GOODHART'S LAW AND MACHINE LEARNING: A STRUCTURAL PERSPECTIVE
Open accessAbstract
We develop a simple structural model to illustrate how penalized regressions generate Goodhart bias when training data are clean but covariates are manipulated at known cost by future agents. With quadratic (extremely steep) manipulation costs, bias is proportional to Ridge (Lasso) penalization. If costs depend on absolute or percentage manipulation, the following algorithm yields manipulation‐proof prediction: Within training data, evaluate candidate coefficients at their respective incentive‐compatible manipulation configuration. We derive analytical coefficient adjustments: slopes (intercept) shift downward if costs depend on percentage (absolute) manipulation. Statisticians ignoring manipulation costs select socially suboptimal penalization. Model averaging reduces these manipulation costs.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
While such a proxy may robustly correlate with our intuitive judgements of harm in normal situations, they may still reliably come apart if those proxies are used as an optimisation target.
“However, while such a proxy may robustly correlate with our intuitive judgements of harm in normal situations, they may still reliably come apart if those proxies are used as an optimisation target.”
From Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems (2024) · cites this paper as Hennessy & Goodhart (2023)
The Introduction describes predictions fitted to clean training data that become biased when future agents manipulate the covariates used for prediction. That supports the general point that a useful proxy can diverge from what it predicts when incentives target it. The paper does not address proxies for intuitive judgements of harm specifically. (LLM, conf=0.85)
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
No references could be resolved yet.
Research works printed but not linked: 11
- Ball, I., "Scoring Strategic Agents," arXiv Working Paper, 2022.
- Björkegren, D., J. E. Blumenstock, and S. Knight, "Manipulation-Proof Machine Learning," arXiv Working Paper, 2020.
- Bruckner, M., and T. Scheffer, "Stackelberg Games for Adversarial Prediction Problems," Journal of Machine Learning Research 13 (2011), 2617-54.
- Dekel, O., F. Fischer, and ---, "Incentive Compatible Regression Learning," Journal of Computing System Science 76 (2010), 759-77.
- Eliaz, K., and R. Spiegler, "The Model Selection Curse," American Economic Review: Insights, 1 (2019), 127-40.
- Fernández-Villaverde, J., "Has Machine Learning Rendered Simple Rules Obsolete?" European Jour- nal of Law and Economics 52 (2021), 251-65.
- Frankel, A., and N. Kartik, "Improving Information from Manipulable Data," Journal of the European Economic Association 20 (2022), 79-115.
- ---, and Kartik, N.,"Muddled Information," Journal of Political Economy 129 (2019), 1739-76.
- Hardt, M., N. Megiddo, C. Papadimitriou, and M. Wooters, "Strategic Classification," Proceedings of the 7th Innovations in Theoretical Computer Science Conference (2016), 111-22.
- Lucas, R., "Econometric Policy Evaluation: A Critique," Carnegie-Rochester Conference Series on Pub- lic Policy 1 (1976), 19-46.
- Parviero, R., "Improving Ridge Regression via Model Selection and Focussed Fine-Tuning," Thesis, Scuola di Economia e Statistica, Milano-Bicocca, 2017. 14682354, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/iere.12633 by London School Of Economics And, Wiley Online Library on [18/04/2023]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License · DOI
Cited by
Sorted by citations
Towards Guaranteed Safe AI: A Framework for Ensuring Robust and Reliable AI Systems
Dalrymple, D. "., Skalse, J., Bengio, Y., Russell, S., Tegmark, M., Seshia, S., Omohundro, S., Szegedy, C., Goldhaber, B., Ammann, N., et al. (2024) · arXiv · doi:10.48550/arxiv.2405.06624 · Cited by 12 · see citing works
- Authors
- Hennessy, C. A., & Goodhart, C. A. E.
- Affiliations
- London Business School (GB)
- London School of Economics and Political Science (GB)
- Corresponding author
- Hennessy, Christopher A.
- Venue
- International Economic Review
- Publisher
- Wiley
- Published
- 2023-04-10
- Volume
- 64
- Issue
- 3
- Pages
- 1075-1086
- ISSN
- 0020-6598 (ISSN-L)
- 1468-2354
- OpenAlex
- W4328128375
- Cited by
- 8
- References
- 0
- Citations by year
- 2026: 4 · 2025: 1 · 2024: 3
- FWCI
- 1.62
- Type
- journal article
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- Locations
- International Economic Review (publishedVersion, cc-by-nc-nd)
- London Business School Research Online (London Business School) (acceptedVersion)
- London School of Economics and Political Science Research Online (London School of Economics and Political Science) (acceptedVersion)
- London School of Economics and Political Science Theses Online (London School of Economics and Political Science) (acceptedVersion)
- RePEc: Research Papers in Economics (submittedVersion)
- Funders
- Economic and Social Research Council (ES/K002309/1, ES/R009724/1)
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Inference
- Physical Sciences › Mathematics › Statistics and Probability › Advanced Statistical Methods and Models
- Physical Sciences › Mathematics › Statistics and Probability › Advanced Causal Inference Techniques
- Keywords
- Goodhart's law, penalized regression, ridge regression, LASSO, incentive compatibility, structural model, model averaging, prediction bias, social optimum
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
