ABSTRACT
An empirical approach to comparative law is challenging: neither are jurisdictions independent, nor can observational data isolate the causal effect of differences in doctrine. One viable way out are experiments. Yet testing the same stimulus material in different jurisdictions is logistically complicated, and hard to truly standardize. Sampling a large language model is a new substitute. Using a vignette about unforeseen changes in circumstances that make contract fulfilment difficult, this paper demonstrates that merely giving the LLM the otherwise identical vignette in different languages leads to strongly different results. When asked in German or English, GPT-3.5-turbo-1106 is much more inclined to grant a price adjustment than when queried in French or Italian.
Engel, Christoph, Experimental Comparative Law 2.0? Large Language Models as a Novel Empirical Tool (July 19, 2024), MPI Collective Goods Discussion Paper No 2024/12.
Leave a Reply