EquiJob: Using AI to Balance Bias in Job Descriptions

Local Chapter Toronto, Canada Chapter

Coordinated byCanada ,

Kausthab Dutta Phukan

Status: Ongoing

Project background.

Despite significant advancements in diversity and inclusion, employment discrimination remains a pervasive issue globally. According to a 2020 World Economic Forum report, gender parity alone is estimated not to be achieved for another 257 years. Unconscious bias often manifests in job descriptions, with a study from the University of Waterloo and Duke University revealing that gender-coded language can discourage certain applicants.

For instance, the use of words like ‘dominant’ or ‘competitive’ can dissuade female candidates, while ‘collaborative’ or ‘supportive’ may deter male applicants. Similarly, age bias, often subtle, can deter qualified candidates from applying.

‘EquiJob: Using AI to Balance Bias in Job Descriptions’ aims to address this issue by creating an AI tool that can detect and highlight potentially biased language in job descriptions, promoting more inclusive hiring practices.

The problem.

This challenge will tackle the very evident and rampant problem of discrimination that ails the job market today. This is a project that will be country agnostic. We will first start with North America and maybe then scale it up to the Rest of the World

Project goals.

- Develop an AI-based Tool. - Understand the implementation power of NLP. - Build a User-Friendly Interface. - Improve Diversity and Inclusion. - Influence Industry Practices.

Project plan.

  • Week 1

    – Project Setup & Planning

  • Week 2

    – Data Collection & Preprocessing
    * Collect job descriptions from various sources.
    * Preprocess the data (cleaning, standardizing, etc.)

  • Week 3

    – Annotation & Ground Truth Development
    * Develop annotation guidelines.
    * Annotate a subset of data and refine guidelines.
    * Complete annotation of the dataset.

  • Week 4

    – Model Selection & Training
    * Select the appropriate NLP model and implement the baseline version.
    * Fine-tune model on annotated data.
    * Evaluate initial model performance and adjust as needed.

  • Week 5

    – Model Refinement & Testing
    * Continue fine-tuning the model, incorporating any insights from initial testing.
    * Conduct thorough testing of the model’s performance on unseen data.
    * Begin work on the user-friendly interface.

  • Week 6

    Week 6 – Integration & User Testing * Complete user interface and integrate with the AI model * Conduct user testing, gather feedback *: Refine product based on feedback, prepare for launch

Share project on: