Training Policymakers in AI

How to cite

Mehmood, Sultan; Naseer, Shaheen; Chen, Daniel. 2021. "Training Policymakers in AI." AEA RCT Registry. AEARCTR-0008431. https://doi.org/10.1257/rct.8431-1.0

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

Sultan Mehmood* , Shaheen Naseer , Daniel Chen

* Primary investigator

Registry ID
AEARCTR-0008431
DOI
https://doi.org/10.1257/rct.8431-1.0
First registered
Last updated
Status
In development
Trial dates
2021-11-09 – 2021-11-10
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/8431

Abstract

In this experiment with senior deputy ministers of Pakistan, we randomize training policymakers in an “AI for Policy” workshop. The first treatment arm randomly receives a training focused on utility or benefits of AI for policymaking. The second treatment arm receives training focused on potential harms or dangers of AI. A control group received no training. Our design will elicit demand for both trainings and study the mediating channel experimentally using a book lottery in a simplified Becker-Degroot-Marschak mechanism. Using this lottery that randomly implements an earlier action to causally isolate the effect of book choice allows us to experimentally disentangle the effect of training from randomly assigned book conditional on the training’s demand. The rich administrative dataset and close collaboration with the Government of Pakistan will give us access to suit of measurements including actual policy outcomes.

Intervention

We randomize training policymakers in an “AI for Policy” workshop.

Treatment 1: The first treatment arm randomly receives a training focused on utility or benefits of AI for policymaking.

Treatment 2: The second treatment arm receives training focused on potential harms or dangers of AI.

A control group receives no training.

The design also studies a mediating channel experimentally using a book lottery in a simplified Becker-Degroot-Marschak mechanism. Using this lottery that randomly implements an earlier action to causally isolate the effect of book choice allows us to experimentally disentangle the effect of training from randomly assigned book conditional on the training’s demand.

Experimental design

Treatment 1: The first treatment arm randomly receives a training focused on utility or benefits of AI for policymaking.

Treatment 2: The second treatment arm receives training focused on potential harms or dangers of AI.

300 deputy ministers are randomly assigned in treatment 1, 2 or control group.

Design details

Randomization method
randomization done in office by a computer at the individual level using Stata.
Randomization unit
individual.
Clusters
300 at the individual level
Observations
300
Arms
300 with 100 clusters per treatment arm.
Primary outcomes
self-reported benefits of AI for policy, actual policy decisions

Cited by

No citation data available for this registration yet.

Keywords