The Data-Dollars Tradeoff: Privacy Harms vs. Economic Risk in Personalized AI Adoption

Paper Title

The Data-Dollars Tradeoff: Privacy Harms vs. Economic Risk in Personalized AI Adoption

Publication Info

• Topic area: Privacy risks and economic trade-offs in AI personalization adoption.
• Keywords: AI personalization, privacy risks, ambiguity, risk, data leaks, economic trade-offs, transparency, user behavior, privacy labels, algorithmic pricing.

Background and Problem

• Problem / challenge: Limited understanding of how different information environments (risk vs. ambiguity) influence user behavior towards AI personalization under privacy leak threats. Prior work has not empirically separated economic and non-monetary harms caused by privacy leaks.
• Significance: Privacy concerns are a major barrier to AI adoption, with potential economic and non-monetary harms from data leaks. Understanding these dynamics is critical for designing user-centric AI systems and regulatory frameworks.
• Motivation and related work: Existing literature highlights the privacy paradox, ambiguity aversion, and the role of transparency in shaping privacy behavior. However, gaps remain in understanding how ambiguity and data type influence AI adoption, and how users value privacy disclosure mechanisms.

Solution

• Proposed approach: A 2 × 3 between-subjects experiment to examine the effects of risk versus ambiguity and data type (preference vs. sensitive) on AI personalization adoption, bargaining behavior, and willingness to pay for privacy labels.
• Novelty:
  1. Separation of risk and ambiguity in privacy-leak environments.
  2. Isolation of economic and non-monetary harms from privacy leaks.
  3. Measurement of willingness to pay for perfectly informative privacy labels.
  4. Analysis of human-algorithm bargaining behavior under privacy risks.
• Procedure and key techniques:
  • Participants (N = 610) chose between standard and AI-personalized product baskets requiring data sharing.
  • Manipulated information environments: Risk (30% leak probability) vs. Ambiguity (10-50% leak probability).
  • Data types: Preference Data (preferences) vs. Sensitive Data (demographics) vs. Neutral (random surcharge).
  • Measured AI adoption, bargaining behavior in an ultimatum game, and willingness to pay for privacy labels using the Becker-DeGroot-Marschak (BDM) method.

Results

• Concrete findings:
  • Under Risk, AI adoption was stable (~50%) across all conditions.
  • Under Ambiguity, AI adoption dropped significantly in privacy conditions (Preference Data: 42%, Sensitive Data: 34%) compared to Neutral (56%).
  • Users overpaid for privacy labels (average bid: 17.7 vs. theoretical maximum: 15), indicating strong demand for transparency.
  • Ambiguity reduced bargaining assertiveness, but privacy threats alone did not alter bargaining behavior.
• Advantage over baselines:
  • Ambiguity, not risk, drives avoidance of AI personalization under privacy threats.
  • Verified privacy labels effectively mitigate privacy concerns and ambiguity.
• Experiments / evaluation:
  • 2 × 3 design with pre-registered hypotheses.
  • Metrics: AI adoption rates, willingness to pay for privacy labels, bargaining rejection rates.
  • Statistical tests: Chi-square tests, OLS regressions, and logit models.
• Limitations and future work:
  • Artificial data leak abstraction limits real-world generalizability.
  • Sample skewed towards English-speaking participants.
  • Statistical power may be insufficient for detecting some interactions.
  • Future work: Cross-cultural studies, real-world privacy label trustworthiness, disentangling baseline data-sharing reluctance.

Summary

This study investigates how risk and ambiguity influence user behavior towards AI personalization under privacy leak threats. Results show that ambiguity significantly reduces AI adoption, while risk does not. Users exhibit strong preferences for transparency, overpaying for privacy labels that eliminate ambiguity. Privacy threats do not directly affect bargaining behavior but reduce assertiveness under ambiguity. These findings highlight the importance of transparency mechanisms, such as third-party verified privacy labels, in fostering informed AI adoption. Future work should explore real-world applications, cross-cultural differences, and adaptive interface designs to address privacy concerns.