Data: Implications for Markets and for Society
Read PDF →Ziani, 2019
Category: Economics/ML
Overall Rating
Score Breakdown
- Latent Novelty Potential: 6/10
- Cross Disciplinary Applicability: 7/10
- Technical Timeliness: 6/10
- Obscurity Advantage: 3/5
Synthesized Summary
-
The most notable insight lies in Chapter 7, where the paper presents a counter-intuitive theoretical finding: an adversarial data provider might strategically choose to reveal less information... because partial revelation can lead to a worse outcome... than full revelation.
-
While the paper's models are stylized, this specific concept—strategic information omission as a potent adversarial tactic—offers a novel angle for modern AI robustness research...
-
...many aspects of the thesis address problems now superseded by advancements in ML and data handling...
-
...presents a distinct, albeit abstract, challenge for AI systems operating on incomplete data streams from untrusted sources.
Optimist's View
-
This thesis, while covering several important topics at the intersection of data, markets, and society, contains a particularly ripe and unconventional insight in Chapter 7 concerning the strategic behavior of third-party data providers in auctions.
-
The core, counter-intuitive finding is that an adversarial data provider... may strategically choose to reveal less information... because partial revelation can be more damaging... than full revelation.
-
This specific result... could fuel highly unconventional research in understanding and defending against modern adversarial information influence on AI systems.
-
This thesis points to a subtler attack vector: exploiting the AI's information processing mechanisms by controlling what is revealed and what isn't.
Skeptic's View
-
The core formulations... grapple with data in a way that feels increasingly detached from modern data paradigms.
-
Modern data applications are dominated by complex machine learning tasks... where the goal is prediction accuracy or model performance, not just unbiased estimation of simple statistics.
-
...rely on a simplification that contemporary research... immediately found problematic.
-
Modern research has significantly advanced beyond the specific problem formulations and solutions presented here, particularly in data markets, privacy, and fairness.
Final Takeaway / Relevance
Watch