AI as scientific instruments: a new vision for science education

A new preprint from Bewersdorff, Rojas, and Zhai envisions a practice-centered approach to integrating AI into science education. Rather than teaching AI as an abstract, standalone topic, the authors argue AI should be treated as a set of scientific instruments—tools students use within the investigative practices defined by the Next Generation Science Standards (NGSS). These instruments are pedagogically bounded: their controls are simplified, but their core scientific functions are preserved, allowing authentic inquiry without overwhelming complexity. The framework has two aims: engaging students in real scientific practices, and building discipline-based AI literacy (DAIL)—an understanding of how AI is used in science and where it can mislead.

The article focuses on three exemplar AI instruments representing the core of inquiry: computer vision for observing phenomena, clustering for analyzing data, and generative modeling for building models. Each instrument includes a distinct reflection point that prompts students to critically evaluate the AI's limitations and biases. The authors also address agentic AI—systems that operate across the entire inquiry process—arguing that students should first develop a solid foundation in scientific inquiry and use of individual AI instruments before relying on autonomous agents that might obscure the reasoning behind each step.