The Seductive Promise of AI Recruiters
The appeal of using artificial intelligence in recruitment is undeniable. In a country with a vast and diverse talent pool, AI-powered tools promise to sift through thousands of resumes in minutes, something that would take human recruiters weeks. Vendors
market these systems as a solution to human bias, suggesting that an algorithm can make more objective decisions than a person. For businesses looking to scale quickly and reduce costs, this proposition is incredibly attractive. The promise is a streamlined process that identifies the best candidates based purely on skills and qualifications, creating a more efficient and theoretically fairer hiring funnel. This has led to rapid adoption, with many organisations implementing AI for everything from initial resume screening to analysing video interviews.
When the Algorithm Gets It Wrong
The core problem is that AI systems learn from data, and historical hiring data is often filled with unconscious human biases. If a company's past hiring favoured candidates from specific universities or penalised applicants with resume gaps (which can disproportionately affect women), the AI will learn these patterns and treat them as rules for success. A famous example is Amazon's experimental recruiting tool, which had to be scrapped after it taught itself to penalise resumes that included the word "women's" and systematically downgraded graduates from all-women's colleges. This happens because the AI is designed to find patterns that correlate with past success, and if past success was male-dominated, it learns to favour male candidates. The bias isn't programmed in; it's learned from the data, making it subtle and dangerous.
Why Vendor 'Fairness' Guarantees Fall Short
A common and costly misconception is that if an AI tool causes discriminatory outcomes, the vendor is solely responsible. This is incorrect. Legally and reputationally, the employer who uses the tool is accountable for its hiring decisions. Vendors may test their models for bias before selling them, but these tests are conducted in a controlled environment, not with your company's unique pool of applicants. An algorithm that seems fair with a generic dataset might perform very differently when exposed to the specific demographics and backgrounds of people applying to your jobs. Relying only on a vendor's pre-deployment audit is like buying a car based only on a test drive on a closed track; you don't know how it will perform in real-world traffic until you drive it there yourself.
What Does 'Continuous Monitoring' Actually Mean?
Effective monitoring isn't about spot-checking; it's a continuous process of governance. It starts with demanding transparency from vendors about how their algorithms work. More importantly, it involves regularly auditing the outcomes of the AI system. This is known as adverse impact analysis, where companies compare the pass rates of different demographic groups (e.g., men vs. women, different age groups). The 'four-fifths rule' is a common benchmark, where the selection rate for any group should be at least 80% of the rate for the group with the highest rate. If disparities are found, the system needs to be investigated and adjusted. This creates a feedback loop that allows the organisation to catch and correct emerging biases before they become systemic problems. It's about maintaining human oversight, not abdicating responsibility to a 'black box'.
The Legal and Reputational Risks in India
While India is yet to enact specific laws directly regulating AI in hiring like the EU's AI Act, that doesn't mean there are no legal risks. India's existing constitutional and labour laws provide a framework against discrimination, which can be applied to decisions made by algorithms. An AI tool that systematically filters out qualified candidates from a particular region, gender, or social background could lead to legal challenges. Beyond the courtroom, the reputational damage can be severe. In a competitive market for talent, being known as a company whose hiring process is unfair or biased can deter the very best candidates from applying, ultimately undermining the goal of finding top talent. Transparency is key; candidates should be informed when AI is being used in their evaluation.
