Artificial Intelligence in Recruitment: The Dark Truth

Many AI models have been released, making lots of promises about improving the recruitment process. For example, an applicant can get AI to write their CV, and equally a hiring manager can get AI to write them a job advert. AI can also be used to shortlist applications, which is ironic if it matches AI-generated CVs against an AI-generated job description.

When Schmidt and Hunter analysed assessment tools in 1998, they wouldn’t have encountered AI. But we can benefit from their heuristic, which identified the best tools were job relevant, objective, consistent, comprehensive, and predictive. Based on this, I don’t think most AI tests are currently valid, although I hope they may be valid in the future when combined with other assessment tools.

AI suffers from four distinct challenges:

1. Garbage in = garbage out.
2. There is no feedback loop.
3. Efficacy of decisions.
4. We prefer human interaction.

‍

1. Garbage in = Garbage Out

In computer science, the adage “garbage in = garbage out” suggests that incorrect or poor quality input will produce faulty output. Unfortunately, some CVs are the equivalent of “garbage in” because they:

• Are not objective or comprehensive: CVs are at best an advertisement, sometimes pure fiction. They contain irrelevant information (I once saw a CV which included the candidates blood type!) yet miss valuable information.
• Are not necessarily written by the applicant: Some are created by professional CV writing services or AI! How can anyone accurately assess an applicant if they didn’t write their CV?
• Come in various formats: Making it challenging to find information and benchmark applicants.

Another piece of “garbage in” are job adverts, particularly when written like job descriptions. With the greatest respect to many hiring managers, many don’t know what they want, reuse an outdated job description, and don’t have the time to make improvements. Unfortunately, this is unlikely to change because recruitment is often a distressed purchase; a business is growing fast so they need someone fast, or an employee has resigned and they urgently need to find a replacement! In most circumstances, there is little time to think, “Do I still need to hire for this job?” and if so, “Who do I really need?”

Therefore, if jobseekers and hiring managers put the equivalent of “garbage in” to the recruitment process, the result will be “garbage out”! For example, an applicant may have GCSE French. A hiring manager asked AI to write a job advert that stated French speaking is required, but doesn’t specify what fluency. The advert “matches” the candidate, but the hiring manager is frustrated with the application as it isn’t what he wanted! To call the applicant “garbage” in this analogy may sound harsh, but the hiring manager certainly didn’t get what he needed, and hasn’t saved time using AI.

‍

<span class="purple-callout"><p>A quick mention: I'm available if you're seeking advice or hands-on support. Get your free consultation.</p><p>You'll also find my talent acquisition book helpful -it's a bestseller!</p><p>I can even advertise a job on top UK job sites for just £199.</p></span>

‍

2. There is No Feedback Loop

Have you seen how a robot vacuum cleaner learns and adapts to clean a room? It constantly bumps into objects and sets off in a new direction until it bumps into another object. It’s learning because it has a feedback loop.

Unfortunately, AI does not have a feedback loop. It is never informed which candidates were suitable in the long-term. Consequently, there is no reinforcement of whether it helped make a good or bad decision. It’s the equivalent of a robot vacuum cleaner hitting a wall and keeping going regardless, because if the wheels are spinning it thinks it’s making progress.

‍

3. Efficacy of Decisions

Studies by Armstrong et al. (2024) have already proven that a type of AI called large language models (LLMs) are biased in the recruitment process, scoring applicants differently based on race, race representation, and gender representation. For example, women’s names and those that were non-native English resulted in less favourable hiring decisions. This isn’t because the LLMs have been intentionally programmed to be biased, but because they have been trained on human decisions.

I’m also concerned by some assumptions being made. In early 2023, I saw a demo for an AI tool that shortlisted candidates. The salesperson explained how, “It even analyses what the applicant looks like to decide their suitability”. He went on to say, “For example, if you’re recruiting for a physical trainer, and the applicant looks overweight, it will reject them”. I find this deplorable, and I hope you do, too.

The broader issue is that AI are “black boxes” where we don’t see their assumptions and calculations, so how do we know they aren’t building biases? For example, many vendors train their AI by watching hiring managers for a few months, identifying patterns, and trying to replicate them. But what if the hiring managers are making bad decisions?

Simply put, most suppliers of AI recruitment tools don’t understand if they are valid, so it’s hard to sign off on a tool like that.

‍

4. We Prefer Human Interaction

Currently we have ingrained human behaviour:

• We don’t like interacting with computers: Even “Press 1 to speak to...” is frustrating when we’d prefer someone to answer the phone. So why would we want such an important interaction made in such an inhumane way?
• We don’t trust computers: While humans are far from perfect, no one likes to think their current value and future worth are determined by a computer. People want meaningful relationships, especially when a job has such a significant impact on their lives.

Overall “human resources” starts with “human” because it needs a human touch. Recruitment decisions are rarely binary; they are nuanced, requiring empathy and careful decision-making. AI isn’t the fate of recruitment, and I expect those employers who stop automating everything will ultimately win because the candidate is in the driving seat.

‍

Additional Resources

• Talent Acquisition Book; The Secrets of Great Recruitment - How to Recruitment Great Employees.
• Article; An Overview of the Recruitment Selection Process.
• Article; Improve Candidate Experience in Recruitment: Best Practice.
• Article; Avoid Discrimination in Recruitment: Ensure Fairness.
• Article; How AI is disrupting Job Boards, Recruitment & HR.