A Review of User Interface Design for Interactive Machine Learning

Analysis

Why This Paper Matters

Interactive Machine Learning (IML) promises to democratize data analysis by enabling non-experts to apply their domain knowledge without deep ML expertise. However, the success of IML hinges critically on user interface design, an area that lacked consolidated guidance. This 2018 review by Dudley and Kristensson fills that gap by systematically characterizing IML from an interactive systems perspective. It matters because it provides a structured framework—a generalized IML system model and a set of design principles—that can guide both researchers and practitioners in building more effective, user-friendly IML interfaces.

The paper is particularly significant for the AI community as it highlights the co-adaptive nature of human-machine interaction in IML, where both the user and the model learn and adapt over time. This perspective shifts the focus from purely algorithmic improvements to the design of the interaction loop itself, which is often the bottleneck in real-world IML adoption. By grounding its principles in established HCI literature, the review ensures that IML interface design is not reinvented in isolation but builds on decades of interaction design knowledge.

Technical Contributions

The paper's main technical contributions are:

• A structural and behavioral model of a generalized IML system: This model decomposes the IML process into components (e.g., data, model, user, interface) and their interactions, providing a common vocabulary for designers.
• Solution principles for IML interfaces: The authors identify actionable principles such as providing real-time feedback, supporting iterative refinement, managing user trust, and balancing automation with user control.
• Contextualization within HCI: Each principle is linked to relevant HCI theories and empirical findings, strengthening its validity and providing a richer design rationale.
• Identification of future research strands: The paper pinpoints key UI research areas—such as explainability, mixed-initiative interaction, and adaptive interfaces—that are critical for advancing non-expert IML.

Results

As a review paper, this work does not present new experimental results or quantitative metrics. Instead, its primary result is a synthesized, structured knowledge base: a taxonomy of IML systems, a set of design principles, and a research agenda. The paper's impact is evidenced by its 377 citations, indicating its adoption as a reference in subsequent IML and HCI research. The principles it outlines have been used to inform the design of various IML tools and systems, though the paper itself does not benchmark these against alternatives.

Significance

The broader significance of this paper lies in its role as a bridge between the machine learning and human-computer interaction communities. By framing IML as a co-adaptive interactive system, it encourages ML researchers to consider user experience and HCI researchers to engage with ML capabilities. This cross-pollination is essential for building AI systems that are not only powerful but also usable and trustworthy. The paper's emphasis on non-expert users aligns with the growing trend of AI democratization, making it a timely and influential contribution that continues to inform interface design for interactive AI systems.