Transparency challenges in policy evaluation with causal machine learning -- improving usability and accountability

Black-box causal models create accountability gaps in government policy decisions: The paper identifies a critical problem where causal machine learning tools used to evaluate policies (such as estimating effects of changing school leaving age) lack transparency, making it difficult for government agencies to explain their decisions to the public or demonstrate that models are functioning fairly.

Existing explainability tools don't work well for causal analysis: The research found that standard explainable AI tools designed for predictive models are "poorly suited to causal machine learning," meaning transparency solutions borrowed from other AI applications cannot simply be applied to policy evaluation contexts.

There's a trade-off between accuracy and interpretability with no easy solution: When the researchers simplified their causal forest model to make it more interpretable and transparent, it resulted in "an unacceptable increase in error," suggesting that policymakers face a difficult choice between using accurate but opaque models or transparent but less accurate ones.

New specialized tools are needed for responsible policy AI: The paper concludes that the field requires new transparency and explainability tools specifically designed for causal machine learning, rather than adapting existing solutions, to enable proper government accountability and public trust in AI-assisted policy decisions.

Transparency in policy evaluation is a governance issue, not just a technical one: The research emphasizes that transparency problems in causal machine learning are particularly acute in government contexts where democratic accountability and fair decision-making are essential requirements.