Calibrating user trust in AI outputs.

Designers should help users question AI outputs by teaching skepticism, supplying explanations, showing rationales, adding frictions, and clarifying the AI's role for balanced trust and critical thinking.

About this paper

The author presents six principles for designing generative AI applications and pairs each with specific design strategies.

These principles, developed through extensive research and validation, aim to provide actionable design recommendations for improving generative AI UX.

Here are some methods used in this study:

Literature Review Modified Heuristic Evaluation

Which part of the paper did the design guideline come from?

“The relatively lower relevance ratings for Design for Appropriate Trust & Reliance, Design for Human Control, and Design for Optimization stemmed from differences in application domain and output modality. In some cases, we accepted that relevance may vary by use case; in other cases, we addressed issues raised by participants to clarify or expand relevance. For example, the four evaluators who rated Design for Appropriate Trust & Reliance as "not relevant" had examined image or music (...)” (Section 8.2.1: Iteration 3: Modified Heuristic Evaluation)

Weisz, J. D., He, J., Muller, M., Hoefer, G., Miles, R., & Geyer, W. (2024). Design Principles for Generative AI Applications. Proceedings of the CHI Conference on Human Factors in Computing Systems.

Inspiration and scope

In this paper, authors focused on users of generative AI systems to determine reliability and quality of AI outputs.

You are designing for wireless engineers, ensuring they trust and utilize generative AI effectively. Both design and paper contexts need trustworthy AI. In academics, tools guide users on AI reliance, while in your context, reliable AI ensures accuracy and efficiency for engineers.

Also, both AI applications impact decision-making. Academically, it’s about trust in AI to enhance decision accuracy, while your context aims to optimize engineers' decisions.

By leveraging these similarities, consider designing interactive explainability features and confidence indicators in your AI. This fosters informed decisions and builds trust with wireless engineers, enhancing decision-making in wireless communication.

Your input

What: Generative AI application for use in Wireless Communication
Who: Wireless Engineers
Design stage: Research

Understanding users

The following user needs and pain points may apply to your design target as well:

Design for appropriate trust and reliance

Designers should ensure their wireless communication AI is clear about its capabilities and limitations to help engineers decide when to trust its outputs. Providing explanations for decisions will enhance trust and prevent overreliance, crucial for optimizing system performance.

Design for imperfection

The generative AI in wireless communication should caution engineers about possible inaccuracies and provide ways to handle them. Highlighting uncertainties will ensure engineers are aware of the potential pitfalls, promoting better decision-making.

Methods for you

Consider the following method(s) used in this paper for your design work:

Literature Review

Conducting a literature review helps identify existing research, guidelines, and frameworks relevant to generative AI and human-computer interaction, ensuring that design approaches are grounded in established knowledge. Designers should ensure the review covers a broad range of sources to capture diverse viewpoints and the latest advancements in the field.

Heuristic Evaluation

Heuristic evaluation involves having design practitioners evaluate existing generative AI applications against established design principles to identify clarity, relevance, and gaps. Designers should focus on collecting diverse feedback from practitioners with varying levels of experience to ensure comprehensive insights.

[Table 2] From this figure, you can gain insights into the iterative process of developing design principles and strategies, which may inform your investigation of the problem space.