How (not) to use AI for HR decision-making?

To what extent can HR managers rely on AI in their work? This pressing question has become increasingly relevant as AI technology continues to evolve. For HR managers, the challenge lies in understanding where AI can truly add value and where it might introduce risks. 

From selecting job candidates and conducting interviews to monitoring performance and mitigating biases, the integration of AI in HR processes is not always straightforward. Can AI accurately screen resumes and identify the best candidates without human oversight? Can AI effectively monitor employee performance and provide unbiased assessments? Is AI capable of reducing biases, or does it introduce new ones? Alma explains.

Decision-Making Bias

In a recent study (2024), 93 decision-makers were tasked with selecting candidates for two job positions using an AI-based dashboard. The dashboard provided AI-generated rankings and detailed candidate information across multiple levels, including an overview with fit scores, detailed professional backgrounds, and chat history with a chatbot. Integrated errors were included in the rankings to test reliance on AI. 

The study found that most users tended to rely heavily on AI-generated rankings without thoroughly evaluating the detailed information, demonstrating a status quo bias. While AI can enhance efficiency and consistency, it also risks overlooking details and perpetuating biases. By leveraging AI for objective tasks, ensuring consistent support, investing in higher-performing models, monitoring biases, and tailoring applications to task nature, HR managers can effectively balance AI's capabilities with human oversight, ensuring ethical and efficient outcomes.

Is Reliance on AI Good or Bad?

The tendency of HR managers to rely heavily on AI-generated rankings without thoroughly evaluating detailed information can be both a good and a bad thing.

Good Aspects:

1. Efficiency: By relying on AI-generated rankings, HR managers can save significant time and effort. This allows them to quickly shortlist candidates based on objective criteria, such as fit scores and keyword matches, streamlining the initial stages of the hiring process.
2. Consistency: AI can provide a consistent evaluation of all candidates, reducing the variability that might come from different human evaluators. This helps in maintaining a standardized approach to candidate selection.

Bad Aspects:

1. Overlooking Details: By not thoroughly reviewing the detailed candidate information, HR managers may miss important nuances about a candidate's qualifications, experiences, and potential cultural fit. This can lead to suboptimal hiring decisions.
2. Bias Reinforcement: If the AI model has biases, such as favoring certain demographic groups, relying solely on AI-generated rankings can perpetuate these biases. This can result in unfair hiring practices and a lack of diversity within the organization.

HR Managers, Use AI in These Ways:

While AI can enhance efficiency and consistency, it also risks overlooking details and perpetuating biases. Here are key insights for HR managers on how to effectively use AI in their decision-making processes:

1. Leverage Established Formulas: Use GPT-4 to calculate employee performance scores based on objective criteria like sales numbers or attendance records. GPT-4 excels at tasks grounded in clear, established formulas, providing accurate and reliable results.
2. Consistent Support Across Contexts: Whether evaluating training programs, managing employee schedules, or analyzing survey results, GPT-4 offers consistent support across various contexts. This reliability ensures that HR tasks are handled efficiently and effectively.
3. Invest in Higher-Performing Models: Although GPT-4 is more expensive than GPT-3.5, it is worth the investment for predictive tasks such as forecasting staffing needs based on current data trends. GPT-4's enhanced performance in these scenarios can significantly benefit HR decision-making.
4. Monitor and Mitigate Biases: Be aware of potential biases in GPT-4, such as favoring certain candidates due to subtle patterns in the data. Regularly reviewing AI decisions and adjusting inputs as necessary can help ensure fairness and equity in hiring processes.
5. Tailor Applications to Task Nature: Use GPT-4 for setting up automated interview scheduling, an objective task, while exercising caution when using it to assess cultural fit during hiring, a subjective task. Adding human oversight in subjective tasks can help check for biases and improve decision quality.

By leveraging AI for objective tasks, ensuring consistent support, investing in higher-performing models, monitoring biases, and tailoring applications to task nature, HR managers can effectively balance AI's capabilities with human oversight, ensuring ethical and efficient outcomes.

Which GenAI Model to Use: GPT-4 vs GPT-3.5 in HR Contexts

A study comparing GPT-4 and GPT-3.5 reveals important insights into their behavioral tendencies, which are crucial for HR decision-making:

• Gambler's Fallacy: GPT-4 is more likely to think a coin will land on tails after several heads in a row than GPT-3.5.
• Ambiguity Aversion: GPT-4 is more afraid of choosing something uncertain, like a mystery prize, than GPT-3.5.
• Risk Aversion: GPT-4 prefers safe choices, like taking a guaranteed $10 instead of a chance to win $20, more than GPT-3.5.
• Confirmation Bias: GPT-4 sticks to its initial beliefs, like always thinking cats are better than dogs, more than GPT-3.5.
• Mental Accounting: GPT-4 shows stronger money-related biases, like thinking it's better to save $5 on two separate purchases rather than $10 on one, than GPT-3.5.
• Framing Effect: GPT-4 avoids risks more consistently, like always choosing a sure gain instead of a gamble, no matter how the choice is presented, than GPT-3.5.

These behavioral patterns highlight the importance of understanding AI biases and integrating human judgment to ensure balanced and ethical HR decisions.

Conclusion

Integrating AI in HR processes offers significant advantages, but it's crucial to balance AI's capabilities with human oversight. By leveraging established formulas, ensuring consistent support, investing in higher-performing models, monitoring and mitigating biases, and tailoring applications to task nature, HR managers can enhance their decision-making processes.