With the development of artificial intelligence advancing rapidly in recent years, the prospects for applications in the social sciences seem more and more within reach. A recent paper on generative agents by Joon Sung Park and colleagues offers a tantalizing example of the possibilities ahead. The paper describes the development of 1000 computational agents that replicate human behavior across domains. By applying large language models to qualitative interviews about the lives of 1052 real people, the authors are able to create agents that simulate their attitudes and behaviors. These agents replicate responses on the General Social Survey 85% as accurately as the original participants replicate their own answers two week later. While additional advances are likely to make this approach valuable in numerous applications, I think that an early application might be the creation of a pool of test subjects for a range of pilot tests currently run with small samples of recruits from undergraduate classes across the social sciences.
Be the first to comment