Study 008 / adaptation
Consider the advice and make your own choice.
Consider three recommendations before the explanation.
This voluntary demonstration changes how information is presented. Some details, including any simulated claims, are explained after the final step. You can reveal them or leave at any time.
Your responses stay in this browser tab. No research data is submitted.
Read the mechanism firstStudy 008 / Human versus machine authority
01 / Observation
The stated source of advice can change how the advice is received.
02 / Hypothesis
Identical advice may receive different trust ratings when attributed to a customer, expert, algorithm, AI assistant, or similar users.
03 / Mechanism
Perceived expertise, similarity, objectivity, or prior experience may influence reliance. The source label alone cannot establish actual competence.
04 / Experiment
Consider the same fictional recommendation in three presentations. One randomly selected human attribution and one randomly selected machine attribution appear first, in shuffled order. The unattributed recommendation appears last. Neither source category is assumed to be more persuasive. The notebooks have equivalent everyday-use titles with different phrasing and identical specifications. Their titles and the randomly assigned recommendation stay fixed across all rounds and replay. Choose an option and rate trust each time. The reveal follows all three rounds; replay removes all source labels.
05 / Result
Your choices and trust ratings are compared alongside the exact invented attributions. No real expert, AI service, or crowd generated the recommendation. Differences do not establish which label caused a response.
06 / Alternative explanations
Preference for one notebook title’s wording, prior attitudes toward AI, a preferred letter, suspicion, order, repetition, and remembering earlier responses may affect the comparison.
07 / Real-world use
AI assistants, expert badges, product recommendations, decision support, and “people like you” suggestions.
08 / Ethical dimension
A useful source label clarifies provenance and limits. Unsupported authority can create confidence without evidence.
09 / Countermeasure
Read the reasoning without the label. Ask what evidence supports the recommendation and how accuracy was assessed.
Short summaries in our own words: what the research found, why it belongs here, and where its conclusions stop. These papers provide background; they do not report results from your visit.
Primary research
Logg and colleagues found greater reliance on advice attributed to algorithms in several estimation tasks. The attribution comparison here asks a related question about the source label around a recommendation. It does not test the advice’s accuracy, and findings about numerical estimates do not automatically extend to every AI-mediated decision.
Forecasting experiments
After seeing an algorithm make errors, participants could become less willing to use it even when it outperformed a human forecaster. This counterpoint matters: reliance depends on experience as well as attribution. Our short comparison holds the recommendation constant and cannot reproduce the history of trust or disappointment.
A demonstration can make a mechanism inspectable. One person’s response, or a replay after disclosure, cannot identify a causal effect or establish nonconscious influence.