Short answer

Focus on building user confidence through consistent accuracy and intuitive onboarding, rather than relying on superficial cues like advisor identity or basic justifications, to encourage effective use of AI assistance.

Field
User-Centred Design
Source
Academic Publication (2023)
Method
Laboratory experiment
Sample
118 participants
Evidence
Strong effect

Users are more likely to accept advice from AI, like ChatGPT, when they are unfamiliar with the topic, have prior experience with the AI, or have received accurate advice from it previously, rather than based on the AI's perceived identity or the presence of justifications. This user-centred design research insight is drawn from a 2023 study published in Academic Publication. Using Laboratory experiment with 118 participants, researchers explored how this design variable affects real-world outcomes. The key design takeaway: Focus on building user confidence through consistent accuracy and intuitive onboarding, rather than relying on superficial cues like advisor identity or basic justifications, to encourage effective use of AI assistance.

Study
User-Centred DesignRecentStrong effect

User Trust in AI Advice is Driven by Familiarity and Past Experience, Not Advisor Identity

Users are more likely to accept advice from AI, like ChatGPT, when they are unfamiliar with the topic, have prior experience with the AI, or have received accurate advice from it previously, rather than based on the AI's perceived identity or the presence of justifications.

Academic Publication · 2023

01

Key Findings

  • 01Advisor identity (AI vs. human expert) did not significantly affect advice weighting.
  • 02Written justifications did not significantly affect advice weighting.
  • 03Participants weighed advice more heavily when they were unfamiliar with the topic, had past experience with ChatGPT, or had received more accurate advice previously.
  • 04Participants underestimated ChatGPT's accuracy on several topics.
  • 05Participants could only leverage correct advice more effectively when written justifications were provided.
02

Application

Design takeaway

Focus on building user confidence through consistent accuracy and intuitive onboarding, rather than relying on superficial cues like advisor identity or basic justifications, to encourage effective use of AI assistance.

How to apply

When designing AI-powered tools, prioritize features that build user familiarity and demonstrate consistent accuracy. Provide clear feedback on AI performance and consider progressive disclosure of justifications based on user needs.

Project actions

  • 01When evaluating AI tools, consider how user familiarity and past performance might influence adoption.
  • 02Think about how to design feedback loops that help users understand the AI's reliability.
03

Method & Evidence

AimTo investigate the factors influencing user reliance on AI-generated advice, specifically examining the impact of advisor identity, justification, task difficulty, user familiarity with the AI, and past advice accuracy.
MethodLaboratory experiment
ProcedureParticipants answered multiple-choice questions and were given advice from a GPT model. Their decision to update their initial responses based on the AI's advice was recorded, with various factors manipulated or measured.
Sample118 participants
ContextEducational/knowledge-based tasks

Variables

IV["Advisor identity (AI vs. human)","Presence of written justification","Topic familiarity","Past ChatGPT usage","Previous advice accuracy"]
DV["Weight placed on AI advice (i.e., likelihood of updating initial response)","Perceived accuracy of AI"]
CV["Number of questions","Academic subjects","Multiple-choice question format"]
04

Strengths & Limitations

Strengths

  • +Uses a large number of questions (2,828) across diverse subjects.
  • +Examines a widely used AI model (ChatGPT).

Limitations

The findings are specific to the AI model tested and the academic context. Generalizing to creative design tasks or different AI architectures requires further investigation.

Reliability & validity

The study's reliability could be supported by the large number of questions and participants. Validity is addressed by testing specific hypotheses about user behavior in a controlled lab setting, though ecological validity might be a concern given the artificial nature of the task.

Think critically

If justifications don't significantly increase advice weighting unless they help users verify correctness, what does this imply about the design of AI explanations for complex or subjective design decisions?

05

Design Principles

"User trust in AI is an emergent property of user experience and perceived reliability, not simply a function of its presentation or stated identity."

Understanding the psychological drivers of AI advice adoption is crucial for designing effective human-AI collaboration tools. Designers can leverage these insights to build interfaces and user experiences that foster appropriate levels of trust and reliance, leading to more efficient and accurate outcomes in various design and problem-solving contexts.

06

What This Means for Your Design

People trust AI advice more when they don't know the answer themselves, have used the AI before, or know it's usually right. They don't care as much if it's called an 'AI' or a 'human expert', or if it explains its answer, unless they can actually use the explanation to check if the advice is good.

How to use in your project

  • 1.Reference this study when discussing user trust in AI tools, the impact of AI on decision-making, or the calibration of user expectations regarding AI performance in your design project.
07

Add to My Project

08

Quick Cite

Paragraph starter

Research by Zhang (2023) indicates that user trust in AI advice, such as from ChatGPT, is significantly influenced by factors like user unfamiliarity with the task, prior experience with the AI, and the AI's demonstrated accuracy, rather than the AI's perceived identity or the presence of justifications. This suggests that for AI-assisted design tools, building consistent reliability and fostering user familiarity are key to effective adoption.

09

Source

Academic Publication

Taking Advice from ChatGPT

journal · 2023

View source

Questions About This Research

What does the research say about user trust in ai advice is driven by familiarity and past experience, not advisor identity?
Focus on building user confidence through consistent accuracy and intuitive onboarding, rather than relying on superficial cues like advisor identity or basic justifications, to encourage effective use of AI assistance. Evidence: Academic Publication (2023).
Why does "User Trust in AI Advice is Driven by Familiarity and Past Experience, Not Advisor Identity" matter for design?
Understanding the psychological drivers of AI advice adoption is crucial for designing effective human-AI collaboration tools. Designers can leverage these insights to build interfaces and user experiences that foster appropriate levels of trust and reliance, leading to more efficient and accurate outcomes in various design and problem-solving contexts.
How can designers apply this research?
Focus on building user confidence through consistent accuracy and intuitive onboarding, rather than relying on superficial cues like advisor identity or basic justifications, to encourage effective use of AI assistance.
What were the main findings?
Advisor identity (AI vs. human expert) did not significantly affect advice weighting.. Written justifications did not significantly affect advice weighting.. Participants weighed advice more heavily when they were unfamiliar with the topic, had past experience with ChatGPT, or had received more accurate advice previously.. Participants underestimated ChatGPT's accuracy on several topics.
What research method was used?
Laboratory experiment with 118 participants.
How strong is the evidence?
Evidence strength is rated Strong effect, based on a 2023 journal from Academic Publication.
What should I do differently in my next project?
When designing AI-powered tools, prioritize features that build user familiarity and demonstrate consistent accuracy. Provide clear feedback on AI performance and consider progressive disclosure of justifications based on user needs.
What are the limitations?
The study involved student participants answering academic questions, which may not generalize to all user groups or task types. The specific AI model used (ChatGPT) and its performance characteristics are also context-specific.