Send message to Authors

Long abstract:

When discussing generative AI systems such as ChatGPT, it is common to discuss them as if they are a single universal system. Although this may be true now, it is unlikely to remain the case. Generative AI systems are products of the data they are trained on, the legality of which has begun to vary. This paper explores how developing data ecologies are affecting regional generative AI implementation. Data ecologies refer to distinctive regulatory environments governing data flows. Data ecologies come into friction with one another when a product is designed in one data ecology but utilized in another. For example, ChatGPT is currently banned in Italy, where it has been accused of violating Italian/EU data privacy laws.

This paper builds from current literature examining the effects of different data ecologies on the targeted/surveillance advertising space. We explore how the effect emerges differently —and with potentially greater economic ramifications— in the case of generative AI due to the productive nature of generative AI. Our analysis is positioned around the regulatory framing of training data, including personal data, non-personal data, and copyrighted material. To do so, we provide a comparative analysis of four cases. 1) The European Union 2) The United States 3) Japan 4) China. Our comparative analysis is positioned to flesh out how each of these regulatory regimes are developing regulations to govern the flows of data that may be used in generative AI systems and highlight how their choices create geopolitical and economic friction.