Learning with Conflicts of Interest

Ali Vakilian; Arash Termehchy; Marianne Winslett; Nischal Aryal

arxiv: 2605.15504 · v1 · pith:ZIFM76DFnew · submitted 2026-05-15 · 💻 cs.LG · cs.AI

Learning with Conflicts of Interest

Nischal Aryal , Arash Termehchy , Ali Vakilian , Marianne Winslett This is my paper

Pith reviewed 2026-05-19 15:59 UTC · model grok-4.3

classification 💻 cs.LG cs.AI

keywords game theorymachine learningconflicts of interestbias mitigationuser protectionscalable algorithmstheoretical guaranteesinformation filtering

0 comments

The pith

A game-theoretic framework enables users to extract useful information from ML systems while shielding themselves from biased and manipulative outputs even when system owners have conflicting goals.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper contends that because owners of ML systems frequently face financial or political incentives that diverge from user interests, they have little reason to adopt bias-reduction protocols. It therefore models the owner-user relationship as a strategic game in which users can respond to potentially manipulative outputs without requiring any cooperation from the provider. If the approach works, users could continue to draw value from existing ML services while systematically favoring information that supports their own decisions and discarding the rest. A sympathetic reader would see this as a way to keep benefiting from powerful tools without waiting for owners to change their behavior.

Core claim

We propose a game-theoretic framework that models the interaction between ML systems and users with conflicts of interest. We present scalable algorithms with theoretical guarantees that maximize the amount of desired information and actions and minimize the amount of biased and manipulative actions in interaction with ML systems.

What carries the argument

A game-theoretic model of the interaction between an ML system and its users that treats biased or manipulative outputs as strategic choices by the system owner.

Load-bearing premise

Real-world conflicts of interest between ML owners and users can be captured accurately enough by a game model that scalable algorithms with theoretical guarantees can protect users without any cooperation from the owners.

What would settle it

A controlled study in which users following the proposed algorithms still make worse decisions or receive more biased information than users who simply accept all outputs from the same ML system would falsify the central claim.

read the original abstract

Financial, social, and political factors often prevent the interests of the owners of ML systems and services and their users from being perfectly aligned. ML systems often produce biased information that can influence users to make decisions that are not in their best interest. Current solution approaches require ML systems to implement protocols to mitigate their biases. However, ML system owners usually do not have any incentive to implement these protocols and often argue that it limits their freedom of expression or business. We believe that a successful solution to this problem must recognize the conflict of interest between the ML systems and their users, and use this information to protect users against information that adversely influences their decisions while allowing users to safely benefit from these systems. To this end, we propose a game-theoretic framework that models the interaction between ML systems and users with conflicts of interest. We present scalable algorithms with theoretical guarantees that maximize the amount of desired information and actions and minimize the amount of biased and manipulative actions in interaction with ML systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper proposes a game-theoretic framework to model interactions between ML system owners and users with misaligned interests due to financial, social, and political factors. It claims to introduce scalable algorithms equipped with theoretical guarantees that enable users to unilaterally maximize access to desired information and actions while minimizing exposure to biased or manipulative outputs, without requiring any cooperation or protocol changes from the ML system owners.

Significance. If the unspecified game model, algorithms, and guarantees can be rigorously constructed and verified, the work would offer a user-centric alternative to owner-dependent bias mitigation techniques. This could be impactful for real-world ML services where owners lack incentives to self-regulate, potentially enabling practical protection mechanisms in high-stakes domains.

major comments (2)

[Abstract] Abstract: The central claim that 'scalable algorithms with theoretical guarantees' are presented lacks any supporting construction; no game definition (player action spaces, information structures, or payoff functions), algorithm description, or proof sketch appears, rendering it impossible to assess whether the claimed bounds are non-vacuous or hold against adversarial ML systems.
[Abstract] The weakest assumption—that real-world conflicts can be accurately captured by a game-theoretic model allowing unilateral user strategies with meaningful guarantees—is load-bearing but unsupported, as the manuscript supplies neither an explicit model nor any analysis showing existence of such strategies.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback. The comments highlight opportunities to better connect the abstract to the technical content of the manuscript. We address each point below and will incorporate revisions to improve clarity without altering the core contributions.

read point-by-point responses

Referee: [Abstract] Abstract: The central claim that 'scalable algorithms with theoretical guarantees' are presented lacks any supporting construction; no game definition (player action spaces, information structures, or payoff functions), algorithm description, or proof sketch appears, rendering it impossible to assess whether the claimed bounds are non-vacuous or hold against adversarial ML systems.

Authors: We agree that the abstract does not preview the technical details and that this makes the central claims difficult to evaluate from the abstract alone. The game is formally defined in Section 2 with explicit player action spaces, information structures, and payoff functions; the algorithms appear in Section 3; and the theoretical guarantees, including bounds that hold against adversarial behavior, are proven in Section 4. We will revise the abstract to include a concise outline of these elements so that the claims are immediately supported by reference to the constructions in the paper. revision: yes
Referee: [Abstract] The weakest assumption—that real-world conflicts can be accurately captured by a game-theoretic model allowing unilateral user strategies with meaningful guarantees—is load-bearing but unsupported, as the manuscript supplies neither an explicit model nor any analysis showing existence of such strategies.

Authors: The manuscript does supply an explicit model and existence analysis, but these appear only in the body (Section 2) rather than being signaled in the abstract. We will update the abstract to reference the model definition and the results establishing existence of unilateral strategies with performance guarantees. If the referee would like additional discussion or a short proof sketch moved into the abstract or introduction, we are happy to do so. revision: yes

Circularity Check

0 steps flagged

No circularity: framework presented as original modeling proposal

full rationale

The paper proposes a game-theoretic framework and scalable algorithms as a new modeling choice to address conflicts of interest. No equations, fitted parameters, self-citations, or ansatzes are shown reducing the central claims to inputs by construction. The abstract and description treat the game model and guarantees as independently introduced rather than derived from prior self-referential results or data fits. This is a standard non-circular proposal of a modeling approach.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The contribution rests primarily on the introduction of a game-theoretic model and associated algorithms; no numerical free parameters are mentioned, and the modeling choice itself is the main added element.

axioms (1)

domain assumption Interactions between ML systems and users can be modeled as a game with conflicting interests
This is the foundational modeling premise invoked to justify the framework.

invented entities (1)

Game-theoretic interaction model for ML-user conflicts no independent evidence
purpose: To capture and mitigate biased information flows
The model is introduced as the core new construct in the paper.

pith-pipeline@v0.9.0 · 5696 in / 1251 out tokens · 55716 ms · 2026-05-19T15:59:59.016753+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We propose a game-theoretic framework that models the interaction between ML systems and users with conflicts of interest... Bayesian equilibrium... user strategy σr : D → Δ(M)
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Theorem 4.1... influential if and only if there exists t such that t + b = (ΦL(t) + ΦH(t))/2

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages · 3 internal anchors

[1]

Information discrepancy in strategic learning

Yahav Bechavod, Chara Podimata, Steven Wu, and Juba Ziani. Information discrepancy in strategic learning. InInternational Conference on Machine Learning (ICML), pages 1691–1715, 2022

work page 2022
[2]

Barry Becker and Ronny Kohavi. Adult. UCI Machine Learning Repository, 1996. DOI: https://doi.org/10.24432/C5XW20

work page doi:10.24432/c5xw20 1996
[3]

Poisoning attacks against support vector machines

Battista Biggio, Blaine Nelson, and Pavel Laskov. Poisoning attacks against support vector machines. InProceedings of the 29th International Coference on International Conference on Machine Learning, pages 1467–1474, 2012

work page 2012
[4]

Persuade me if you can: A framework for evaluating persuasion effectiveness and susceptibility among large language models, 2025

Nimet Beyza Bozdag, Shuhaib Mehri, Gokhan Tur, and Dilek Hakkani-Tür. Persuade me if you can: A framework for evaluating persuasion effectiveness and susceptibility among large language models, 2025. URLhttps://arxiv.org/abs/2503.01829

work page internal anchor Pith review arXiv 2025
[5]

Bayesian strategic classification

Lee Cohen, Saeed Sharifi-Malvajerdi, Kevin Stang, Ali Vakilian, and Juba Ziani. Bayesian strategic classification. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, editors,Advances in Neural Information Processing Sys- tems, volume 37, pages 111649–111678. Curran Associates, Inc., 2024. doi: 10.52202/ 079017-3546. URL ...

work page 2024
[6]

Crawford and Joel Sobel

Vincent P. Crawford and Joel Sobel. Strategic information transmission.Econometrica, 50 (6):1431–1451, 1982. ISSN 00129682, 14680262. URL http://www.jstor.org/stable/ 1913390

work page 1982
[7]

Strategic classification from revealed preferences

Jinshuo Dong, Aaron Roth, Zachary Schutzman, Bo Waggoner, and Zhiwei Steven Wu. Strategic classification from revealed preferences. InProceedings of the 2018 ACM Conference on Economics and Computation, pages 55–70, 2018. 10

work page 2018
[8]

M. E. Dyer and A. M. Frieze. On the complexity of computing the volume of a polyhedron. SIAM Journal on Computing, 17(5):967–974, 1988. doi: 10.1137/0217060. URL https: //doi.org/10.1137/0217060

work page doi:10.1137/0217060 1988
[9]

Dickerson

Seliem El-Sayed, Canfer Akbulut, Amanda McCroskery, Geoff Keeling, Zachary Kenton, Zaria Jalan, Nahema Marchal, Arianna Manzini, Toby Shevlane, Shannon Vallor, Daniel Susser, Matija Franklin, Sophie Bridgers, Harry Law, Matthew Rahtz, Murray Shanahan, Michael Henry Tessler, Arthur Douillard, Tom Everitt, and Sasha Brown. A mechanism-based approach to miti...

work page internal anchor Pith review doi:10.48550/arxiv 2024
[10]

Is amazon’s search engine biased? it’s hard to tell.Atlantic, September 2019

Sidney Fussell. Is amazon’s search engine biased? it’s hard to tell.Atlantic, September 2019

work page 2019
[11]

Goodfellow, Jonathon Shlens, and Christian Szegedy

Ian J. Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adver- sarial examples. InInternational Conference on Learning Representations, 2015

work page 2015
[12]

Hardt, N

Moritz Hardt, Nimrod Megiddo, Christos Papadimitriou, and Mary Wootters. Strategic classification. InProceedings of the 2016 ACM Conference on Innovations in Theoreti- cal Computer Science, ITCS ’16, page 111–122, New York, NY , USA, 2016. Association for Computing Machinery. ISBN 9781450340571. doi: 10.1145/2840728.2840730. URL https://doi.org/10.1145/28...

work page doi:10.1145/2840728.2840730 2016
[13]

Manipulating machine learning: Poisoning attacks and countermeasures for regression learning

M Jagielski, A Oprea, B Biggio, C Liu, C Nita-Rotaru, B Li, et al. Manipulating machine learning: Poisoning attacks and countermeasures for regression learning. In2018 IEEE Symposium on Security and Privacy (SP), pages 19–35, 2018

work page 2018
[14]

Jones and Benjamin K

Cameron R. Jones and Benjamin K. Bergen. Lies, damned lies, and distributional language statistics: Persuasion and deception with large language models, 2024. URL https://arxiv. org/abs/2412.17128

work page arXiv 2024
[15]

American Economic Review 101, 6 (October 2011), 2590–2615

Emir Kamenica and Matthew Gentzkow. Bayesian persuasion.American Economic Review, 101(6):2590–2615, October 2011. doi: 10.1257/aer.101.6.2590. URL https://www.aeaweb. org/articles?id=10.1257/aer.101.6.2590

work page doi:10.1257/aer.101.6.2590 2011
[16]

Towards deep learning models resistant to adversarial attacks

Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. Towards deep learning models resistant to adversarial attacks. InInternational Conference on Learning Representations, 2018

work page 2018
[17]

The planar k-means problem is np-hard.Theoretical computer science, 442:13–21, 2012

Meena Mahajan, Prajakta Nimbhorkar, and Kasturi Varadarajan. The planar k-means problem is np-hard.Theoretical computer science, 442:13–21, 2012

work page 2012
[18]

Amazon changed search algorithm in ways that boost its own prod- ucts.Wall Street Journal, September 2019

Dana Mattioli. Amazon changed search algorithm in ways that boost its own prod- ucts.Wall Street Journal, September 2019. URL https://www.wsj.com/articles/ amazon-changed-search-algorithm-in-ways-that-boost-its-own-products-11568645345

work page 2019
[19]

Using machine teaching to identify optimal training-set attacks on machine learners

Shike Mei and Xiaojin Zhu. Using machine teaching to identify optimal training-set attacks on machine learners. InProceedings of the aaai conference on artificial intelligence, volume 29, 2015

work page 2015
[20]

The social cost of strategic classification

Smitha Milli, John Miller, Anca D Dragan, and Moritz Hardt. The social cost of strategic classification. InProceedings of the conference on fairness, accountability, and transparency, pages 230–239, 2019

work page 2019
[21]

How apple’s apps topped rivals in the app store it controls.The New York Times, September 2019

Jack Nicas and Keith Collins. How apple’s apps topped rivals in the app store it controls.The New York Times, September 2019. URL https://www.nytimes.com/interactive/2019/ 09/09/technology/apple-app-store-competition.html

work page 2019
[22]

School Admission, 1998

Dan Ofer. School Admission, 1998. https://www.kaggle.com/datasets/danofer/law-school- admissions-bar-passage

work page 1998
[23]

Prosper Load, 2021

Henry Okam. Prosper Load, 2021. https://www.kaggle.com/datasets/henryokam/prosper-loan- data. 11

work page 2021
[24]

O'Brien and Carrie Jun Cai and Meredith Ringel Morris and Percy Liang and Michael S

Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Interactive simulacra of human behavior. InProceed- ings of the 36th Annual ACM Symposium on User Interface Software and Technology, UIST ’23, New York, NY , USA, 2023. Association for Computing Machinery. ISBN 9798400701320. d...

work page doi:10.1145/3586183.3606763 2023
[25]

J. R. Quinlan. Credit Approval. UCI Machine Learning Repository, 1987. DOI: https://doi.org/10.24432/C5FS30

work page doi:10.24432/c5fs30 1987
[26]

Certified defenses for data poisoning attacks

Jacob Steinhardt, Pang Wei Koh, and Percy Liang. Certified defenses for data poisoning attacks. InAdvances in Neural Information Processing Systems, volume 30, 2017

work page 2017
[27]

How will advanced ai systems impact democracy?, 2024

Christopher Summerfield, Lisa Argyle, Michiel Bakker, Teddy Collins, Esin Durmus, Tyna Eloundou, Iason Gabriel, Deep Ganguli, Kobi Hackenburg, Gillian Hadfield, Luke Hewitt, Saffron Huang, Helene Landemore, Nahema Marchal, Aviv Ovadya, Ariel Procaccia, Mathias Risse, Bruce Schneier, Elizabeth Seger, Divya Siddarth, Henrik Skaug Sætra, MH Tessler, and Matt...

work page arXiv 2024
[28]

Intriguing properties of neural networks

Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfel- low, and Rob Fergus. Intriguing properties of neural networks.arXiv preprint arXiv:1312.6199, 2013

work page internal anchor Pith review Pith/arXiv arXiv 2013
[29]

Position: Challenges and future directions of data-centric AI alignment

Min-Hsuan Yeh, Jeffrey Wang, Xuefeng Du, Seongheon Park, Leitian Tao, Shawn Im, and Yixuan Li. Position: Challenges and future directions of data-centric AI alignment. InForty- second International Conference on Machine Learning Position Paper Track, 2025. URL https://openreview.net/forum?id=bXfF6Dqe9s

work page 2025
[30]

Fairness in ranking, part i: Score-based ranking.ACM Comput

Meike Zehlike, Ke Yang, and Julia Stoyanovich. Fairness in ranking, part i: Score-based ranking.ACM Comput. Surv., 55(6), dec 2022. ISSN 0360-0300. doi: 10.1145/3533379. URL https://doi.org/10.1145/3533379

work page doi:10.1145/3533379 2022
[31]

Fairness in ranking, part ii: Learning-to-rank and recommender systems.ACM Comput

Meike Zehlike, Ke Yang, and Julia Stoyanovich. Fairness in ranking, part ii: Learning-to-rank and recommender systems.ACM Comput. Surv., 55(6), dec 2022. ISSN 0360-0300. doi: 10.1145/3533380. URLhttps://doi.org/10.1145/3533380

work page doi:10.1145/3533380 2022
[32]

Farmers sue over deletion of climate data from government websites.The New York Times, February 2024

Karen Zraik. Farmers sue over deletion of climate data from government websites.The New York Times, February 2024. URLhttps://www.nytimes.com/2025/02/24/climate/ agriculture-farmer-website-data-lawsuit.html. 12 A Limitations, Future Work and Broader Impact Limitations and Future Work.We assume that both the user and the learner know the prior over learner...

work page 2024
[33]

In the noiseless model, a report is a region of parameters. For example, if the equilibrium pools both parameters, then a dataset with w∗ ℓ (D) = 0.8 sends M ⋆(D) ={0.8,0.9} , and the learner responds with E[w∗ ℓ (D)| w∗ ℓ (D)∈ {0.8,0.9}] = 0.8+0.9 2 = 0.85. In the noisy model, a report is a numeric. Let b= 0.03 and σ= 0.2 . For the endpoint pattern 0< M ...

work page 2090

[1] [1]

Information discrepancy in strategic learning

Yahav Bechavod, Chara Podimata, Steven Wu, and Juba Ziani. Information discrepancy in strategic learning. InInternational Conference on Machine Learning (ICML), pages 1691–1715, 2022

work page 2022

[2] [2]

Barry Becker and Ronny Kohavi. Adult. UCI Machine Learning Repository, 1996. DOI: https://doi.org/10.24432/C5XW20

work page doi:10.24432/c5xw20 1996

[3] [3]

Poisoning attacks against support vector machines

Battista Biggio, Blaine Nelson, and Pavel Laskov. Poisoning attacks against support vector machines. InProceedings of the 29th International Coference on International Conference on Machine Learning, pages 1467–1474, 2012

work page 2012

[4] [4]

Persuade me if you can: A framework for evaluating persuasion effectiveness and susceptibility among large language models, 2025

Nimet Beyza Bozdag, Shuhaib Mehri, Gokhan Tur, and Dilek Hakkani-Tür. Persuade me if you can: A framework for evaluating persuasion effectiveness and susceptibility among large language models, 2025. URLhttps://arxiv.org/abs/2503.01829

work page internal anchor Pith review arXiv 2025

[5] [5]

Bayesian strategic classification

Lee Cohen, Saeed Sharifi-Malvajerdi, Kevin Stang, Ali Vakilian, and Juba Ziani. Bayesian strategic classification. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, editors,Advances in Neural Information Processing Sys- tems, volume 37, pages 111649–111678. Curran Associates, Inc., 2024. doi: 10.52202/ 079017-3546. URL ...

work page 2024

[6] [6]

Crawford and Joel Sobel

Vincent P. Crawford and Joel Sobel. Strategic information transmission.Econometrica, 50 (6):1431–1451, 1982. ISSN 00129682, 14680262. URL http://www.jstor.org/stable/ 1913390

work page 1982

[7] [7]

Strategic classification from revealed preferences

Jinshuo Dong, Aaron Roth, Zachary Schutzman, Bo Waggoner, and Zhiwei Steven Wu. Strategic classification from revealed preferences. InProceedings of the 2018 ACM Conference on Economics and Computation, pages 55–70, 2018. 10

work page 2018

[8] [8]

M. E. Dyer and A. M. Frieze. On the complexity of computing the volume of a polyhedron. SIAM Journal on Computing, 17(5):967–974, 1988. doi: 10.1137/0217060. URL https: //doi.org/10.1137/0217060

work page doi:10.1137/0217060 1988

[9] [9]

Dickerson

Seliem El-Sayed, Canfer Akbulut, Amanda McCroskery, Geoff Keeling, Zachary Kenton, Zaria Jalan, Nahema Marchal, Arianna Manzini, Toby Shevlane, Shannon Vallor, Daniel Susser, Matija Franklin, Sophie Bridgers, Harry Law, Matthew Rahtz, Murray Shanahan, Michael Henry Tessler, Arthur Douillard, Tom Everitt, and Sasha Brown. A mechanism-based approach to miti...

work page internal anchor Pith review doi:10.48550/arxiv 2024

[10] [10]

Is amazon’s search engine biased? it’s hard to tell.Atlantic, September 2019

Sidney Fussell. Is amazon’s search engine biased? it’s hard to tell.Atlantic, September 2019

work page 2019

[11] [11]

Goodfellow, Jonathon Shlens, and Christian Szegedy

Ian J. Goodfellow, Jonathon Shlens, and Christian Szegedy. Explaining and harnessing adver- sarial examples. InInternational Conference on Learning Representations, 2015

work page 2015

[12] [12]

Hardt, N

Moritz Hardt, Nimrod Megiddo, Christos Papadimitriou, and Mary Wootters. Strategic classification. InProceedings of the 2016 ACM Conference on Innovations in Theoreti- cal Computer Science, ITCS ’16, page 111–122, New York, NY , USA, 2016. Association for Computing Machinery. ISBN 9781450340571. doi: 10.1145/2840728.2840730. URL https://doi.org/10.1145/28...

work page doi:10.1145/2840728.2840730 2016

[13] [13]

Manipulating machine learning: Poisoning attacks and countermeasures for regression learning

M Jagielski, A Oprea, B Biggio, C Liu, C Nita-Rotaru, B Li, et al. Manipulating machine learning: Poisoning attacks and countermeasures for regression learning. In2018 IEEE Symposium on Security and Privacy (SP), pages 19–35, 2018

work page 2018

[14] [14]

Jones and Benjamin K

Cameron R. Jones and Benjamin K. Bergen. Lies, damned lies, and distributional language statistics: Persuasion and deception with large language models, 2024. URL https://arxiv. org/abs/2412.17128

work page arXiv 2024

[15] [15]

American Economic Review 101, 6 (October 2011), 2590–2615

Emir Kamenica and Matthew Gentzkow. Bayesian persuasion.American Economic Review, 101(6):2590–2615, October 2011. doi: 10.1257/aer.101.6.2590. URL https://www.aeaweb. org/articles?id=10.1257/aer.101.6.2590

work page doi:10.1257/aer.101.6.2590 2011

[16] [16]

Towards deep learning models resistant to adversarial attacks

Aleksander Madry, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, and Adrian Vladu. Towards deep learning models resistant to adversarial attacks. InInternational Conference on Learning Representations, 2018

work page 2018

[17] [17]

The planar k-means problem is np-hard.Theoretical computer science, 442:13–21, 2012

Meena Mahajan, Prajakta Nimbhorkar, and Kasturi Varadarajan. The planar k-means problem is np-hard.Theoretical computer science, 442:13–21, 2012

work page 2012

[18] [18]

Amazon changed search algorithm in ways that boost its own prod- ucts.Wall Street Journal, September 2019

Dana Mattioli. Amazon changed search algorithm in ways that boost its own prod- ucts.Wall Street Journal, September 2019. URL https://www.wsj.com/articles/ amazon-changed-search-algorithm-in-ways-that-boost-its-own-products-11568645345

work page 2019

[19] [19]

Using machine teaching to identify optimal training-set attacks on machine learners

Shike Mei and Xiaojin Zhu. Using machine teaching to identify optimal training-set attacks on machine learners. InProceedings of the aaai conference on artificial intelligence, volume 29, 2015

work page 2015

[20] [20]

The social cost of strategic classification

Smitha Milli, John Miller, Anca D Dragan, and Moritz Hardt. The social cost of strategic classification. InProceedings of the conference on fairness, accountability, and transparency, pages 230–239, 2019

work page 2019

[21] [21]

How apple’s apps topped rivals in the app store it controls.The New York Times, September 2019

Jack Nicas and Keith Collins. How apple’s apps topped rivals in the app store it controls.The New York Times, September 2019. URL https://www.nytimes.com/interactive/2019/ 09/09/technology/apple-app-store-competition.html

work page 2019

[22] [22]

School Admission, 1998

Dan Ofer. School Admission, 1998. https://www.kaggle.com/datasets/danofer/law-school- admissions-bar-passage

work page 1998

[23] [23]

Prosper Load, 2021

Henry Okam. Prosper Load, 2021. https://www.kaggle.com/datasets/henryokam/prosper-loan- data. 11

work page 2021

[24] [24]

O'Brien and Carrie Jun Cai and Meredith Ringel Morris and Percy Liang and Michael S

Joon Sung Park, Joseph O’Brien, Carrie Jun Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. Generative agents: Interactive simulacra of human behavior. InProceed- ings of the 36th Annual ACM Symposium on User Interface Software and Technology, UIST ’23, New York, NY , USA, 2023. Association for Computing Machinery. ISBN 9798400701320. d...

work page doi:10.1145/3586183.3606763 2023

[25] [25]

J. R. Quinlan. Credit Approval. UCI Machine Learning Repository, 1987. DOI: https://doi.org/10.24432/C5FS30

work page doi:10.24432/c5fs30 1987

[26] [26]

Certified defenses for data poisoning attacks

Jacob Steinhardt, Pang Wei Koh, and Percy Liang. Certified defenses for data poisoning attacks. InAdvances in Neural Information Processing Systems, volume 30, 2017

work page 2017

[27] [27]

How will advanced ai systems impact democracy?, 2024

Christopher Summerfield, Lisa Argyle, Michiel Bakker, Teddy Collins, Esin Durmus, Tyna Eloundou, Iason Gabriel, Deep Ganguli, Kobi Hackenburg, Gillian Hadfield, Luke Hewitt, Saffron Huang, Helene Landemore, Nahema Marchal, Aviv Ovadya, Ariel Procaccia, Mathias Risse, Bruce Schneier, Elizabeth Seger, Divya Siddarth, Henrik Skaug Sætra, MH Tessler, and Matt...

work page arXiv 2024

[28] [28]

Intriguing properties of neural networks

Christian Szegedy, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfel- low, and Rob Fergus. Intriguing properties of neural networks.arXiv preprint arXiv:1312.6199, 2013

work page internal anchor Pith review Pith/arXiv arXiv 2013

[29] [29]

Position: Challenges and future directions of data-centric AI alignment

Min-Hsuan Yeh, Jeffrey Wang, Xuefeng Du, Seongheon Park, Leitian Tao, Shawn Im, and Yixuan Li. Position: Challenges and future directions of data-centric AI alignment. InForty- second International Conference on Machine Learning Position Paper Track, 2025. URL https://openreview.net/forum?id=bXfF6Dqe9s

work page 2025

[30] [30]

Fairness in ranking, part i: Score-based ranking.ACM Comput

Meike Zehlike, Ke Yang, and Julia Stoyanovich. Fairness in ranking, part i: Score-based ranking.ACM Comput. Surv., 55(6), dec 2022. ISSN 0360-0300. doi: 10.1145/3533379. URL https://doi.org/10.1145/3533379

work page doi:10.1145/3533379 2022

[31] [31]

Fairness in ranking, part ii: Learning-to-rank and recommender systems.ACM Comput

Meike Zehlike, Ke Yang, and Julia Stoyanovich. Fairness in ranking, part ii: Learning-to-rank and recommender systems.ACM Comput. Surv., 55(6), dec 2022. ISSN 0360-0300. doi: 10.1145/3533380. URLhttps://doi.org/10.1145/3533380

work page doi:10.1145/3533380 2022

[32] [32]

Farmers sue over deletion of climate data from government websites.The New York Times, February 2024

Karen Zraik. Farmers sue over deletion of climate data from government websites.The New York Times, February 2024. URLhttps://www.nytimes.com/2025/02/24/climate/ agriculture-farmer-website-data-lawsuit.html. 12 A Limitations, Future Work and Broader Impact Limitations and Future Work.We assume that both the user and the learner know the prior over learner...

work page 2024

[33] [33]

In the noiseless model, a report is a region of parameters. For example, if the equilibrium pools both parameters, then a dataset with w∗ ℓ (D) = 0.8 sends M ⋆(D) ={0.8,0.9} , and the learner responds with E[w∗ ℓ (D)| w∗ ℓ (D)∈ {0.8,0.9}] = 0.8+0.9 2 = 0.85. In the noisy model, a report is a numeric. Let b= 0.03 and σ= 0.2 . For the endpoint pattern 0< M ...

work page 2090