@misc{37118,
title = {Documenting Differences Between Humans and AI in High-Stakes Decisions: A Labor Market Turing Test},
author = {Abril Arteaga, Andres Sebastian and Rangel, Marcos and Zanoni, Wladimir},
year = {2025},
doi = {10.18235/0013729},
abstract = {We developed a Labor Market Turing Test (LMTT) to measure human-AI decision alignment using data from 277 human recruiters engaged in a field experiment set in Quito, Ecuador. We augmented the pool of recruiters by creating AI teams, each of which with differing impersonation of human-like traits, and compared their choices to humans and a benchmark AI model. While AI teams were more consistent, they selected candidates with a pattern that markedly different from human choices. In fact, random decisions mir- rored human choices more closely than our most human-like AI agents. These findings reveal a fundamental tension between algorithmic consistency and human judgment. That humans were closer to a random process when com- paring candidates with equal productivity might be seen as a fairer outcome. Our LMTT framework, which involves isolating and estimating a machina la- tent trait, provides a quantitative tool for assessing human-AI alignment which can be employed across critical domains, such as healthcare, justice, and edu- cation, thereby informing the design and AI governance.},
url = {https://doi.org/10.18235/0013729}
}