<p>Artificial intelligence (AI), a technology rooted in scientific advancements, is increasingly reshaping the research practice. Although references to AI have become more frequent in scientific publications, its specific roles remain insufficiently clarified. This study explores the bidirectional relationship between AI and scientific research–namely, how AI supports scientific inquiry and how scientific knowledge contributes to AI development. Referring to Ihde’s theory of human-technology relations and its implication for the philosophy of science, we construct a two-level classification framework to categorize the different ways in which AI is mentioned in research articles. The first level categorizes AI mentions in articles into three main types–<i>Science for AI</i>, <i>AI for Science</i>, and <i>Others</i>. The second level further divides the former two types into three and four subtypes, respectively, capturing nuanced differences in how AI is framed or utilized in scientific research. Empirical analysis is conducted in the fields of oncology, nanoscience and nanotechnology, as well as in meteorology and atmospheric sciences. The findings indicate that while the mention of AI in research articles has become widespread, the ways in which AI is mentioned are limited in variety. We not only identify the conceptual differences in the focus of AI mentions, but also uncover the disparities in the intensity and ways of AI mentions across research articles from various countries. We have also observed that articles focusing on <i>Science for AI</i> are likely to receive more citations, whereas those focusing on <i>AI for Science</i> tend not to yield an additional gain in terms of citation impact. By distinguishing AI’s roles across research fields, this study provides a structured bibliometric framework for analyzing how emerging technologies become differentially embedded in scientific knowledge production.</p>