Optimising Signals in Human-AI Interaction: Experiment II

Last registered on May 14, 2025

View Trial History

Pre-Trial

Trial Information

General Information

Title

Optimising Signals in Human-AI Interaction: Experiment II

RCT ID

AEARCTR-0015758

Initial registration date

April 07, 2025

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published

April 22, 2025, 9:25 AM EDT

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated

May 14, 2025, 11:06 AM EDT

Last updated is the most recent time when changes to the trial's registration were published.

Locations

Country

United States of America

Region

Primary Investigator

Name

Ruru (Juan Ru) Hoong

Affiliation

Harvard University

Contact Primary Investigator

Other Primary Investigator(s)

PI Name

Bnaya Dreyfuss

PI Affiliation

Contact Investigator

Additional Trial Information

Status

On going

Start date

2025-04-14

End date

2025-06-07

Keywords

Behavior, Finance & Microfinance, Firms & Productivity, Labor

Additional Keywords

JEL code(s)

Secondary IDs

Prior work

This trial is based on or builds upon one or more prior RCTs.

Abstract

Artificial intelligence (AI) has caught pace with — and in some contexts even surpassed — humans in the ability to make predictions from data, purporting to improve decision-making. However, in cases where humans are still responsible for the final decision, biases in probabilistic reasoning can render even informative AI predictions detrimental to decision-making outcomes.

Using a randomised experiment with loan underwriters, we showed previously (#13716 in registry) that the provision of optimised AI signals can improve overall decision-making in spite of information loss. In this experiment, we explore the optimal AI policies further by adapting interventions to individual loan underwriters based on their historical decisions.

External Link(s)

Registration Citation

Citation

Hoong, Ruru (Juan Ru) and Bnaya Dreyfuss. 2025. "Optimising Signals in Human-AI Interaction: Experiment II." AEA RCT Registry. May 14. https://doi.org/10.1257/rct.15758-2.0

Sponsors & Partners

Experimental Details

Interventions

Intervention(s)

Participants see different types of AI signals to aid their decision-making, personalised based on their historical decisions.

Intervention (Hidden)

Two-stage experiment with loan officers.

In stage 1, randomisation is within-individual into 5 treatments. We use this data to inform personalised thresholds for the second stage.

In the second stage, we randomise loan officers into three groups: (i) personalised AI thresholds (calculated from Stage I experimental data.) vs. (ii) generalised universal binary threshold implied by theoretical model from experiment 1, (iii) top-performing policy on average in the population from Stage I. This is our core randomisation.

The randomisation is across-individual, 50% in control (general binary threshold), 50% in treatment (personalised thresholds). See PAP for more details.

Intervention Start Date

2025-05-15

Intervention End Date

2025-05-31

Primary Outcomes

Primary Outcomes (end points)

The main dependent variable we will look at is decision accuracy in State 2 (i.e. whether the decision to approve/deny the loan was ex-post correct).

Primary Outcomes (explanation)

Secondary Outcomes

Secondary Outcomes (end points)

Secondary Outcomes (explanation)

Experimental Design

We recruit 200-300 loan underwriters/officers/processors to make loan application decisions in two stages.

Our randomisation happens int the second stage, where we personalise interventions to each underwriter for half of the underwriters, and compare it to our control benchmarks.

Experimental Design Details

Two-stage experiment with loan officers, randomising into three groups: (i) personalised AI thresholds (calculated from Stage I experimental data.) vs. (ii) generalised universal binary threshold implied by theoretical model from experiment 1, (iii) top-performing policy on average in the population from Stage I.

The randomisation is across-individual, 25% in each of the two controls (general binary threshold; top or second-top performing policy in Stage 1), 50% in treatment (personalised thresholds). See PAP for more details.

Randomization Method

Randomisation done through Otree programme

Randomization Unit

Individual

Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters

200-300 loan officers/processors/underwriters.

Sample size: planned number of observations

We will aim to recruit 200-300 loan officers/processors/underwriters, unless otherwise limited by our recruitment. We will filter out subjects in our analysis based on attention and time checks, as well as Stage I performance as specified by our PAP.

Sample size (or number of clusters) by treatment arms

100-150 in treatment, 50-75 in each control.

Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

See PAP.

Supporting Documents and Materials

IRB