AI is fast transforming HR and business decision-making, but it’s not without inclusion risks. That’s why it’s key to involve diversity and inclusion leaders and representatives from under-represented and marginalised groups early in AI model design and evaluation.
Bias in AI systems has already led to public controversies. For example, the Apple Card investigation in 2019 revealed that men were given significantly higher credit limits than women with comparable financial profiles.
But AI can also drive inclusion – if built well. Models trained to remove bias in the recruitment process, for example, can help deliver a more diverse candidate pool.
When AI models are trained on diverse data sets and consider inclusivity and fairness, there’s many advantages.
A framework we like for its focus on fairness and bias mitigation in recruitment is the Fair AI for Recruitment (FAIR) framework developed by sapia.ai, which presents a set of measures and guidelines to implement and maintain fairness in AI based candidate selection tools.
There are other fairness toolkits (e.g., IBM AI Fairness 360, Microsoft Fairlearn, Google What-If Tool) to identify and mitigate unwanted bias in model predictions.
What Business and HR Leaders can do now
As AI adoption accelerates, HR and business leaders have a crucial role to play – not just in deploying AI, but in shaping it responsibly. Here’s some ways your organisation can ensure equitable outcomes.
- Involve D&I leaders and representatives from under-represented and marginalised groups early in model design and evaluation, to ensure inclusivity of the system.
- Consider accessibility and cultural sensitivity at every step.
- Ensure your AI tools are trained on datasets that reflect the diversity of the population the model will impact (across gender, ethnicity, ability, socioeconomic status, geography, etc.).
- Ongoing audits of AI systems to detect disparities, bias, or unintended exclusion.
- Explore fairness toolkits (e.g., IBM AI Fairness 360, Microsoft Fairlearn, Google What-If Tool) to identify and mitigate unwanted bias in model predictions.
- Ensure equitable support, redeployment, and reskilling for people in the labour market at greater risk of disadvantage from AI technology. Women, older workers, First Nations Australians, and people with disability may face disproportionate risks because of occupational segregation and digital access gaps, as the recent Job and Skills Australia report Our Gen AI Transition highlighted.
