A COMPARATIVE ASSESSMENT OF GENERATIVE, SYMBOLIC, AND MACHINE LEARNING TOOLS IN STEM TEACHING PEDAGOGY

Abstract

The rapid integration of Artificial Intelligence (AI) in engineering science requires a critical evaluation of tool reliability and algorithmic limits. Grounded in contemporary frameworks of AI literacy and the "plausibility trap" of using probabilistic engines for deterministic tasks, this paper investigates the performance profiles of generative and symbolic platforms. We evaluate ChatGPT and Wolfram Alpha across localized statistical and algebraic computations foundational to engineering curricula. The empirical results demonstrate that the symbolic computation engine significantly outperforms the probabilistic language model in arithmetic accuracy and graphical replication, underscoring the necessity of algorithmic verification frameworks. Additionally, this study benchmarks engineering software environments by developing ordinary least squares and polynomial regression models within MATLAB and Python, utilizing the comprehensive Yacht Hydrodynamics dataset to predict residuary hull resistance. While linear formulations show numerical convergence, nonlinear variations emerge due to differences in baseline model design design parameters, with MATLAB's structural matrix capturing a higher coefficient of determination  than the default Python setup . Mirroring recent regional findings on the double-edged impact of AI dependence on academic achievement and data constraints, these findings indicate that while AI and software tools accelerate engineering analytics, explicit cross-platform verification remains essential. Ultimately, this work provides actionable insights for optimizing data science applications and automated engineering design pipelines.