When it comes to violence against women and girls, shifting the mindsets of judicial actors is critical for proper redress for survivors. Yet, manual case law analysis takes 1 hour per case, when accounting for review, analysis, and validation, making scaling across jurisdictions impossible.
However, recent advances in Large Language Models (LLMs) now make automation viable, creating an opportunity for massive impact.
Our initiative, ImpartialAI, is putting this technology into practice with the human rights infrastructure baked into its development and design. This tool will be gold standard justice-tech, ready to scale to jurisdictions first in the Commonwealth and then around the world.
Our Progress to Date:
- Our automated data pipelines are now parsing court websites, identifying GBV judgments, and extracting 60+ sentencing variables in seconds
- We have achieved over 90% accuracy on 65% of the features we track. The team has been working to ensure the system reviews complex interpretive indicators accurately and consistently. We are on track for >90% precision and recall across all features.
- We’re working with a team from MIT Sloan’s GenAI Lab to provide an independent review of our LLM performance and "human-in-the-loop" architecture.
By automating the detection of gender stereotypes (like victim-blaming and rape myths), we provide advocates with near-real-time evidence to demand judicial reform.