Microbiome science needs AI competence, not just AI literacy

Abstract

Artificial intelligence is rapidly becoming embedded in microbiome research, yet the way microbiome scientists are trained has not kept pace. Existing discussions frequently emphasize AI literacy and responsible tool use. We argue that literacy alone is insufficient in a field where data leakage, incomplete metadata, ecological confounding, inappropriate model selection and unverified generative-AI outputs can produce convincing but biologically misleading conclusions. We propose an AI-competence framework comprising five connected capabilities: data and platform fluency, machine-learning competence, advanced model competence, large-language-model and agent competence, and biological closure. The final capability—the ability to connect computational predictions to independent data, mechanistic reasoning or experimental evidence—is essential for ensuring that predictive performance becomes meaningful biology. We further propose that competence should be demonstrated through reproducible scientific artifacts rather than course completion alone. Shared datasets, computational platforms and benchmark challenges could make such training broadly accessible while establishing common standards for evaluating AI competence in microbiome science.