People + Planet + Parity Governance Framework
Joshua Williams · Team Build People + Planet + Parity AI
Submitted to Apart x Martian Mechanistic Router Interpretability Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.
This project introduces a governance framework to improve interpretability and ethical alignment in AI routing systems. It tackles challenges like late-stage ethics integration, disconnected fairness metrics, and ambiguous accountability by embedding ethical governance throughout all deployment stages.
The framework utilizes three dedicated “Judges” (Accessibility, Carbon, and Bias) to evaluate AI outputs against WCAG 2.2, SCI, and OWASP GenAI standards. These assessments guide routing decisions and ensure ethical oversight is measurable and auditable.
Aligned with Track 2: Intelligent Routing Systems, this solution helps AI teams address bias, environmental impact, and accessibility issues in real-time, fostering safer, fairer, and more sustainable AI systems prioritizing People + Planet + Parity.
Reviews
Decomposing ethics into modular judges is compelling. Reliance on third party standards outsource quality metrics, which can be a strength or a weakness depending on third party metric robustness. I'd be curious how this can be translated to technical solutions.
The paper surfaces relevant considerations for routing systems. It would be great to move to the implementation and evaluation stage to judge whether the system would actually work as expected and experience potential trade-offs between the proposed dimensions for judges and other desiderata for model outputs.
I really like the idea of adding a "carbon" judge that looks also at the environmental impact. Just to ask, is it not likely that this would correlate strongly with simply minimizing the dollar cost (since larger models are expensive to run)?
Under the people + parity + planet framework, it might also be interesting to consider things like robustness to other languages (under accessibility), or say perceptiveness to emotions(under people). I think this framework is a good starting point that could be built upon further.
Cite this project
@misc{williams2025people,
title = {{People + Planet + Parity Governance Framework}},
author = {Joshua Williams},
year = {2025},
month = jun,
note = {Submitted to Apart x Martian Mechanistic Router Interpretability Hackathon, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/people-planet-parity-governance-framework-h3ks}},
url = {https://apartresearch.com/sprints/projects/people-planet-parity-governance-framework-h3ks}
}More from Apart x Martian Mechanistic Router Interpretability Hackathon
- 1st place by peer reviewView project: Manipulating Self-Preference for Large Language Models
Manipulating Self-Preference for Large Language Models
Team Preference
Large language models (LLMs) carry great value as evaluators of synthetic data for research and production settings. However, recent research shows that language models exhibit bias towards their own responses in blind …
- 2nd place by peer reviewView project: Approximating Human Preferences Using a Multi-Judge Learned System
Approximating Human Preferences Using a Multi-Judge Learned System
AutoBox
In this work, we introduced a learned approach to aggregating multi-judge scores: using a GAM and a simple MLP as an alternative to traditional, non-learned methods like averaging. Our models outperform the naive …
- 3rd place by peer reviewView project: Judge using SAE Features
Judge using SAE Features
SAEwhat?
The key idea of this project was to explore model judgement using Sparse Autoencoder (SAE) features for mathematical reasoning tasks involving addition, multiplication, and subtraction operations. We compared this …