Applied Statistics in the Era of Artificial Intelligence: A Review and Vision

TL;DR

The paper surveys how AI reshapes applied statistics and, in turn, how statistics can advance AI, outlining a non-linear eight-step workflow and tracing historical, traditional, and emerging areas. It presents concrete engineering-statistics examples (GPU reliability, sensor data clustering) and emerging domains (HPC performance, renewable energy, social networks, additive manufacturing), while detailing AI-enabled statistics (uncertainty quantification, XAI, reliability) and AI-assisted statistics (automated analysis, computing tools). It also articulates AI-for-statistics topics (AI models for statistics, computing enhancements) and statistics-for-AI topics (model properties, UQ, OOD detection, AI assurance, test planning) and ends with a forward-looking view on automation and the evolving roles of statisticians. The work aims to guide practitioners and researchers in leveraging AI to advance applied statistics and to prepare statisticians for an AI-augmented landscape. It emphasizes a human-centered approach, balancing automation with domain knowledge, rigorous inference, and responsible decision-making.

Abstract

The advent of artificial intelligence (AI) technologies has significantly changed many domains, including applied statistics. This review and vision paper explores the evolving role of applied statistics in the AI era, drawing from our experiences in engineering statistics. We begin by outlining the fundamental concepts and historical developments in applied statistics and tracing the rise of AI technologies. Subsequently, we review traditional areas of applied statistics, using examples from engineering statistics to illustrate key points. We then explore emerging areas in applied statistics, driven by recent technological advancements, highlighting examples from our recent projects. The paper discusses the symbiotic relationship between AI and applied statistics, focusing on how statistical principles can be employed to study the properties of AI models and enhance AI systems. We also examine how AI can advance applied statistics in terms of modeling and analysis. In conclusion, we reflect on the future role of statisticians. Our paper aims to shed light on the transformative impact of AI on applied statistics and inspire further exploration in this dynamic field.

Figures (11)

• Figure 1: Flowchart for the main steps in applied statistics.
• Figure 2: Illustration of the layout of Titan supercomputer. Figure reproduced with permission from Ostrouchov et al. (2020).
• Figure 3: Heatmap for the DBE failure proportion over the $8\times 25$ spatial domain.
• Figure 4: Examples of data from two sensors, with each line shows a trajectory from the senor. Figure reproduced with permission from INFORMS.
• Figure 5: Illustration of clustering results based on data from the four sensors stayed in the model. Figure reproduced with permission from INFORMS.