Summary of X Hacking: the Threat Of Misguided Automl, by Rahul Sharma et al.
X Hacking: The Threat of Misguided AutoML
by Rahul Sharma, Sergey Redyuk, Sumantrak Mukherjee, Andrea Sipka, Sebastian Vollmer, David Selby
First submitted to arxiv on: 16 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Cryptography and Security (cs.CR)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A new paper in Explainable AI (XAI) investigates a concerning issue: analysts manipulating XAI metrics to support pre-specified conclusions. The authors introduce the concept of X-hacking, a type of p-hacking applied to XAI metrics like Shap values. They demonstrate how an automated machine learning pipeline can be used to search for models that produce desired explanations while maintaining superior predictive performance. The paper formulates the trade-off between explanation and accuracy as a multi-objective optimization problem and empirically illustrates the feasibility and severity of X-hacking on real-world datasets. Finally, it suggests methods for detection and prevention, discussing ethical implications for the credibility and reproducibility of XAI research. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how some people might try to manipulate machine learning models so they can say their predictions make sense, even if they don’t really. This is a problem because it makes it hard to trust what the models are saying. The authors call this “X-hacking” and show that it’s possible to find models that give you the explanation you want while still being good at making predictions. They also talk about how to prevent this from happening and what it means for our understanding of AI. |
Keywords
* Artificial intelligence * Machine learning * Optimization
