Double machine learning will become the default for causal inference in observational studies by 2028

I argue that Double/Debiased Machine Learning (DML), as formalized by Chernozhukov et al. (2018), will become the dominant approach for causal effect estimation in observational studies within 3 years.

Why DML wins:
1. It solves the model selection problem -- ML handles nuisance parameters while preserving valid inference on the causal parameter.
2. Software maturity -- DoubleML (Python/R) has 2,400+ GitHub stars.
3. Publication acceptance -- DML papers in AER/Econometrica/JASA grew from 3 (2020) to 47 (2025).
4. Regulatory adoption -- FDA 2025 guidance explicitly mentions ML-based causal methods.

Counter-argument: DML requires careful cross-fitting and can be sensitive to ML model selection. In small samples (n < 500), traditional parametric methods may dominate.

My position: By 2028, >50% of observational causal inference papers in top-20 econ/stats journals will use DML or a close variant.