This article critiques the assumption that AI universally enhances access to information. It introduces the concept of 'knowledge collapse,' where uneven reliance on AI-generated content results in a narrowing of public knowledge, particularly with eccentric viewpoints. The authors present a model to illustrate how reduced costs for AI-generated content can lead to significant deviations from true knowledge distributions. Their findings suggest that while AI may reduce information costs, it paradoxically risks under-representing diverse perspectives, thus degrading the overall quality of public knowledge.
Excessive reliance on AI-generated content leads to a curtailing of eccentric and rare viewpoints, causing public knowledge to collapse toward the center, which undermines a comprehensive vision.
As the discount rate for AI-generated content decreases, the distribution of public knowledge diverges significantly from the true distribution, indicating a deeper risk of model collapse.
Collection
[
|
...
]