AI Empowerment or AI Replacement
How to cite
XUE, Lian; Wei, Lijia; Bai, Lu. 2023. "AI Empowerment or AI Replacement." AEA RCT Registry. AEARCTR-0012196. https://doi.org/10.1257/rct.12196-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
In our earlier experiment, centered around social and AI scoring, we painted an optimistic picture of AI's introduction. Not only did we find that AI scoring systems enhance human cooperation, but individuals also demonstrated a growing acceptance of these systems, particularly as they gained more experience or encountered deteriorating cooperative environments.
However, concerns arise regarding the potential for excessive automation resulting from AI implementation. The replacement of humans by AI may lead to undesirable social consequences such as job loss, stagnant wages, and increased inequality. To address these concerns, we introduced a new treatment aimed at discussing the potential of AI restatement – a type of AI development that augments workers' productivity and creates new tasks for labor-intensive roles.
Our study delved into the dual functions of AI – replacement and reinstatement – and employed experimental methods to investigate individual preferences and their associated outcomes concerning these two AI applications.
Intervention
Our specific objective is to introduce a novel intervention, granting participants the autonomy to determine the dynamics of their relationship with the decision-making process, which involves the integration of their own choices with AI assistance.
In conjunction with the existing AI-score-punish and Social-Score-punish treatments, which incorporate punitive measures, we provide participants with the opportunity to decide whether they wish to receive guidance from the AI while retaining ultimate decision-making authority. We refer to this distinct treatment as the AI-assistant (AI-assist) treatment.
Experimental design
Our experimental design involves a public goods game with participants grouped into foursomes, playing 20 rounds. Like in the "Digital Punishment" setup, two participants get 20 tokens, while the other two get 40 tokens initially. We set a marginal per capita return (MPCR) at 1.6. We depart from standard practices by introducing rematching every 10 rounds, closely related to our treatment groups. In this new treatment, instead of voting for whether or not to incorporate the AI or Social score system in their group task, participants choose between two ways AI could influence their group decision-making - one in which AI scores everyone and implements corresponding punitive measures, the other in which AI makes scoring suggestions and participants themselves make final judgment and decisions.
Design details
Cited by
No citation data available for this registration yet.