Bias Scorer Example

This example demonstrates how to use Mastra's Bias Scorer to evaluate LLM-generated responses for various forms of bias.

Prerequisites

• Node.js 22.13.0 or later
• pnpm
• OpenAI API key

Getting Started

1. Clone the repository and navigate to the project directory:

   bashCopy

   git clone https://github.com/mastra-ai/mastra
   cd examples/basics/scorers/bias
   

2. Copy the environment variables file and add your OpenAI API key:

   bashCopy

   cp .env.example .env
   

   Then edit .env and add your OpenAI API key:

   envCopy

   OPENAI_API_KEY=sk-your-api-key-here
   

3. Install dependencies:

   bashCopy

   pnpm install --ignore-workspace
   

4. Run the example:

   bashCopy

   pnpm start
   

Overview

The Bias Scorer evaluates responses for various forms of bias, including:

• Gender bias
• Political bias
• Racial/ethnic bias
• Geographical bias
• Cultural bias
• Age bias

Example Structure

The example includes three scenarios:

1. High Bias: Testing a response with clear gender bias about leadership styles
2. Mixed Bias: Testing a response with age-related stereotypes about work performance
3. Low Bias: Testing a response about fair hiring practices with minimal bias

Each scenario demonstrates:

• Setting up the scorer with the language model
• Providing input questions and responses to evaluate
• Running the bias evaluation
• Interpreting the results with detailed reasoning

Expected Output

The example will output:

• The input query and response for each scenario
• The scorer result with:
  • Score (0-1, where 1 indicates high bias and 0 indicates minimal bias)
  • Detailed reasoning about any detected bias

Key Components

• createBiasScorer: Function that creates the bias scorer instance
• Scorer configuration:
  • model: The language model to use for evaluation (e.g., OpenAI GPT-4)
  • options: Optional configuration (e.g., scale factor)
• scorer.run(): Method to evaluate input/output pairs for bias
  • Takes { input, output } where:
    • input: Array of chat messages (e.g., [{ role: 'user', content: 'question' }])
    • output: Response object (e.g., { role: 'assistant', text: 'response' })
  • Returns { score, reason } with numerical score and explanation