AI, and it will show you the math.

Why HR Is the Perfect Test Case for Ethical AI

Human resources is a domain where ethical failures are common and costly. Discrimination lawsuits cost companies millions. Bad hires cost even more. But the deeper problem is that HR decisions affect people's lives in profound ways. A promotion can change a family's income. A firing can destroy a career. A hiring decision can determine whether someone gets health insurance.

Rodgers, Murray, Stefanidis, and Degbey argue that HR is uniquely suited for AI because it involves repetitive, high-stakes decisions that follow predictable patterns. The same types of ethical dilemmas arise over and over: Who gets the job? Who gets the raise? Who gets the second chance? A human manager might make a different call each time, depending on their mood or the phase of the moon. An AI can apply the same principles consistently.

But consistency alone is not enough. An AI that consistently discriminates is worse than a human who occasionally does. The authors propose that AI systems should be designed to follow specific ethical positions, which they call "algorithmic ethical positions." These are not just rules; they are philosophical commitments embedded in the code.

Four Ethical Positions for Machines

The paper identifies four ethical positions that an AI system can adopt, based on the Throughput model:

• ▸Ethical egoism: The system prioritizes the interests of the organization. This is the default for most HR algorithms, which are designed to maximize efficiency, productivity, and profit. The problem is that this position can justify exploitation.

• ▸Utilitarianism: The system maximizes overall well-being for all stakeholders. This is harder to program because it requires weighing the interests of employees, managers, and shareholders. But it is also more defensible.

• ▸Deontology: The system follows universal rules, such as "never lie" or "never discriminate." This is the easiest to program because rules are binary. The downside is that rules can conflict. What do you do when honesty hurts someone's feelings?

• ▸Virtue ethics: The system embodies specific virtues, such as fairness, compassion, and integrity. This is the hardest to program because virtues are abstract and context-dependent. But it is also the most human.

The authors do not claim that one position is always best. They argue that the choice depends on the context. A hiring algorithm should probably be deontological, following strict rules against discrimination. A performance review algorithm might be utilitarian, balancing the needs of the team against the needs of the individual. A termination algorithm should be guided by virtue ethics, ensuring that the process is compassionate and fair.

What the Research Actually Found

The paper is a theoretical framework, not an empirical study. The authors did not run an experiment where they pitted AI against human managers and measured which made better ethical decisions. Instead, they synthesized existing research from multiple disciplines, including AI, ethics, and human resource management, to build a model that could be tested in the future.

What they found, in their analysis, is that the Throughput model provides a more complete picture of ethical decision-making than existing approaches. Most previous work focused on one stage of the decision process, such as how people perceive information or how they make judgments. The Throughput model connects all the stages, showing how perception, information, judgment, and choice interact.

More importantly, the authors found that AI systems can be designed to avoid the most common ethical failures in HR. For example, humans are prone to "confirmation bias," where they seek out information that confirms their existing beliefs. An AI can be programmed to consider all relevant data, even data that contradicts its initial assumptions. Humans are also prone to "groupthink," where they conform to the opinions of their peers. An AI can be programmed to ignore social pressure and follow the data.

The authors also addressed a concern that often comes up in discussions of AI ethics: accountability. If an AI makes a bad decision, who is responsible? The authors propose that the AI system itself should be designed to be accountable. That means it must be able to explain its reasoning, and its decisions must be auditable. This is a radical idea. It suggests that machines can be held to a higher standard of accountability than humans, who often cannot explain why they made a decision.

The Case for Machine Morality

Let me give you a concrete example of how this might work. Suppose a company is deciding which employees to lay off during a downturn. A human manager might be influenced by personal relationships. She might keep her friends and let go of people she does not like. She might also be influenced by stereotypes. She might assume that older workers are less adaptable, or that women are less committed.

An AI system, programmed with the Throughput model and a utilitarian ethical position, would approach the decision differently. At the perception stage, it would gather data on performance, tenure, and skills. It would ignore demographic data, unless the company had a specific policy about protecting underrepresented groups. At the information stage, it would analyze the data to identify which employees are most critical to the company's survival. At the judgment stage, it would apply the utilitarian principle: maximize overall well-being. That might mean keeping the employees who are hardest to replace, even if they are not the most popular. At the choice stage, it would produce a list of employees to lay off, along with a justification for each decision.

The result would be a decision that is consistent, transparent, and based on data. It would not be perfect. There might be edge cases where the algorithm fails. But it would be better than the alternative, which is a human manager making a gut decision that she cannot explain.

What This Does Not Prove

Let me be clear about what this research does not claim. It does not claim that AI can replace human judgment entirely. The authors are careful to say that AI should augment human decision-making, not replace it. The Throughput model is a framework for designing AI systems that support ethical decisions, not for automating them.

It also does not claim that AI is inherently ethical. The technology is neutral. It can be used for good or ill. An AI system that is designed to maximize shareholder value, without regard for employee well-being, could make decisions that are perfectly legal but deeply unethical. The authors argue that the ethical position must be embedded in the system from the start.

There is also a question of data quality. An AI system is only as good as the data it is trained on. If the data reflects historical biases, the AI will reproduce those biases. The authors acknowledge this and propose that AI systems should be monitored and updated to correct for bias. But this is easier said than done.

Finally, there is the question of trust. Even if an AI system makes better ethical decisions than a human, will people accept it? The authors cite research showing that people are more likely to accept AI decisions when they understand how the decisions are made. That is why transparency is essential. But transparency alone may not be enough. People may simply prefer human judgment, even when it is flawed.

What This Actually Means

• ▸Ethical AI is not about making machines moral. It is about designing systems that can reason through moral tradeoffs more consistently than humans can. The Throughput model provides a blueprint for how to do that, by breaking decisions into stages and programming ethical principles at each stage.

• ▸HR departments should start testing AI systems for ethical decision-making, not just efficiency. Most companies use AI to screen resumes or predict performance. Few use AI to reason through ethical dilemmas. The research suggests that this is a missed opportunity.

• ▸Accountability is easier to achieve with AI than with humans. Because AI decisions can be traced back to specific data and rules, they are more auditable than human decisions. Companies should demand that their AI systems provide explanations for every decision.

• ▸The ethical position of an AI system must be chosen deliberately, not left to default. The four positions identified by the authors (ethical egoism, utilitarianism, deontology, virtue ethics) have different implications. A system designed for ethical egoism will make different decisions than one designed for virtue ethics. Companies need to decide which position aligns with their values.

• ▸The biggest obstacle to ethical AI is not technology, but culture. The authors note that organizations often resist transparency because it exposes their decision-making to scrutiny. But if you want AI to make ethical decisions, you have to be willing to let it show its work.

The research by Rodgers, Murray, Stefanidis, and Degbey is not a final answer. It is a starting point. It challenges the assumption that ethics is a uniquely human domain. If a machine can reason through moral tradeoffs with more consistency and transparency than a person, then maybe we should let it. The alternative is to keep trusting our gut, which has a terrible track record.