Question 1

How does this paper distinguish between personalization that helps users versus personalization that exploits them?

Accepted Answer

Acemoglu et al. argue that the critical variable is durability of deception. When false product attributes are quickly corrected by market feedback, AI-driven personalization genuinely improves user welfare. But when platforms learn to exploit durable misconceptions—what the authors call 'glossiness'—the same personalization engine becomes a tool against users, even as it appears to serve them. For product teams, this means your recommendation architecture's impact depends entirely on whether the information environment allows users to discover they've been deceived.

Question 2

What does 'glossiness' mean, and why is that vocabulary useful for product decisions?

Accepted Answer

Glossiness refers to product attributes that make something appear better than it actually is—a formal concept for what practitioners call dark patterns or attention manipulation. Rather than vague terms, this gives product leaders a precise way to analyze whether their interface choices, recommendation logic, or feature prominence systematically exploit user misperception. The paper's model shows platforms dynamically learn which glossy attributes work on which user segments, making this especially relevant for personalization systems.

Question 3

How should I think about the tension between optimizing for engagement and optimizing for user welfare?

Accepted Answer

The paper reveals this isn't a simple trade-off—it depends on market structure. When users can quickly learn the truth about products (short-lived deception), optimizing for engagement and user welfare align. But in markets with sticky misconceptions, the infrastructure that powers engagement optimization (dynamic learning, personalization, behavioral targeting) becomes optimized for exploitation instead. The uncomfortable implication: your personalization system's ethical status isn't determined by intent but by whether users can ever correct their mistakes.