What Happens When AI Decides Your Career

Here is where the benefits start to feel more real. When AI is trained on genuinely diverse, fair data, it can outperform humans on certain consistency metrics. It doesn't get excited about a candidate who went to the same school as the hiring manager. It doesn't unconsciously favour names that sound familiar. Studies have shown that structured, criteria-based hiring tends to produce better long-term employee performance than gut-feel interviews, and AI can enforce that structure at scale.

AI also gives smaller companies access to tools that only large enterprises used to have. A 30-person startup can now use intelligent resume screening without building a whole HR department. Scheduling software powered by AI means candidates get faster responses and fewer of those weeks-long silences that make job hunting feel so demoralising. For candidates, faster processes are genuinely better.

There is also something worth saying about the data. AI hiring tools can track which candidate attributes actually correlate with long-term success at a company, not just who seemed impressive in a 30-minute conversation. If the data is good and the model is well-designed, the output could genuinely be better than what experienced human recruiters consistently produce.

But then you look at what can go wrong, and the problems are serious. AI hiring systems learn from historical data. And historical data reflects historical hiring decisions, which in many industries reflect decades of systemic bias. Amazon famously scrapped an AI recruiting tool in 2018 after discovering it had learned to penalise resumes that included the word 'women's,' as in 'women's chess club,' because most of the successful hires in the training data were men. The model didn't intend to discriminate. It just learned from a biased world and reproduced that bias at scale.

The keyword-matching problem is another one that quietly disadvantages huge numbers of qualified candidates. If your resume doesn't contain the exact phrasing the AI is scanning for, you disappear, even if you have the exact experience the role requires. This disproportionately affects people who are career changers, self-taught, or who learned skills through nontraditional paths. It also creates a strange arms race where candidates learn to game the system by stuffing resumes with keywords, which means the AI is increasingly selecting for people who are good at optimising for AI, not necessarily good at the job.

And then there is the video interview scoring. Some platforms analyse candidate responses for things like word choice, speaking pace, eye contact, and micro-expressions. The research on this is weak at best, and the equity concerns are significant. These systems have shown lower accuracy for people with accents, for people with certain disabilities, and for people who are simply not comfortable performing in front of a camera. That is not efficiency. It is just bias with a more sophisticated wrapper.

The deeper issue is accountability. When a human recruiter makes a bad call, you can at least ask why. When an algorithm rejects you, the reasoning is often opaque, even to the company using it. Some jurisdictions are starting to push back on this, with laws requiring algorithmic transparency or human review of AI-driven hiring decisions. New York City passed legislation in 2023 requiring employers to audit AI hiring tools for bias. More regulation is coming, slowly, and it probably needs to come faster.

What I think we actually need is a clear-eyed view of what AI is good at in hiring and what it is not. Sorting thousands of resumes based on objective criteria? Sure, that's a reasonable use. Scoring the emotional authenticity of someone's voice during a job interview? That feels like a step too far. AI should be handling the genuinely mechanical parts of recruiting, not making character judgments about people's potential based on data proxies that have never been rigorously validated.

The goal, at least the goal worth working toward, is a hiring process that is faster and fairer, not just faster. That means using AI to reduce the grunt work while keeping humans in the loop for anything that requires nuanced judgment. It means auditing these systems regularly and being honest when they produce unfair outcomes. And it means remembering that behind every resume is a person who wants a chance to show what they can actually do. No algorithm, no matter how sophisticated, should get to be the last word on that.