Filter Bubbles in the Age of Generative AI: A Literature Review and In-depth Analysis

This literature review explores the evolution of filter bubbles in the age of Generative AI and analyzes the impact of personalized algorithms on information acquisition and opinion formation. GenAI shifts from filtering to generating content, with a deeper level of personalization, bringing new characteristics such as opinion homogenization; Its impacts cover democratic discussions, education, business decision-making, and mental health; Responses require multi-level efforts in technology (diversity injection, transparency), policy (platform responsibility), and education (digital literacy). Future research should focus on long-term impacts and interdisciplinary studies.

The filter bubble was proposed by Eli Pariser in 2011, describing how personalized algorithms isolate users in an environment of consistent opinions. Its theoretical foundations span disciplines such as communication and psychology. It is necessary to distinguish related concepts: information cocoons (proposed by Sunstein, where individuals' active choices lead to self-isolation), echo chamber effect (opinions are reinforced in closed groups), and confirmation bias (tendency to accept information that supports existing views). Each concept has different focuses.

This paper uses the literature review method to systematically sort out studies related to filter bubbles in the age of Generative AI, integrate multi-disciplinary perspectives from communication, psychology, computer science, and sociology, analyze their formation mechanisms, impact scope, and response strategies, aiming to comprehensively present the current research status and trends in this field.

Multiple studies show: 1. The AI-generated content received by different users on the same topic varies significantly and is highly correlated with historical preferences; 2. The degree of opinion polarization increases among users who frequently use personalized AI services, especially in sensitive areas; 3. Cross-cultural differences exist: users in collectivist cultures have a strong tendency to conform, while users in individualist cultures are more likely to form personalized information environments.

The risks of filter bubbles include: 1. Threatening democratic public discussions and exacerbating social division; 2. Affecting the comprehensive dissemination of knowledge in education and hindering the formation of critical thinking; 3. Causing business decision-makers to ignore market signals and suppress innovation; 4. Long-term exposure to bubbles may bring psychological comfort, but it is easy to produce cognitive shock when exposed to different views.

Responses require multi-level efforts: 1. Technology: Inject diversity mechanisms (adjust algorithm weights), improve transparency (let users understand the generation logic), and enhance user control (allow adjustment of personalization levels); 2. Policy: Clarify platform responsibilities (regular audits) and formulate industry standards; 3. Education: Strengthen digital literacy (understand algorithms) and cultivate media literacy (identify information reliability).

Filter bubbles in the GenAI era are more hidden and in-depth, with negative impacts covering multiple fields. Responses require collaboration between technology, policy, and education; future research should focus on long-term impacts, interdisciplinary integration, the impact of emerging technologies, and the effectiveness of intervention measures. It is necessary to find a balance between personalized services and information diversity to build a healthy and inclusive information environment.