AI as Doctors

How to cite

Chan, Alex; Cutler, David. 2026. "AI as Doctors." AEA RCT Registry. AEARCTR-0017897. https://doi.org/10.1257/rct.17897-1.2

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

Alex Chan* , David Cutler

* Primary investigator

Registry ID
AEARCTR-0017897
DOI
https://doi.org/10.1257/rct.17897-1.2
First registered
Last updated
Status
In development
Trial dates
2026-02-14 – 2027-01-01
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17897

Abstract

Recent advances in artificial intelligence (AI) offer opportunities to improve healthcare delivery through AI-driven primary care consultations. However, patient acceptance remains uncertain. To inform healthcare policy and insurance benefit design, it is essential to understand the conditions under which patients would accept AI consultations instead of traditional doctor visits, especially the financial compensation or savings necessary to encourage adoption.

This study aims to:
Measure patients' willingness-to-accept (WTA) compensation for switching from traditional doctor visits to AI-based consultations for primary care.
Identify how WTA varies based on AI attributes, including:
Price differences
Wait time differences
Medical liability of AI for errors
AI’s ability to prescribe medications
AI’s clinical diagnostic performance compared to doctors (above-average or top 10% of U.S. medical licensing exams)
Familiarity with the doctor (regular vs. new)

and to do so across different types of healthcare situations.

Intervention

The study is conducted as an online survey experiment via Prolific, which presents participants with hypothetical scenarios about AI systems and asks them to indicate their preferences. We vary attributes of the AI and doctor alternatives, and elicit preferences using a multiple price list design.

Experimental design

See attached Qualtrics file for the full design and randomization.

Participants are randomized to arms where we hold some attributes of AI/doctor fixed (e.g. whether it is a new doctor or the usual doctor), then they will make binary choices using an adaptive multiple price list (MPL) to gauge switching points in price.

*This MPL design is different from usual MPLs in that it will show a random price level as a starting point and present the choices as binary choices instead of the full list (to address the criticism that MPL participants might be biased to just pick a switching price towards the middle of the list), it will then show prices based the choice to hone into the switching price.

Design details

Randomization method
Qualtrics
Randomization unit
individual
Clusters
2400 individuals
Observations
2400 individuals
Arms
For the experiment, we expect to enroll 2400 respondents. 50 in each “arm”. The 48 arms are simply 3 possible hypothetical Ai memory*4 vignettes*2 types of doctors (new vs usual)* 2 AI performance levels as described above and shown in the instrument. Because we will need to compare across arms, the rough power calculation and minimum detectible effect sizes suggest that this is the smallest sample we need to have adequate power.
Primary outcomes
Willingness to pay between binary choices (multiple price list), used to infer the outcomes of interest, namely valuation of various AI and doctor attributes (price, waiting time, ability to prescribe, etc.)

Cited by

No citation data available for this registration yet.

Keywords