Statistics in the Age of AI 

Jae-Kwang Kim

Artificial intelligence is reshaping nearly every data-driven discipline, but the challenge is unusually existential for statistics. The question is no longer whether AI will change how statistics is practiced, but whether university statistics departments can adapt quickly enough to remain intellectually and institutionally central in this new environment. 

This concern is no longer hypothetical. The recent closure of the department of statistics at the University of Nebraska-Lincoln has drawn widespread attention within the statistical community. Regardless of the specific local factors behind the decision, it underscores a broader reality: The long-term survival of statistics departments can no longer be taken for granted. 

In an era in which computation, model fitting, and even statistical explanation are increasingly automated, the traditional sources of disciplinary authority are being fundamentally reconfigured.  

The challenges facing statistics departments can be understood through an evolutionary biology framework. Concepts such as environmental shift, selection pressure, and speciation offer a lens for diagnosing why some responses to AI are likely to fail—and why others may allow statistics to not only survive, but to reassert leadership across science, industry, and public policy. 

AI as an Environmental Shift 

In evolutionary biology, survival depends on how organisms respond to changes in their environment. Evolution is driven by four fundamental elements: environmental shift; variation; selection; and adaptation (or speciation). The rise of AI satisfies all four simultaneously. 

AI represents a rapid environmental shift characterized by the democratization of computational power, the automation of model implementation, and the partial substitution of tasks that once required specialized statistical expertise—coding, derivation, and even explanation. As these capabilities become widely accessible, the scarcity that once defined many statistical skills is disappearing. What once served as a competitive advantage is rapidly becoming a commodity. 

This does not imply that statistics is becoming obsolete. Rather, it means the basis on which statisticians are evaluated and selected is changing. 

Shifting Selection Pressures 

Historically, statistical expertise was closely associated with technical mastery: precise computation; efficient model implementation; and command of specialized inferential techniques. In the AI era, these abilities remain necessary, but they are no longer sufficient. 

Increasingly, selection pressure favors problem formulation over computation, assumption assessment over model execution, and integration across methods over mastery of isolated techniques. Knowledge, itself, is no longer enough. What matters is meta-knowledge—understanding when methods apply, when they fail, and how conclusions depend on often untestable assumptions. 

In short, we are witnessing a shift from technical skill to epistemic competence—from “how to compute” to “how to reason.”

Three Evolutionary Paths for Statistics Departments 

Against this backdrop, statistics departments appear to face the following three broad strategic paths: 

Conservative Evolution: The Risk of Stagnation

The first path is conservative adaptation—maintaining existing curricula while defending them by asserting that “statistical foundations still matter.” This claim is undoubtedly true. Probability theory, inference, and statistical reasoning remain essential. However, without reinterpretation and recontextualization, this defense risks sounding inward-looking and disconnected from the realities students and employers face. In evolutionary terms, this is stabilizing selection in a rapidly changing environment, a strategy that often leads to marginalization or eventual extinction. 

Mimicry: Competing on Unfavorable Terms

A second path is mimicry: reshaping statistics programs to resemble computer science or machine learning degrees. While tempting, this strategy is inherently risky. In an academic ecosystem in which computer science departments already dominate this niche—with greater scale, resources, and institutional momentum—statistics departments that pursue mimicry risk becoming second-tier substitutes.

To be clear, the problem is not engagement with machine learning methods. Modern statistics education must grapple seriously with these tools. The distinction lies in the pedagogical lens through which these methods are taught.

The danger lies in teaching ML purely as a set of computational techniques, without statistical re-interpretation. Engaging with machine learning through an epistemic lens—asking when strong training performance fails to generalize, how to quantify uncertainty in black-box models, and what assumptions justify a prediction—is not mimicry; it is the integration speciation requires.

Mimicry occurs only when we surrender that lens. In evolutionary biology, mimicry succeeds when it confers protection. In this case, it does the opposite: It strips statistics of its comparative advantage, leaving it to compete solely on computation—a battle it cannot win. 

Speciation: Creating a Distinct Statistical Niche

The most promising path is speciation: the development of a clearly differentiated niche that builds on statistics’ unique comparative advantages. This strategy does not reject AI or machine learning; rather, it situates them within a broader inferential and epistemological framework that statistics is uniquely positioned to provide.

The rapid rise of causal inference offers a concrete example. Demand for principled reasoning about causality, assumptions, and generalizability has grown across science, industry, and policy. Statisticians, trained to think carefully about data-generating mechanisms and uncertainty, are well positioned to lead in this space. This is not accidental; it reflects a successful instance of evolutionary differentiation already underway.

Diversity of Background as an Adaptive Advantage 

From an evolutionary perspective, diversity is not merely beneficial; it is essential for adaptation. As AI increasingly automates the syntax of statistics—coding and calculation—the discipline’s value shifts toward semantics: understanding data generation, bias, and context. 

Here, the intellectual diversity of statistics students becomes a strategic asset. Students entering from economics, biology, or the social sciences bring distinct inferential instincts: counterfactual reasoning; hierarchical thinking; or sensitivity to measurement error. In the AI era, these domain-specific insights are the raw materials of epistemic competence. They provide the context necessary to ask the questions AI cannot: Is this proxy valid? Does this causal claim align with theory? 

By integrating these varied perspectives with a shared core of probability, statistics can achieve speciation—colonizing intellectual niches that pure computation cannot reach. Diversity does not dilute the discipline; it powers its survival. 

Conclusion: Survival Through Renewal 

The rise of AI does not signal the end of statistics. But it does mark the end of a particular equilibrium, one in which technical expertise alone guaranteed relevance and authority. 

Whether statistics emerges from this period diminished or renewed will depend on how deliberately it adapts: embracing speciation rather than mimicry; cultivating epistemic competence; and leveraging intellectual diversity. 

But this evolutionary process is ongoing, not a one-time adjustment. Going forward, statistics departments must repeatedly ask themselves: What are the core “genes” of statistics? What can be safely de-emphasized in the AI era? How should undergraduate and graduate training evolve differently? Should statistics remain an independent discipline, or thrive through deeper forms of symbiosis? 

Evolution offers no guarantees—only opportunities for those willing to adapt deliberately.

Jae-Kwang Kim

Jae-Kwang Kim is dean’s professor in the department of statistics at Iowa State University. He is a Fellow of the American Statistical Association and Institute of Mathematical Statistics. He has published two books, Statistical Methods for Handling Incomplete Data and Statistics in Survey Sampling. He is serving as the vice chair of the ASA Professional Issues and Visibility Council.