Reimagining the Philosophy of Experimentation in Chemistry Education: Embracing AI as a Tool for Scientific Inquiry

Synthesis: Reyes and Regala (2026) offer a philosophical examination of how AI reshapes the epistemology, ontology, and methodology of experimentation in chemistry education. AI-driven methods challenge the traditional realism/positivism of the chemistry laboratory, placing AI predictions in a "liminal ontological space" — neither wholly hypothetical nor fully real until empirically proven. The authors trace ontological shifts (e.g., AI-predicted compounds and titration curves blur the virtual/physical boundary), agency shift (AI's role in decision-making raises questions about the locus of scientific Creativity), and the risk of over-reliance on AI that automates routine tasks and alters how experiments are conceived. They argue for a balanced human–AI framework that maximizes AI's benefits while preserving the essential elements of scientific reasoning, intuition, creativity, and discovery.

AI and the epistemology of experimentation

The article examines how AI challenges the epistemological foundations of chemistry experimentation — how knowledge is generated, validated, and trusted. AI predictive models and Simulation function by analyzing datasets to produce models and forecasts that may not be observable in a physical laboratory, raising questions about what counts as evidence, how results are validated, and whether AI-generated conclusions can be distinguished from direct observation.

AI and the ontology of experimentation

Chemistry's ontology has been rooted in realism — substances and phenomena exist independently of the observer, and experiments offer direct insight into them. AI complicates this:

• AI systems can forecast reaction results, propose novel reactions/molecules not yet created, and generate predicted results that occupy a liminal ontological space between abstraction and potential reality.
• In an acid–base titration, students may interact with three representations simultaneously: an AI predictive model, the physical titration, and real-time AI analysis — introducing a complex landscape where virtual and physical chemistry become increasingly permeable.
• This shifts chemistry from realism toward a Constructivism or post-Constructivism perspective where reality is a collaborative creation of human and machine entities.

Agency shift and methodological implications

Using the synthesis of a new compound as a worked example, the authors detail how AI integration changes experimental methodology and introduces agency shift:

• Experimental control — AI can optimize reaction conditions in real time, raising questions about the locus of scientific creativity and responsibility for outcomes.
• Randomization — AI automates randomization with high precision but may introduce algorithmic bias and opacity.
• The recommended approach is a cooperative human–AI methodology where AI optimizes design and provides data-driven insights while the educator/researcher provides theoretical foundations and evaluates the feasibility of AI-suggested conditions.

Over-reliance and the need for balance

The article cautions against over-reliance on AI in chemical experimentation, where AI automates routine tasks and fundamentally alters how experiments are conceptualized and conducted. It calls for a framework for integrating AI into chemistry education that maximizes benefits while preserving the essential elements of scientific reasoning and discovery — a theme that resonates with the knowledge base's Reducing AI Misuse and Cognitive Offloading concerns, and with the philosophical treatment of Philosophy of AI in Education.

What this means for practice

• Instructors. Run AI predictions and wet-lab measurements side by side — comparing AI-predicted reaction yields or titration behavior with what actually happens — and spend class time on where and why the two diverge.
• Instructors. Make the "black box" the object of study: have students evaluate the quality of data fed to a model, question whether the algorithm suits the specific question asked, and demand a transparency route before accepting an AI-generated inference.
• Instructors. Assess the interpretive work explicitly. As AI absorbs data interpretation, the experimenter's role shifts to problem formulation, oversight, and validation, so grade students on their ability to contextualize AI output within a scientific framework rather than on producing it.
• Researchers. Report AI-predicted results with their validation route attached — reproducibility checks, cross-validation against traditional methods, and stated limits — because AI-generated knowledge occupies a liminal space between the hypothetical and the verified until it is confirmed empirically.

Limitations

• This is a philosophical/conceptual analysis, not an empirical study: the authors position it as proposing "a foundational framework to map the landscape of change" rather than resolving the questions it raises, and they explicitly bracket issues such as whether AI requires redefining empirical knowledge.
• The concrete cases — the three simultaneous titration representations and the worked synthesis of a new compound — are illustrations, not findings from a classroom intervention, so the paper reports no measured effect on student learning.
• Its scope is chemistry education rather than professional chemistry practice, framed through curriculum, instruction, engagement, and assessment; the claims about epistemology in research settings are argued rather than tested.

Citation

Reyes, R. L., & Regala, J. D. (2026). Reimagining the philosophy of experimentation in chemistry education: Embracing AI as a tool for scientific inquiry. Science & Education, 35, 709–754.