Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Aug 4·5 min read
Who Should Be Generated? Justifying Demographic Targets in Open-Ended Generation
TL;DR
A framework for justifying demographic targets in generative models was developed, revealing significant discrepancies between generated outputs and geographic targets. This highlights the need for explicit justification in fairness evaluations.
✦ Why It Matters
Evaluate your generative models against justified demographic targets to ensure fairness in outputs.
Key Takeaways
- Understand the importance of justified demographic targets in generative models.
- Utilize the proposed framework for evaluating fairness in outputs.
- Assess your model's demographic distribution against geographic-derived targets.
- Consider replacing targets with equal-category comparators for better fairness assessment.
How It Works
The framework decomposes target construction into four commitments, allowing for a structured approach to defining demographic targets. It incorporates geographic and occupational priors to ensure that the generated outputs align with real-world distributions.