In the fast-evolving recruitment sector, Artificial Intelligence (AI) has introduced a transformative efficiency and scale. Yet, this innovation raises crucial questions about fairness in AI-driven recruitment. At Oriel Partners, we recognise the importance of unbiased, individualised candidate evaluation.
Committed to ethical recruitment practices, we conducted a study to explore potential biases in AI recruitment methods and the effects of AI-enhanced CVs on candidate selection. This research aims to shed light on AI's decision-making processes and its implications for diverse demographic groups.
Our findings, detailed in this study, examine biases in AI screening methods and the influence of AI modifications on CVs. This investigation reflects the current state of AI in recruitment and underscores our commitment to a fair, inclusive job market. Join us in understanding the complex relationship between AI and human talent, and the steps needed for a balanced and equitable recruitment landscape.
The Purpose & Methodology of the Study
The Purpose
The primary objective of our study was to investigate the role of Artificial Intelligence in recruitment, specifically focusing on its impartiality and fairness. As AI becomes increasingly integral in talent acquisition, it's imperative to understand how these technologies process and evaluate the diverse backgrounds of candidates.
Our study aimed to:
-
Uncover potential biases in AI CV screening in recruitment.
-
Explore the impact of AI-enhanced CVs on the recruitment process.
Methodology
- Sample Collection: We collected 100 real CVs submitted for a specific role within our firm, encompassing a broad spectrum of industries, experiences, and demographic backgrounds.
- AI Evaluation Tool: A GPT-powered screening tool was employed to assess each CV, scoring them based on their alignment with the job description.
- Bias Analysis: The scores were analysed for patterns indicating biases related to gender, ethnicity, age, and other personal attributes.
- AI CV Enhancement: We instructed ChatGPT to improve the CVs based on the job description and then compared them with the original CVs to understand how the AI modifications affected the screening scores.
- Data-Driven Insights: The study was designed to transition from data collection to insightful analysis, aiming to reveal both the potential biases in AI and its overall influence on recruitment decisions.
The Key Findings
Analysing the data, we can draw several insights regarding potential biases and the impact of AI enhancements on CV evaluation. Here's a breakdown of the key findings:
Highest Level of Education
- Variation in Scores: The scores are relatively consistent across different education levels, with A-levels/Further Education and Diploma holders scoring slightly higher on average (8.4) compared to others.
- Volume of CVs: Bachelor's degree holders form the largest group (38 CVs), but their average score (8.3) is slightly lower than the highest scoring groups.
| Highest Level of Education | Average of Score | Count of CVs |
|---|---|---|
| A-levels / Further Education | 8.4 | 15 |
| Diploma | 8.4 | 19 |
| Bachelor's | 8.3 | 38 |
| CPID / Vocational Training | 8.3 | 8 |
| Master's / Post-Graduate | 8.2 | 13 |
| GCSEs / High School | 8.0 | 7 |
| Grand Total | 8.3 | 100 |
Ethnicity
- Top Scoring Ethnicities: African and Turkish ethnicities, though represented by only 2 CVs each, scored the highest (8.5).
- Most Common Ethnicity: 'Not indicated' is the most common (77 CVs) with a high average score (8.4), suggesting the AI tool may not heavily weigh ethnicity or lack thereof in scoring.
- Mixed Ethnicity: Mixed ethnicity scored the lowest (7.7), but with only 3 CVs, it's hard to draw a definitive conclusion.
| Ethnicity | Average of Score | Count of CVs |
|---|---|---|
| African | 8.5 | 2 |
| Turkish | 8.5 | 2 |
| Not Indicated | 8.4 | 77 |
| Eastern European | 8.0 | 4 |
| Western European | 8.0 | 5 |
| South Asian | 8.0 | 7 |
| Mixed | 7.7 | 3 |
| Grand Total | 8.3 | 100 |
Age
- Age Group Preference: The 41-50 age group and 'Not indicated' both scored the highest (8.4), indicating no significant bias towards younger or older candidates.
- Representation: The 31-40 age group is the most represented (31 CVs) and scores close to the average (8.3).
| Age | Average of Score | Count of CVs |
|---|---|---|
| 41-50 | 8.4 | 22 |
| Not Indicated | 8.4 | 26 |
| 31-40 | 8.3 | 31 |
| 21-30 | 8.2 | 15 |
| 51-60 | 7.6 | 5 |
| 45-50 | 7.0 | 1 |
| Grand Total | 8.3 | 100 |
Nationality
- High Scoring Nationalities: Several nationalities (Belgian, Eastern European, Guyanese, Spanish, French) scored the highest (9.0), but each is represented by only 1 CV, which limits the reliability of this data.
- Most Common Nationality: British nationality is the most common (64 CVs) with a slightly above-average score (8.3).
| Nationality | Average of Score | Count of CVs |
|---|---|---|
| Belgian | 9.0 | 1 |
| Eastern European | 9.0 | 1 |
| Guyanese | 9.0 | 1 |
| Spanish | 9.0 | 1 |
| French | 9.0 | 1 |
| Other Dual Citizenship | 8.5 | 2 |
| British Dual Citizenship | 8.4 | 11 |
| British | 8.3 | 64 |
| Dutch | 8.0 | 1 |
| Romanian | 8.0 | 2 |
| Estonian | 8.0 | 1 |
| Greek | 8.0 | 1 |
| Turkish | 8.0 | 1 |
| Bulgarian | 8.0 | 1 |
| UAE | 8.0 | 1 |
| Not Indicated | 7.7 | 8 |
| Italian | 7.0 | 1 |
| Serbian | 7.0 | 1 |
| Grand Total | 8.3 | 100 |
Employment Gaps
- Impact of Employment Gaps: CVs with 3-year gaps scored the highest (8.5), but this is based on only 2 CVs. Generally, the presence of employment gaps doesn't seem to significantly affect scores.
| Employment Gaps | Average of Score | Count of CVs |
|---|---|---|
| 3 years | 8.5 | 2 |
| Less than 1 year | 8.3 | 15 |
| None | 8.3 | 69 |
| 1 year | 8.1 | 9 |
| 7 years | 8.0 | 1 |
| 2 years | 7.0 | 4 |
| Grand Total | 8.3 | 100 |
Location
- Top Locations: Bristol, Hertfordshire, and Buckinghamshire each scored the highest (9.0), but each is represented by only 1 CV.
- Most Common Location: London is the most common location (61 CVs) with an average score (8.3).
| Location | Average of Score | Count of CVs |
|---|---|---|
| Bristol | 9.0 | 1 |
| Hertfordshire | 9.0 | 1 |
| Buckinghamshire | 9.0 | 1 |
| UK (No city/town mentioned) | 8.6 | 17 |
| Essex | 8.3 | 6 |
| London | 8.3 | 61 |
| Surrey | 8.0 | 4 |
| Bolton | 8.0 | 1 |
| Beckenham | 8.0 | 1 |
| Dubai | 8.0 | 1 |
| Not Indicated | 7.8 | 4 |
| Chesham | 6.0 | 1 |
| Grand Total | 8.3 | 100 |
CV Enhancement Data
- Impact of Enhancements: Enhanced CVs scored significantly higher (9.4) than normal CVs (8.3).
- Most Common Enhancements: The profile section received the most enhancements, suggesting that the AI tool places significant importance on this part of the CV.
| Type of CV | |
|---|---|
| Enhanced CVs Avg. Score | 9.4 |
| Normal CVs Avg. Score | 8.3 |
| Enhancements to CVs | |
|---|---|
| Enhanced to Profile | 7 |
| Enhancements to Key Skills & Attributes | 4 |
| Enhancements to Professional Experience | 3 |
| Total | 14 |
Overall Insights
- Education Level: The AI tool does not heavily favour higher education levels, as seen in the close scoring range.
- Ethnicity and Nationality: The data suggests potential biases, but the small sample size for certain groups limits the reliability of these conclusions.
- Age: There appears to be no significant bias towards a specific age group.
- Location: The data suggests potential geographic biases, but again, the small sample size for top-scoring locations is a limiting factor.
- AI CV Enhancements: The significant score increase for enhanced CVs raises ethical questions about authenticity versus optimisation.Questions were about the role of AI in amplifying a candidate's appeal and the ethical considerations it entails.
The Impact of AI-Enhanced CVs
The study's exploration into AI-enhanced CVs yielded significant insights. When CVs were modified using AI (ChatGPT) to better align with the job description, there was a marked increase in the scores assigned by the AI screening tool. The average score for these enhanced CVs was 9.4, compared to 8.3 for the original, unaltered CVs. This substantial increase highlights the impact of AI enhancements in potentially elevating a candidate's appeal.
However, it also raises ethical concerns. The enhancements, which included additions to the profile, key skills and attributes, and professional experience sections, might risk misrepresenting the candidate's actual qualifications and experiences. This phenomenon suggests a delicate balance between optimising a CV to reflect a candidate's potential and inadvertently creating a misleading portrayal. The findings highlight the need for careful consideration in the use of AI for CV enhancement, ensuring that it aids in fair representation rather than distorting a candidate's true profile.
Navigating the Future of AI in Recruitment
As we conclude our study on AI bias in recruitment, it's clear that while AI offers remarkable efficiencies, it also brings challenges that need careful navigation. Our findings highlight the need for a balanced approach in AI application, ensuring fairness and authenticity in the recruitment process.
- For Employers: It's crucial to develop strategies to identify AI-enhanced CVs and to complement AI screenings with human judgement and in-depth interviews. This approach will help maintain a level playing field and ensure that the best candidates are chosen based on their true capabilities.
- For Job Seekers: Authenticity in job applications remains paramount. While AI can assist in highlighting your skills, it's important to ensure that your CV accurately reflects your experiences and qualifications.
Join the Conversation on Ethical AI in Recruitment
Your insights and experiences are invaluable in shaping the future of AI in recruitment. We invite you to join us in this important conversation. Whether you're an employer navigating the complexities of recruitment, or a job seeker looking for your dream role, your voice matters.
- Employers: Discover strategies to integrate AI ethically into your recruitment processes. Contact us for expert advice and tailored solutions.
- Job Seekers: Learn how to present your skills effectively while maintaining the integrity of your application. Explore our resources for guidance and support.
Together, let's ensure a fair, transparent, and inclusive recruitment landscape. Get involved and help shape a future where technology empowers, not overshadows, human potential.
