Quick Summary
A recent study reported in MIT Technology Review revealed a worrying trend: artificial intelligence—now widely used to screen job applicants—often develops biases, sometimes even stronger than human recruiters. Previously, it was assumed that AI simply mirrored the stereotypes present in its training data. But new findings show that large language models (LLMs) can actually create their own novel biases. This raises serious questions about the fairness and effectiveness of AI-driven hiring.
How AI Becomes Biased in Hiring
Automating recruitment with AI sounds promising: algorithms can sift through hundreds of resumes quickly, identify top candidates, and reduce human error. But in reality, things are more complicated. Here’s what actually happens:
- AI learns from historical data. If past hiring was biased—by age, gender, education, etc.—the model will pick up and perpetuate those imbalances.
- Models spot patterns—even the wrong ones. LLMs find every possible correlation, including random or socially undesirable ones.
- New biases can emerge. Research shows LLMs don’t just replicate old stereotypes—they can invent their own, even if those weren’t present in the original data.
Example: When AI Gets It Wrong
Imagine your training data contains lots of hires from a certain university. The AI notices this and starts favoring that school, even though the real reason those candidates succeeded might have nothing to do with the institution. Worse, the model might create a connection between something like a candidate’s address and their likelihood of success—even if no such relationship exists.
Why This Matters for AI Creators and Marketers
1. Reliability of AI Solutions Is at Stake
If you’re building AI products—especially for HR or automation—your solutions might not just fail to help, but actually harm your clients. Bias leads to errors, lost customers, legal problems, and PR disasters.
2. Ethics Are a Real-World Issue
Your users, clients, and stakeholders expect fairness and transparency. If your AI makes unfair decisions, word will spread fast. You’ll need to both explain how your system works and prove you’re managing risks.
3. Hiring and Marketing Are High-Stakes Use Cases
Whenever AI is “judging” people—whether in recruitment, ad targeting, or lead scoring—you need to be extra careful. Black-box decisions can result in discrimination and widespread complaints.
4. Business Misses Opportunities
If AI wrongly filters out talented candidates due to spurious correlations, your company loses out on strong employees and fresh ideas. In marketing, you risk ignoring potential customers who’d be a perfect fit.
What to Do Right Now
- Audit your training data. Regularly check datasets for hidden biases and clean them up.
- Implement model audits. Run test scenarios to spot unexpected biases. Bring in independent experts for review.
- Make AI decisions explainable. Build transparent reports explaining why the AI chose a particular candidate or lead score.
- Don’t rely exclusively on AI. Add human review at key decision points during hiring or evaluation.
- Educate your team on AI ethics. Study real-world cases, follow the latest research, and discuss risks openly.
More hands-on AI tools and practical ethics training — in my courses.
Based on materials from MIT Tech Review



