External Help Under a Preference for Independent Work: AI versus Human Assistance.

2026-09-28

2026-12-31

The primary outcome is external-help uptake, defined as a binary indicator equal to 1 if the participant submits the external resource’s guess rather than their own initial guess, and 0 if the participant submits their own initial guess.

Because each participant encounters both sources of external assistance, this outcome is measured separately for:

1. AI-help uptake: whether the participant chooses the AI’s guess over their own initial guess.
2. Human-help uptake: whether the participant chooses the human’s guess over their own initial guess.

Our primary comparisons examine (i) whether external-help uptake differs between AI and human assistance and (ii) whether the expressed preference against using external help changes uptake relative to the Baseline condition.

Secondary outcomes include the following measures:
Assistive-tool perception, Descriptive norm belief, Injunctive norm belief, Perceived accuracy of external help

These outcomes will be used to examine potential mechanisms underlying differences in external-help uptake, including whether AI and human assistance differ in how tool-like, socially prevalent, socially appropriate, or capable they are perceived to be.

We conduct an online experiment in which participants complete a text-based judgment task using transcripts from the television game show To Tell the Truth. For each transcript, participants first identify which of the three contestants is the genuine individual described in an accompanying affidavit and report their confidence and perceived task difficulty.

The study varies the source of external assistance available to participants. Using a within-subjects design, each participant encounters two transcripts: in one, the available external help comes from an AI system; in the other, it comes from a human source. The assignment of the help source to the transcript and the order of transcript presentation are randomized. Participants are informed that the two sources of assistance are equally capable.

After completing their initial judgments for both transcripts, participants revisit each decision and choose whether to submit their own initial answer or the answer provided by the external resource. The primary outcome is participants’ uptake of external assistance.

Across experimental treatments, we additionally vary whether participants are explicitly told that independent work is preferred and that they should not use external help (Baseline vs Nudge treatments). This allows us to examine how the source of assistance and an expressed preference for independent work affect decisions to rely on external help.

Following each decision, participants answer questions about how they perceive the external resource and about their beliefs regarding the prevalence and social appropriateness of using such assistance. Participants also complete a brief demographic and prior-AI-use questionnaire at the end of the study.

Not available

computer software

transcript, order of assignment of help

No

200 participants for each treatment, contingent on funding resources

For the pilot, we will have 30 participants for the first two treatments. Later, we plan to enroll 100 participants per treatment if sufficient funding is available.

For the pilot, we will have 30 participants for the first two treatments. Later, we plan to enroll 100 participants per treatment if sufficient funding is available. Each participant submits a guess for two transcripts, so 200 observations pr treatment.