Human-AI Interaction is a field that examines how humans use, interact with, and are affected by AI. From these insights, we can formulate practical design guidelines for developing new AI-based products and features, guide and inform new regulations and laws for AI, and support the ethical use of AI in society.
Human-AI interaction is not a new field, but with the rise of consumer-facing generative AI models and large language models, such as the launch of ChatGPT in 2022, it’s become a much more relevant topic in society. Research and development in the field comes from both academic institutions, e.g., Stanford’s Human-Centered AI Institute, and from industry, from big tech companies such as OpenAI and Anthropic.
Since development is moving incredibly fast, it’s hard for research, education, and larger institutions to stay up to date. Thus, it falls to the individual contributor, such as designers, developers, teachers, and other subject-matter experts, to continuously develop their competencies and skills.
Central Topics in Human-AI Interaction Research
Human-AI Interaction is a broad field encompassing topics such as human-AI collaboration, human-AI co-creation in creative processes, and responsible AI.
A central topic is keeping humans in the loop, meaning that we aim for AI to enhance our human abilities rather than replace them, also known as augmentation over automation. This stance is the exact opposite of the worries about using AI to replace human workers. Instead, we want to use AI to enhance our abilities, make us more productive, and automate certain boring, repetitive tasks. But the goal, in the end, is still to have a human evaluating the outputs, making the final decisions, and keeping an overview of the AI systems.
Another central topic is human-centered AI, meaning that we design and implement AI from a user-centered perspective, aligning well with UX practices. Human-centered AI also argues that AI should be developed and implemented for human well-being and should not harm us. Critical aspects of AI include concerns about data privacy, security, and bias, which can harm individuals and society.
Designing for Human-AI Interaction
In design practice, research on Human-AI interaction has led to new design guidelines, such as Microsoft’s Human-AI Interaction Guidelines. These new design guidelines are important because AI, as a design material, has properties distinct from those of static websites or interfaces. Generative AI, both for text, images, and other creative outputs, is not as static as what we’re used to designing for.
For example, when you use ChatGPT, you can ask it questions and prompt it on an unlimited number of topics. This makes it highly dynamic and impossible for designers to account for all possible states and outputs. Traditional methods, such as Wizard of the Oz, are, for example, particularly unsuitable for dynamic AI systems.
Instead, design guidelines for AI-based products and features focus more on educating users about the AI system, for example, through explainable AI (which explains how the AI model works), providing feedback to the user, allowing the user to provide feedback to the system, error handling, and allowing for global controls.
Related: How Generative AI is Revolutionizing Design Workflows