When AI Starts Making HR Decisions

INTRODUCTION

AI is slowly becoming part of HR. So, today AI can help companies screen candidates, predict employee turnover, recommend training, monitor performance, and even support decisions about who should be hired or promoted. And honestly, this sounds very useful. AI can look at a huge amount of data much faster than a human HR manager can. It can find patterns, give recommendations, and save a lot of time. But the interesting part is that HR decisions are not just about numbers. They affect real people, their careers, their income, and sometimes even their future. So the big question is not really, “Can AI make HR decisions?” The bigger question is, “Should HR allow AI to make these decisions without enough human control?”

This is where things become complicated. Like, if an AI system says that a candidate is unlikely to accept a job offer, should the recruiter trust it? If an AI system says an employee is likely to leave, should the manager treat that employee differently? And if the AI makes a wrong decision, who is responsible? The HR manager? The company? The technology provider? Or can everyone simply say, “The AI said so”? So, AI can make HR faster, but HR still needs to make sure that technology does not remove human responsibility.

PROBLEM

The main problem starts when companies move from using AI as a support tool to using AI as a decision-maker. There is a big difference between the two. If AI looks at 10,000 applications and helps a recruiter find 500 possible candidates, the recruiter can still review those candidates. But if the company automatically rejects candidates because an AI system says they are not suitable, the AI is no longer just helping. It is influencing the actual employment decision.

So, this creates a few problems. First, AI systems learn from data, and data can be incomplete or wrong. Second, employees and candidates may not understand why an AI system made a particular recommendation. Third, people may trust AI too much simply because they think computers are more objective than humans. And fourth, companies may collect and use personal information without thinking carefully about whether they should be using it at all.

Like, imagine a candidate who has a 70% chance of rejecting a job offer according to an AI system. A recruiter might think, “Why waste time on this person?” and stop considering them. But what if the prediction is wrong? The candidate may actually be the best person for the job. So, the danger is that a prediction can slowly become a decision.

ANALYSIS OF PROBLEM

One important concept here is automation bias. Automation bias basically means that people can trust a computer-generated recommendation more than their own judgement, even when the recommendation may be wrong. So, if an AI system gives a candidate a low score, a recruiter might accept that score without asking enough questions. This is especially risky in HR because people are not machines and their behaviour cannot always be predicted perfectly.

Another issue is the accountability gap. Basically, accountability means knowing who is responsible when something goes wrong. When an HR manager makes a decision, we know who made it. But when an AI system recommends a decision, responsibility can become confusing. The company may blame the software, the software provider may blame the data, and the HR manager may say that they only followed the recommendation. So, HR needs to make sure that responsibility never disappears just because technology is involved.

There is also the idea of privacy by design. This means privacy should be considered while creating and using a system, instead of being treated as an issue after something goes wrong. So, HR should ask questions like, “Do we really need this employee data?” “Did the person know how their data would be used?” and “Who can access this information?” This becomes especially important when AI systems use large amounts of personal data.

Another useful framework is the socio-technical systems approach. It basically says that technology and people should be looked at together. So, HR cannot introduce an AI system and assume the technology will solve everything. The company also needs the right people, processes, rules, training, and controls around that technology. Basically, good AI in HR is not just about having a smart algorithm. It is about creating a system where the technology and the people work together properly.

ELABORATE REAL LIFE CASE

A very interesting real-life example is Recruit Career's Rikunabi DMP Follow service in Japan. Rikunabi is a job-search platform operated by Recruit Career. The company developed a system that used information about students' activity on the platform to estimate the probability that a student would reject a job offer. The idea was to provide this type of information to companies so they could better understand potential candidates.

But the problem was that the system was not just using data to understand recruitment trends. The information was being used in a way that affected how companies viewed individual job applicants. And the Personal Information Protection Commission of Japan found that Recruit Career had provided personal data to third parties without obtaining the required consent. The service was discontinued on August 4, 2019.

Recruit later found another problem involving consent. After reviewing its privacy-policy screens, the company said that 7,983 students had not properly given the required consent because information about the service was missing from some screens after a privacy-policy change. Recruit Career apologised and reported the issue to Japan's Personal Information Protection Commission.

So, the problem here was not simply that AI was being used. The bigger problem was how the data was collected, how the prediction was created, how it was shared, and whether people properly understood and agreed to this use. The Tokyo Labour Bureau also directed Recruit Career to properly respond to students whose information had been used and to provide appropriate explanations about the information involved.

This case shows something important. AI can make a prediction very quickly, but a prediction can still have a real human impact. Like, if a company believes that a candidate is likely to reject an offer, it might change how much effort it puts into that candidate. So, an AI prediction that looks like just a number can actually influence a person's opportunity.

CASE AND HR THEORIES, FRAMEWORKS, MODELS AND CONCEPTS REFLECTED IN THE PROBLEM

The first concept reflected here is automation bias. When a system produces a score or prediction, humans may give that output more importance than it deserves. In HR, this can be dangerous because recruiters may start treating an AI recommendation as a fact instead of treating it as one piece of information.

The second concept is the accountability gap. HR decisions need a clearly responsible human or team. So, even if