Inteligência artificial como hipótese a ser testada: análise empírica de uma aula prática investigativa de Química no Ensino Médio
DOI:
https://doi.org/10.33871/23594381.2026.24.2.12305Abstract
This paper presents an empirical analysis of an investigative chemistry lab class conducted with 52 first- and second-year high school students at a public full-time school in Anchieta, Espírito Santo, Brazil. The activity integrated generative AI tools (ChatGPT and Gemini) into the experimental determination of ethanol content in gasoline samples. The methodological distinction lies in repositioning AI from a source of answers to a hypothesis to be tested against experimental evidence. A post-lesson questionnaire with 12 Likert-scale items and two open-ended questions was administered, organized into five constructs: learning, critical thinking, AI as a tool, responsible use, and the role of the teacher and the experiment. Quantitative data were analyzed through descriptive statistics, internal consistency (Cronbach's α), and Mann-Whitney tests between grade levels. Open responses underwent thematic content analysis. The instrument showed high overall reliability (α = 0.815). Means were positive across all constructs, particularly for "Teacher and experiment" (M = 4.59) and "Responsible use" (M = 4.47). First-year students reported higher acceptance of AI as a tool (p = 0.034). Qualitative analysis identified six categories, the most frequent being omission/incompleteness of responses (21.2%) and factual errors in identifying adulterated samples (13.5%). The findings support the claim that productive pedagogical use of AI depends less on the tool itself than on activity design that demands its verification.