Here are three steps HR leaders can implement to ensure AI bias is minimised
Machine learning models can have bias, not intentionally, but because of the data it uses to make decisions.
Luli Adeyemo, executive director of TechDiversity Foundation noted that AI reflects human decisions.
“AI systems don't start with bias, they inherit it. The bias doesn't come just from data. It comes from decisions made long before the first line of code is written,” Adeyemo explained.
“Who decided what data matters? Who defined what ‘success’ looks like? Who was in the room when the problem statement was written? These are human choices. These are governance choices. And if those choices were made by a homogeneous group, the same backgrounds, the same experiences, the same blind spots, then those blind spots get baked into the system.”
Adeyemo has experienced the impact of bias in AI systems, which can have affect HR processes.
From facial recognition systems that can’t identify darker skin tones to voice recognition software that flags some accents as suspicious, these trends can have a devastating impact on company diversity.
Adeyemo has witnessed instances where women’s employment gaps due to maternity leave were factored into viability for recruitment.
“These aren't accidents. They're the result of choices made by people who didn't have the perspective to see the problem,” she said.
“Unaddressed AI bias costs your organisation money, reputation, and talent.”
What can be done?
There are a variety of measures that can be taken to mitigate biases in AI systems. According to Adeyemo, a great starting point is the three layer approach:
Layer 1: Data and technical auditing: “You need to audit your training data. You need data scientists who can spot patterns of bias in datasets. You need technical testing to identify where systems perform differently across demographic groups,” she said.
Layer 2: Governance and decision-making: “This is where most organisations fail. You need diverse people in the room before the technical work begins. You need someone asking: ‘Who are we building this for? What are we optimising for? What outcomes are we willing to accept? What trade-offs are we making?’
“You need governance frameworks that require diverse perspectives at every stage, from problem definition through to deployment and monitoring. This is where the real work happens. Not in the code. In the thinking," Adeyemo explained.
Layer 3: Continuous learning and accountability: “You need ongoing monitoring and updates. AI systems don't stay fair. Societies change. New data comes in. Markets shift. If you audit your system once and then forget about it, you've solved yesterday's problem. You need processes that treat AI governance as continuous practice, not a one-time checkbox,” said Adeyemo
AI bias is not merely a technical flaw – it is a reflection of the human decisions and governance structures that shape technology from the outset.
As Adeyemo highlighted, unchecked bias in AI can undermine a company’s financial performance, damage its reputation, and drive away valuable talent.
To address this, HR leaders must adopt a proactive, multi-layered approach. By embedding these practices, organisations can not only minimise AI bias but also foster fairer, more inclusive workplaces that are better positioned for long-term success.
