Generative AI Meets Data Quality: Innovation or Risk?
演讲者
张龙天
时间
2026年01月09日 15:00 至 16:00
地点
A3-2-303
线上
Zoom 435 529 7909
(BIMSA)
摘要
The widespread adoption of Generative AI raises concerns about potential risks, particularly those arising from excessive reliance on AI. This paper examines both the benefits and drawbacks of this emerging technology through the lens of data quality. We develop a semi-endogenous growth model in which production depends on two types of data: AI-generated data and producer data, the latter representing real-world information. Although AI-generated data are substantially cheaper to produce, their use involves a trade-off in the form of lower data quality, which leads to higher error rates in production. Our analysis shows that firms, operating under competitive equilibrium, tend to underutilize both types of data relative to the optimal allocation. We further demonstrate that, while multiple Generative AI firms exist in the market, the optimal number is one. These findings support the case for government intervention in the AI industry.