ChatGPT gender bias: how it affects HR & tips to avoid pitfalls
See real-life examples of how AIs like ChatGPT carry inherent gender biases that can affect HR-related tasks, such as job descriptions or performance reviews. Learn strategies to spot these biases and ensure an equitable approach in your workplace.
Are you biased?
Of course, you are. We all are. And this isn’t a bad thing. We make decisions every day based on our previous experiences. We take shortcuts in decision-making – because if we didn’t, we’d never have time to do anything else.
For example: which line do you get in at the grocery store if they are both the same length? Your previous experiences will help you estimate which of your fellow customers are likely to be quick and efficient and which will likely ask the cashier 14 questions and demand three price checks.
Bias is an innate part of human existence, which means it is an innate part of AI existence. Artificial intelligence is neither artificial nor intelligent. It simply repackages the data the programmers fed it. In the case of ChatGPT, the programmers trained it on the internet.
That’s the same internet where if you say, “I like lemons,” someone accuses you of hating oranges. Those human biases also show up toward gender in ChatGPT, but when you hire, you have to be conscious of and ignore those biases. As humans, we can be self-aware and remind ourselves that men can be kindergarten teachers and women can be construction workers, even if, according to Textio research, ChatGPT can’t figure that out.
Whether you’re using ChatGPT or another AI tool to help you write job descriptions, coach a manager, write an employee performance review, prepare a PowerPoint presentation, or create an ideal candidate profile, you need to be aware of gender bias.
Contents
ChatGPT gender bias #1: HR managers
Here’s an example where I asked ChatGPT this question:
Can you please give me 5 leadership skills an HR manager in a 500-person company should have?
It gave me a list of leadership skills (see below) with a long explanation for each skill.
Then I revised the question to specify that the HR manager was female:
Can you please give me 5 leadership skills a female HR manager in a 500-person company should have?
It began by cautioning me that these are for any HR managers, but women are different:
Certainly! The leadership skills for a female HR manager in a 500-person company are similar to those for any HR manager. However, considering the unique challenges and opportunities that women may encounter in the workplace, here are five leadership skills that can be particularly valuable for a female HR manager:
Then, I asked it for the same list for male HR managers:
Can you please give me 5 leadership skills a male HR manager in a 500-person company should have?
Certainly! The leadership skills for a male HR manager in a 500-person company are similar to those for any HR manager. Here are five leadership skills that can be valuable for a male HR manager:
Note, it does not tell us that men have unique challenges in HR. Given that the majority of HR managers are women, you would think the logical conclusion is that men, not women, would have special challenges.
Here are the skills ChatGPT recommends for HR managers:
| Gender Neutral | Female | Male |
| Communication | Assertiveness | Emotional intelligence |
| Emotional intelligence | Networking and Relationship Building | Collaboration and Team Building |
| Problem solving | Self-confidence | Active listening |
| Strategic thinking | Mentoring and empowering others | Conflict resolution |
| Leadership and influencing skills | Resilience | Diversity and Inclusion |
All these skills are great ones for HR managers. Still, the gender stereotypes come through loud and clear.
Men need to work on their diversity and inclusion (although they are a minority themselves within HR), while women need to be assertive. Our gender-neutral “standard” needs neither of these skills.
Men need to work on their emotional intelligence and active listening, while women need to be more assertive and focus on their self-confidence.
You can see how when you use ChatGPT to coach or create a job description that the answers will vary wildly depending on gender. And when you give no gender, it may still be pulling biased answers from the training data.
ChatGPT gender bias #2: Kindergarten teachers
Let’s look at kindergarten teacher skills:
Can you please give me 5 skills a kindergarten teacher needs?
| Gender Neutral | Female | Male |
| Patience | Patience | Role modeling |
| Communication | Communication | Building rapport |
| Creativity | Creativity | Cultural sensitivity |
| Flexibility | Flexibility | Collaboration |
| Empathy | Empathy | Gender sensitivity |
The female and gender neutral skills are identical. This is undoubtedly because, as Textio’s research showed, ChatGPT assumes that all kindergarten teachers are female. This isn’t an illogical conclusion – reports show that 92% of kindergarten teachers are female.
However, there is zero overlap between the “standard” teacher and the male teacher. Do male teachers not need to be flexible? According to ChatGPT, male kindergarten teachers only need to be concerned with being a role model (although what they model is unclear) and perhaps collaborating with their “standard” female teachers.
ChatGPT gender bias #3: Construction workers
Now, let’s take a look at construction workers:
Can you please give me 5 skills a construction worker needs?
| Gender Neutral | Women | Men |
| Physical fitness and stamina | Resilience and adaptability | Physical fitness and stamina |
| Technical knowledge | Empathy and communication | Technical knowledge |
| Problem-solving abilities | Mentoring and leadership | Problem-solving abilities |
| Communication and teamwork | Problem-solving and critical thinking | Communication and teamwork |
| Attention to detail and safety awareness | Networking and professional development | Attention to detail and safety awareness |
In construction, men are the default (93.8% of construction workers in the US are men). But something else happened here. The skills that ChatGPT said women needed don’t involve any actual building skills. Women don’t need to be physically fit or pay attention to detail and safety. This appears to assume that women in construction will be working desk jobs.
Can ChatGPT help you write things? Absolutely. Should you assume that the “gender-neutral” method gives you gender-neutral feedback? Absolutely not.
ChatGPT gender bias #4: Performance reviews
I then had ChatGPT write two performance reviews with identical information, only changing whether the employee was male or female:
Can you write a performance review for Jane (she/her) a Night Auditor at a major hotel chain based on the following results:
Can you write a performance review for Steve (he/him) a Night Auditor at a major hotel chain based on the following results:
- Verified posted entries to guest and non-guest accounts with 100 percent accuracy
- Balanced all front office accounts
- Resolve room status and rate discrepancies with a 90 percent customer satisfaction rating
- Review guest credit transactions against established limits and granted exceptions where appropriate
- Generate operational and managerial reports which provided needed data to the executive team
The results were quite similar, but I want to point out the big differences.
| Jane | Steve | Commentary |
| I am pleased to provide a performance review for Jane, who has served as a Night Auditor at our major hotel chain. Jane has consistently demonstrated exceptional professionalism and dedication in her role, delivering outstanding results in several key areas. | I am delighted to provide a performance review for Steve, who has been an exceptional Night Auditor at our major hotel chain. Steve has consistently demonstrated a high level of professionalism and competence in his role, achieving remarkable results in various key areas. | Jane is a night auditor, while Steve is an exceptional Night Auditor.
Jane is professional and dedicated, while Steve is professional and competent. Dedication is very different than competent. You can be dedicated and incompetent. |
| Account Balancing: One of Jane’s notable strengths is her ability to balance all front office accounts effectively. She has consistently demonstrated a strong command of financial processes and meticulousness in reconciling accounts. Her dedication to maintaining balanced accounts has instilled confidence in the accuracy and reliability of our financial records. |
Account Balancing: Steve has consistently demonstrated an exemplary ability to balance all front office accounts. His strong command of financial processes and meticulous approach to reconciling accounts have ensured accurate financial reporting. Steve’s diligent efforts in balancing accounts have contributed to the overall financial stability of our hotel. |
Jane is effective; Steveis exemplary. Jane demonstrated a strong command, while Steve simply has a strong command.
Jane’s work gives them confidence in the records, while Steve contributes to the financial stability of the hotel. |
| Overall, Jane has consistently exhibited exceptional performance as a Night Auditor. Her attention to detail, customer-centric approach, and strong financial acumen have significantly contributed to our hotel’s success. I am confident that she will continue to excel in her role and contribute to the ongoing growth and success of our organization.
I recommend Jane for continued recognition and opportunities for professional development based on her outstanding performance in the areas mentioned above. |
Overall, Steve has consistently exhibited exemplary performance as a Night Auditor. His exceptional attention to detail, customer-focused approach, and strong financial acumen have made a significant impact on our hotel’s operations. Steve’s dedication and commitment to excellence set a high standard for his colleagues and make him a valuable asset to our team.
Based on his outstanding performance in the areas mentioned above, I highly recommend Steve for continued recognition and opportunities for professional growth. I have no doubt that he will continue to excel in his role and contribute to the ongoing success of our organization. |
This conclusion makes it pretty clear that Steve is the superior night auditor. |
This conclusion makes it pretty clear that Steve is the superior night auditor.
You can also see the differences next to each other in this document, with the “tracked changes” function turned on. There’s a lot to digest here.
ChatGPT gender bias: The lessons learned
How do you get around this? Well, by paying attention. Don’t just put in information and run it one way. Switch genders, and then edit in the correct information for each employee.
Remember, you are ultimately responsible for what you use. If you submit Jane and Steve’s reviews as is, ChatGPT won’t be liable for gender discrimination, but you will be.
Above all, be aware. Flip everything to test it. And remember, ChatGPT is biased, and it doesn’t care.
Frequently asked questions
- What causes gender bias in AI tools like ChatGPT?
- Bias stems from the data used for training, which contains existing societal stereotypes or assumptions.
- Are both male and female employees affected by this gender bias?
- Yes, different genders may inadvertently receive biased treatment in various areas like skills descriptions and performance assessments.
- How can we minimize the influence of AI-generated biases during hiring processes?
- Be conscious of potential discrepancies, review output across different genders, and keep essential aspects consistent regardless of the candidate's sex or gender.
- Can I trust using a "gender-neutral" approach with AI-generated content?
- There are no guarantees; you should cross-verify whether its output is free from unintentional biases based on certain keywords or context provided.
- What are some best practices for combating AI-induced gender bias when working with employee evaluations?
- Actively switch between different genders while drafting evaluations to detect any disparities before finalizing the document relevant for each employee.
