Research Proposal: AI Bias in Hiring and Surveillance

Part 1: Information Collection

Recommended Frameworks & Rationale

• Framework 1: Intersectional Analysis
  • Teaching: Intersectionality is a framework for understanding how different aspects of a person's social and political identities (e.g., race, gender, class, disability) combine to create unique experiences of discrimination and privilege. Instead of looking at bias along a single axis (like just gender), it examines the compounded effects (e.g., the specific bias faced by an older woman of color).
  • Applicability: This is essential for your first objective to "compare AI bias impacts across different demographic groups." It ensures you don't just find that "AI is biased against women and racial minorities," but can articulate the unique, amplified bias experienced by individuals at the intersection of these identities.
• Framework 2: Causal Loop Diagram (Systems Thinking)
  • Teaching: A Causal Loop Diagram is a tool used to visualize complex systems and understand how different variables interact. It helps map out feedback loops—where the output of an action feeds back to influence the input, either amplifying (reinforcing loop) or stabilizing (balancing loop) the effect.
  • Applicability: This is perfect for your second objective: "understand the mechanisms of bias amplification." An AI hiring system is a classic complex system. Biased historical data leads to a biased algorithm, which makes biased decisions, creating new biased data that is then fed back into the system, creating a reinforcing loop of inequality. This framework will help you map and explain that process visually.
• Framework 3: Policy Gap Analysis
  • Teaching: This is a straightforward method for evaluating the effectiveness of a policy or law. You start by identifying the stated goals of the policy. Then, you collect evidence on the actual outcomes. The "gap" is the difference between the intended goals and the reality on the ground.
  • Applicability: This directly addresses your third objective to "evaluate legal/policy frameworks." It provides a clear structure to compare what regulations like the EU AI Act or EEOC guidelines are supposed to do versus what they are actually achieving in practice, based on the evidence you collect.

Information Gathering Plan

A. Web Search Content

• For Intersectional Analysis:
  • Search Queries: "AI hiring bias racial and gender intersectionality," "studies on AI surveillance impact on workers with disabilities 2024," "algorithmic discrimination against older women in hiring," "AI bias against Black men in recruiting 2025".
  • Purpose: The goal is to collect quantitative data and qualitative findings from recent academic studies, NGO reports, and journalistic investigations. This evidence will form the factual basis for comparing how different combinations of demographic traits lead to unique and disproportionate harms. For example, some studies show AI systems penalize Black men's resumes most harshly, an intersectional finding that is more specific than separate analyses of race and gender bias.
• For Causal Loop Diagram:
  • Search Queries: "how AI algorithms create feedback loops in hiring," "mechanisms of algorithmic bias amplification," "proxy discrimination in machine learning models," "challenges in correcting bias in AI training data".
  • Purpose: This search is to understand the technical drivers of bias. You will look for whitepapers and technical articles that explain concepts like how AI models trained on historical data learn to reuse existing biases, and how this is amplified when the model's biased outputs become part of future training data. This information is critical for mapping the variables and feedback loops in your diagram.
• For Policy Gap Analysis:
  • Search Queries: "EEOC enforcement AI hiring tools 2025," "effectiveness of EU AI Act on high-risk AI systems," "compliance challenges with the EU AI Act for HR tech," "joint statements federal agencies AI discrimination 2024".
  • Purpose: You need to gather the official text of regulations, guidance documents from bodies like the EEOC, and analyses of their real-world impact. This will allow you to compare the stated intent of these laws (e.g., to prevent discrimination) with the practical outcomes (e.g., companies struggling to audit their complex systems), which is the core of a gap analysis.

B. User Interview Plan

• Interview Subjects:
  1. Affected Individuals: Individuals from diverse backgrounds (e.g., a woman of color in tech, an older worker re-entering the workforce, a person with a speech disability who has done a video interview) who have recently interacted with automated hiring or workplace surveillance systems.
  2. AI/HR Tech Professionals: Data scientists, product managers, or compliance officers working at companies that build or implement AI hiring tools.
  3. Legal and Policy Experts: Academics, civil rights lawyers, and policy advocates specializing in AI ethics and anti-discrimination law.
• Interview Purpose & Core Questions:
  • Interviews with Affected Individuals (to inform Intersectional Analysis):
    • Purpose: To gather rich, qualitative narratives that bring the quantitative data to life.
    • Core Question: "Can you describe your experience with a recent automated job application or a workplace monitoring system? What parts of the process felt confusing, unfair, or opaque to you?"
    • Analysis Value: These stories provide the "why" and "how" behind the statistics, illustrating the human impact of bias and providing powerful examples for your final report. A person with a disability might describe how an eye-tracking tool flagged them for being disengaged, adding critical context to your findings.
  • Interviews with AI/HR Tech Professionals (to inform the Causal Loop Diagram):
    • Purpose: To understand the practical, on-the-ground mechanisms that create and perpetuate bias.
    • Core Question: "When your team identifies a potential bias in a model, what are the technical and business challenges you face in trying to correct it? What trade-offs do you have to make between fairness, accuracy, and speed?"
    • Analysis Value: This helps you move beyond a purely technical explanation. You can add variables to your Causal Loop Diagram like "Time-to-market pressure" or "Lack of diverse training data," making your model a more accurate reflection of reality.
  • Interviews with Legal and Policy Experts (to inform the Policy Gap Analysis):
    • Purpose: To identify the specific weaknesses and loopholes in current legal frameworks.
    • Core Question: "From your perspective, where is the biggest gap between what a regulation like the EU AI Act intends to do and what is practically enforceable when it comes to auditing a complex 'black box' hiring algorithm?"
    • Analysis Value: These experts can pinpoint specific shortcomings—for example, that a law requires "human oversight" without clearly defining what meaningful oversight looks like. This insight is crucial for developing specific and actionable policy recommendations.

Part 2: Information Analysis

Analysis and Synthesis Plan

1. Develop the Comparative Impact Analysis (Using Intersectional Analysis):
   • Process: Start by synthesizing the findings from your web search and interviews. Create profiles for 2-3 specific intersectional groups (e.g., Black women, men with disabilities, older women). For each profile, document the types of bias encountered (e.g., resume de-ranking, negative sentiment analysis in video interviews) and the evidence (e.g., "Study X found resumes with Black-associated names were ranked 10% lower," "Interviewee Y reported being penalized by a surveillance tool for taking breaks to manage a medical condition").
   • Output: This analysis will directly answer the first part of your research question by providing a textured comparison of how different groups are harmed in unique ways.
2. Construct the Bias Amplification Model (Using Causal Loop Diagram):
   • Process: On a whiteboard or with diagramming software, start by placing your key variables identified from web searches and professional interviews (e.g., "Historical Data Bias," "Algorithm Training," "Biased Hiring Outcomes," "Proxy Variables," "Lack of Data Diversity"). Draw arrows to connect them based on cause and effect. Critically, identify the reinforcing feedback loop: Biased Outcomes → More Biased Training Data for the Next Cycle → Amplified Bias in the Algorithm.
   • Output: This diagram will serve as a powerful visual explanation of the mechanisms of bias amplification, directly addressing the second objective of your research.
3. Formulate the Policy Evaluation and Recommendations (Using Policy Gap Analysis):
   • Process: Create a three-column table for each major policy framework (e.g., EU AI Act, EEOC Guidance).
     • Column 1: Stated Policy Goal (e.g., "Ensure transparency for high-risk AI systems.").
     • Column 2: Observed Reality (e.g., "Expert interviews and industry reports reveal that 'explainability' is often technically difficult and companies provide only vague explanations.").
     • Column 3: The 'Gap' (e.g., "A legal requirement for transparency without a clear, enforceable standard for what constitutes a meaningful explanation.").
   • Output: Use the identified "Gaps" to formulate your specific, actionable recommendations. For the gap above, a recommendation would be: "Policymakers should mandate specific transparency standards, such as requiring companies to disclose the key features and weights used in a hiring decision to the affected individual." This directly addresses the third part of your research and forms the basis of your recommendations. This process will also inform the creation of a best-practices guide for corporations.