Discrimination based on race and ethnicity in personnel selection is a well known and pervasive issue highlighted in numerous studies (Bertrand & Mullainathan, 2004; Kline et al., 2021; Pager et al., 2009).
Most of these studies report name-based inference of race and ethnicity by human reviewers leading to differential outcomes in the recruitment process. Linguistic racism is a form of discrimination that occurs based on one’s use of language, especially English (De Costa, 2020), and is highly associated with race and ethnicity.
As machine learning models are adopted to automate tasks like interview scoring, race or ethnicity encoded signals in language can lead to biased outcomes, if not mitigated. Hence understanding the level of ethnicity encoded signals in language is important when building natural language-based machine learning models in order to avoid biased outcomes, for example by using feature scores rather than raw text to score responses (Jayaratne, Jayatilleke, Dai, 2022).
In this work, we sought to quantify and compare the amount of ethnicity encoded information in over 300,000 candidates’ raw text interview responses to language-derived feature scores, including personality, behavioral competencies, and communication skills.
First, we trained machine learning models to predict candidate ethnicity from raw-text chat interview responses. Specifically, we trained an Attention-Based Bidirectional Long Short-Term Memory (Attn-BiLSTM) (Zhou et al., 2016) model for predicting ethnicity from textual responses.
Secondly, we tested the same for the language-derived features used in the automated scoring of the interview responses. We trained multiple models using a variety of machine learning algorithms (a linear model, tree models with bagging and boosting, and a neural network model with a single hidden layer) suitable for tabular data for predicting ethnicity from the 21 derived features.
Each model was then used to predict ethnicity for the 10% of the sample left out of the training dataset. The results from the classification tasks show a clear distinction between the ability to infer ethnicity based on natural language and inferred features. As hypothesized, we found that features derived according to a clearly defined rubric contain significantly less ethnicity information compared to raw candidate responses. That is, the models based on derived features recorded consistently weaker accuracy, precision, recall, and F1 values across all models compared to the model for the raw text candidate responses.
This research demonstrates the benefit of using algorithmically derived feature values in mitigating ethnicity related biases when scoring structured interview responses. Specifically, our results show that natural language responses to interview questions carry higher amounts of ethnicity information compared to features derived according to a clearly defined rubric for assessing interview responses. This further strengthens the case for using structured interviews that have been shown to reduce bias over unstructured interviews (Levashina et al., 2014) with much stronger criterion validity (Sackett et al., 2021).
References:
Bertrand, M., & Mullainathan, S. (2004). Are Emily and Greg More Employable Than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination. American Economic Review, 94(4), 991–1013.
De Costa, P. I. (2020). Linguistic racism: Its negative effects and why we need to contest it. International Journal of Bilingual Education and Bilingualism, 23(7), 833–837.
Jayaratne, M., Jayatilleke, B., & Yimeng Dai (2022). Identifying and mitigating gender bias in structured interview responses [Paper presentation]. 2022 Society for Industrial Organizational Psychology Conference. Seattle, Washington, United States.
Kline, P. M., Rose, E. K., & Walters, C. R. (2021). Systemic Discrimination Among Large U.S. Employers (NBER Working Papers No. 29053). National Bureau of Economic Research, Inc.
Levashina, J., Hartwell, C. J., Morgeson, F. P., & Campion, M. A. (2014). The Structured Employment Interview: Narrative and Quantitative Review of the Research Literature. Personnel Psychology, 67(1), 241–293.
Pager, D., Western, B., & Bonikowski, B. (2009). Discrimination in a Low-Wage Labor Market: A Field Experiment. American Sociological Review, 74(5), 777–799.
Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2021). Revisiting meta-analytic estimates of validity in personnel selection: Addressing systematic overcorrection for restriction of range. Journal of Applied Psychology.
Zhou, P., Shi, W., Tian, J., Qi, Z., Li, B., Hao, H., & Xu, B. (2016). Attention-Based Bidirectional Long Short-Term Memory Networks for Relation Classification. Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), 207–212.
Related
Blog
Sapia.ai Wrapped 2024
It’s been a year of Big Moves at Sapia.ai. From welcoming groundbreaking brands to achieving incredible milestones in our product innovation and scale, we’re pushing the boundaries of what’s possible in hiring.
And we’re just getting started 🚀
Take a look at the highlights of 2024
All-in-one hiring platform
This year, with the addition of Live Interview, we’re proud to say our platform now covers screening, assessing and scheduling.
It’s an all-in-one volume hiring platform that enables our customers to deliver a world-leading experience from application through to offer.
Supercharging hiring efficiency
Every 15 seconds, a candidate is interviewed with Sapia.ai.
This year, we’ve saved hiring managers and recruiters hours of precious time that can now be used for higher-value tasks.
Giving candidates the best experience
Our platform allows candidates to be their best selves, so our customers can find the people that truly belong with them. They’re proud to use a technology that’s changing hiring, for good.
Leading the way in AI for hiring
We’ve continued to push the boundaries in leveraging ethical AI for hiring, with new products on the way for Coaching, Internal Mobility & Interview Builders.
Blog
Situational Judgement Tests vs. AI Chat Interviews: A Modern Perspective on Candidate Assessment
Choosing the right tool for assessing candidates can be challenging. For years, situational judgement tests (SJTs) have been a common choice for evaluating behaviour and decision-making skills. However, they come with limitations that can make the hiring process less effective and less inclusive.
AI-enabled chat-based interviews, such as Sapia.ai, provide organisations with a modern alternative. They focus on understanding candidates as individuals and creating a hiring experience that is both fair and insightful while enabling efficient screening and selection.
This shift raises important questions: Are SJTs still a tool that should be considered for volume hiring? And what do AI assessments offer in comparison?
1. The Static Nature of SJTs
Traditional SJTs use predefined multiple-choice questions to assess behavioural tendencies and situational knowledge. While useful for screening, these static frameworks lack the flexibility to adapt based on real-world performance data or evolving role requirements.
Once created, SJTs don’t adapt to new data or evolving organisational needs. They rely on fixed scenarios and responses that may not fully reflect the dynamic realities of modern workplaces, and as a result, their relevance may diminish over time.
AI-enabled chat interviews, on the other hand, are inherently adaptive. Using machine learning, these tools can continuously refine their models based on feedback from real-world outcomes such as hiring or turnover data. This ability to evolve ensures the assessments align with organisations’ needs.
2. Richer Data Through Open-Ended Responses
One of the main critiques of SJTs is their reliance on multiple-choice responses. While structured and straightforward, these options may not capture the full scope of a candidate’s thinking, communication skills, or problem-solving ability. The approach is often limiting, reducing complex human behaviour to a few predefined choices.
AI-enabled chat interviews work more holistically and dynamically. These tools provide a more complete picture of a person by allowing candidates to answer questions in their own words. Natural language processing (NLP) analyses their responses, offering insights into personality traits, communication skills, and behavioural tendencies. This open-ended format lets candidates express themselves authentically, giving employers a deeper understanding of their potential.
3. The Candidate Experience: Stressful or Supportive?
SJTs often include time constraints and rigid formats, which can create pressure for candidates. This is especially true when candidates feel forced to choose options that don’t fully reflect how they would actually behave. The process can feel impersonal, even transactional.
In contrast, chat-based interviews are designed to be conversational and low-pressure for candidates. By removing time limits and adopting a familiar chat interface, these tools help candidates feel more at ease. They also frequently include personalised feedback, turning the assessment into a valuable experience for the candidate, not just the employer.
4. Addressing Bias and Fairness
Traditional SJTs are prone to transparency issues, as candidates can often identify and select the “best practice” answers without revealing their true tendencies. Additionally, static test designs can unintentionally embed bias; due to the nature of the timed test, SJTs have been found to disadvantage some groups.
AI chat interviews, when developed ethically within a framework like Sapia.ai’s FAIR Hiring Framework, eliminate explicit bias by relying solely on the content of a candidate’s responses. Their machine learning models are continuously validated for fairness, ensuring that hiring decisions are free from subjective judgments or irrelevant demographic factors.
5. An Assessment That Improves Over Time
Workplaces are constantly changing, and hiring tools need to keep up. SJTs’ fixed nature can make them less effective as roles evolve or organizational priorities shift. They provide a snapshot but not a dynamic view of what’s needed.
AI-enabled chat interviews are built to adapt. With feedback loops and continuous learning, they incorporate real-world hiring outcomes—like retention and performance data—into their models. This ensures that assessments stay relevant and effective over time.
Rethinking Candidate Assessment
As hiring demands grow more complex, so does the need for tools that can capture the whole person, not just their response to hypothetical scenarios. While SJTs have played an important role in hiring practices, they are increasingly being replaced by tools like AI-enabled chat interviews.
These modern approaches provide richer data, adapt to changing needs, and create a richer and more engaging experience for candidates. Perhaps most importantly, they emphasise fairness and inclusivity, aligning with the growing demand for unbiased hiring practices.
For organisations evaluating their assessment tools, the question isn’t just which method is “better.” Understanding the specific needs of your roles, teams, and candidates will help you choose tools that help you make decisions that are both informed and equitable.
Blog
Keeping Interviews Real with Next-Gen AI Detection
It’s our firm belief that AI should empower, not overshadow, human potential. While AI tools like ChatGPT are brilliant at assisting us with day-to-day tasks and improving our work efficiency, employers are increasingly concerned that they’re holding candidates back from revealing their true, authentic selves in online interviews.
As an assessment technology provider, we are responsible for ensuring the authenticity and integrity of our platform. That’s why we’re thrilled to unveil the latest upgrade to our flagship Chat Interview: the AI-Generated Content Detector 2.0. With groundbreaking accuracy and a candidate-friendly design, this innovation reinforces our mission to build ethical AI for hiring that people love.
Artificially Generated Content (AGC) is content created by an AI tool, such as ChatGPT, Claude, or Pi. We initially rolled out the first version of our AGC detector last year and have continued to improve it as our data set has grown and these AI tools have evolved.
What’s New?
Our updated AGC Detector 2.0 achieves an impressive 98% detection rate for AI-assisted responses, with a false positive rate of just 1%. This gives organisations peace of mind that they’re getting the most authentic assessment of every candidate.
This cutting-edge system builds on Sapia.ai’s proprietary dataset of over 2 billion words, derived from more than 20 million interview question-answer pairs spanning diverse roles, industries, and regions. It’s trained on real-world data collected before and after the release of tools like ChatGPT, ensuring it remains robust and reliable even as AI tools evolve.
The Challenge of AI in Chat-based Interviews
Our data shows that around 8% of candidates use tools like GPT-4 to generate responses for three or more interview questions. While these tools may offer a quick way for candidates to complete their interview, they can inadvertently hide a person’s true personality and potential – qualities our customers are most interested in understanding through our platform. In fact, research from Sapia Labs shows that these tools have their own personality traits, which may be quite different from the candidate applying for the role.
For Candidates: Enabling Authenticity
When a response is flagged as potentially AI-generated, the system doesn’t disqualify candidates. Instead, a real-time warning pops up, allowing them to revise their answers or submit them as-is. This ensures that candidates are encouraged to present themselves authentically, reflecting their unique communication styles and sharing their genuine experiences.
For Hiring Teams: Actionable Insights
Responses flagged as AI-generated are highlighted in the candidate’s Talent Insights profile, accessible via Sapia.ai’s Talent Hub or ATS integrations. These insights give hiring teams the transparency to make informed decisions, fostering trust while accelerating hiring timelines.
Built on Unmatched AI Interview Expertise
“Our detection model’s strength lies in its foundation of real-world interview data collected from diverse roles and regions,” says Dr Buddhi Jayatilleke, Sapia.ai’s Chief Data Scientist. This depth of understanding enables the AGC Detector to maintain its industry-leading accuracy – even when candidates subtly modify AI-generated answers to appear more human.
Why This Matters
The AGC Detector 2.0 embodies Sapia.ai’s commitment to ethical AI that amplifies human potential. As our CEO Barb Hyman explains:
“The hiring landscape has fundamentally changed since ChatGPT, but our commitment remains clear: AI should amplify human potential, not penalise it. This breakthrough fosters authentic hiring conversations. Our real-time warning system helps candidates make better choices and gives enterprises confidence in their selection decisions.”
Testing and Validation of the AGC Detector 2.0
The new detector has been rigorously tested on over 25,000 interview responses generated by humans and leading AI models like GPT-4, Claude-3.5, and Llama-3. The results speak for themselves, reinforcing the reliability and fairness of this game-changing technology.
Fairness & Transparency in AI-Enabled Hiring
By detecting AI-generated content while allowing candidates to correct their responses, our AGC Detector 2.0 ensures every applicant has the chance to put their best, most authentic foot forward when applying for a role powered by Sapia.ai. For enterprises, it provides confidence in the integrity of their hiring decisions and ensures they’re connecting with real candidates at scale.