Strategic Interactions with the Assistant of an Algorithm: Information Acquisition Treatment

How to cite

XUE, Lian; Bai, Lu; Gui, Zhengqing; Wei, Lijia. 2022. "Strategic Interactions with the Assistant of an Algorithm: Information Acquisition Treatment." AEA RCT Registry. AEARCTR-0010264. https://doi.org/10.1257/rct.10264-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Lian XUE* , Lu Bai , Zhengqing Gui , Lijia Wei

* Primary investigator

Registry ID
AEARCTR-0010264
DOI
https://doi.org/10.1257/rct.10264-1.0
First registered
Last updated
Status
On going
Trial dates
2022-05-01 – 2022-12-01
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/10264

Abstract

We present a laboratory experiment that examines individuals’ willingness to take advice from algorithms provided by artificial intelligence (AI). We explored two distinguished channels behind the algorithm, the data and the mechanism. In this additional experiment, we will examine individual willingness to pay for the algorithm advice using a within-subject information acquisition treatment design. We will also compare welfare effects, in terms of subject payoff, with regard to information provision and acquisition. The benchmark for comparison where no information is available is borrowed from a previously conducted RCT which we also pre-registered on this website.

Intervention

This is a laboratory experiment, we will randomly invite the student subjects to one of the randomized experimental treatments.
Subjects will be instructed to play a repeated strategic interaction game and will have the opportunity to buy advice from an algorithm.
Subjects' payoff will be determined by the decision they and their counterpart made in the game and the amount they paid for the algorithm.
In the experiment, the maximum amount participants could pay will be lower than the minimum amount they could get from the interaction game.

Experimental design

Our experimental design combines between- and within-subject design.
In the control group, subjects will be instructed to play the one-shot centipede game for a repeated iteration of 10 rounds. Among these, 5 rounds will be played with free algorithm advice and 5 rounds will be played with costly algorithm advice. The order of these 5 rounds will be counterbalanced across sessions.
In the treatment group, subjects will play the same centipede game for 10 rounds with a costly-to-free/free-to-costly counterbalance. The only difference is that in the treatment group, subjects will be informed of the underlying data (including 5 aspects of main determinants) that the algorithm used to generate the predictions.

Design details

Randomization method
Randomization is conducted by a computer.
Randomization unit
Experimental sessions
Clusters
Each treatment will have 4 sessions
Observations
Each session will have 15-25 student subjects
Arms
We are planning to have 60-80 subjects per treatment. In total there will be 2 treatments, which will give us 120-160 subjects in total.
Primary outcomes
Willingness to pay (WTP) for the algorithm
Secondary outcomes
Subject payoff (welfare analysis)

Cited by

No citation data available for this registration yet.

Keywords