The AI revolution has been one of the fastest to occur. In the workplace, AI has become almost part and parcel of everyday life. In fact, one could go to the extent to say that AI has changed workplace decision-making and overall functionality. The question we are addressing today is not whether this change is good or bad, but rather, if the tools we use, that is, the AI models, are ready to take on the challenge of understanding and dealing with a diverse workforce.
Diversity comes in many forms. Inclusion more so. With AI, we walk on a tightrope of reiterating our biases or creating a system that is outside of our human bias. Many organisations are adopting AI because of the “neutrality” that an AI professes to maintain. However, we forget that even an AI is coded by a human being, a human being who has their own biases, non-conscious and otherwise.
An AI, for all intents and purposes, is only a machine after all. A machine that learns from previous decision-making, thereby creating patterns from the data it has collected. The reality is that the decisions made by an AI are completely dependent on whose data it is built on and the values of the designer.
Understanding Algorithmic Bias
When AI tools are developed without the voices of marginalised communities, they automate bias at scale. But it is important to remember that when these same communities shape AI development from the ground up, these tools can actively dismantle structural inequities that human processes have long perpetuated.
When systems learn from historical data to identify patterns that predict success, that data tends to reflect patterns that have historically been biased towards male, able-bodied, cisgender, and neurotypical candidates.
Let us take an employee recruitment AI tool as an example. A recruitment algorithm trained on a decade of hiring decisions learns and perpetuates biased lessons. Employment gaps become automatic disqualifiers. Communication styles outside narrow corporate norms trigger rejection. Research has documented these patterns—Amazon discontinued its AI recruiting tool after discovering it systematically downgraded resumes containing the word “women’s.”
Yet an algorithm that has been trained to recognise nuanced diversity markers can change these outcomes for the better, thereby creating a workforce that is stronger in viewpoints and more diverse in thought, action, and orientation.
AI as a Tool for Equity
Equity in AI does not happen by accident. However, when AI is designed on purpose, with marginalised communities at the centre of its development, these systems can achieve what traditional human-led processes have struggled to accomplish for decades.
An AI system built by and with queer individuals, people with disabilities, neurodivergent thinkers, and women from diverse backgrounds embeds fundamentally different values. It asks different questions about what constitutes merit and what barriers should be accounted for rather than penalised.
When recruitment algorithms are trained on truly diverse successful employees rather than historical patterns of privilege, they identify talent that human reviewers overlook. Employment gaps are understood as periods that may bring valuable perspectives. Communication styles that differ from corporate norms are recognised as reflections of diverse backgrounds. Language can be learnt, after all, and is a skill that is constantly growing. By giving space for this growth, companies become more inclusive of thought, idea, and mind.
Well-designed systems flag when certain groups are disproportionately filtered out, alerting organisations before problems compound. They expand talent pools by identifying skills in non-traditional places. For example, candidates from regional language backgrounds, those with non-conventional educational paths, or individuals whose community leadership demonstrates qualities that traditional screening misses.
Building Systems That Work
So now comes the question: what must organisations do? The answer requires viewing decision-making through the lens of intersectionality. People with disabilities, queer employees, neurodivergent staff, and women from diverse backgrounds must be brought in, not as consultants to merely review, but as designers from the start.
Before deployment, organisations must conduct equity audits using diverse sample profiles. After deployment, outcomes must be monitored and disaggregated by gender, disability status, caste, class, and where disclosed, sexual orientation and neurodivergence. When certain groups consistently fare worse, organisations must investigate rather than accept algorithmic outputs as neutral truth.
The fact of the matter is that AI should assist human judgment, not replace it.
The Way Ahead
The question is not whether AI will be part of workplace decision-making—that shift has occurred. The question is whether these systems will replicate historical exclusion or actively advance equity.
This requires prioritising equity even when it conflicts with efficiency, centering marginalised voices as foundation rather than afterthought, and interrogating comfortable assumptions about what productivity and merit mean.
The alternative is automating exclusion at scale, embedding bias so deeply that it becomes nearly impossible to challenge, thereby creating systems that appear objective while perpetuating the very inequities they were designed to eliminate.
There is a way ahead, however. Inclusively designed AI can expand opportunity, reduce human bias, and create pathways for talent that traditional systems overlook. But only if equity is the deliberate priority from the beginning. The technology itself is neutral. The choices about how to build and deploy it are not.
