Summary of Under Manipulations, Are Some Ai Models Harder to Audit?, by Augustin Godinot et al.
Under manipulations, are some AI models harder to audit?
by Augustin Godinot, Gilles Tredan, Erwan Le Merrer, Camilla Penzo, Francois Taïani
First submitted to arxiv on: 14 Feb 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 This paper explores the feasibility of robust audits for web platforms that comply with laws. Auditors face a significant challenge as they lack access to the algorithm, implementation, or training data used by the platform. The study falls within the framework of manipulation-proof auditing and examines the effectiveness of audit strategies in realistic settings where models exhibit large capacities. The researchers prove a fundamental result: no audit strategy can outperform random sampling when estimating properties like demographic parity if the web platform uses models that can fit any data. To better understand the conditions under which state-of-the-art auditing techniques remain competitive, the study relates the manipulability of audits to the capacity of targeted models using Rademacher complexity. Empirical validation on popular models of increasing capacities confirms that large-capacity models commonly used in practice are particularly hard to audit robustly. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps auditors make sure web platforms follow the law. It’s a big challenge because auditors don’t get to see how the platform works or what data it uses. The study looks at how well different auditing strategies work when dealing with complex models that can learn from any data. They found that some models are too good, making it hard for audits to be effective. This means auditors need new ways to check if platforms comply with the law. |
