This study explores an innovative approach in chemical education by engaging undergraduate students with no prior coding experience in generating interactive HTML-based models of borane clusters using generative artificial intelligence (GenAI) tools. The project centers on developing student's undrstanding on Wade's rules, which is taught in an undergraudate Inorganic Chemistry Course in The Univesrity of Hong Kong. Upon completing this activity, student is expected to have a solid understanding for predicting the geometries of electron-deficient borane polyhedra.
Students were tasked with prompting GenAI models to produce JavaScript/HTML code (often incorporating libraries like Three.js or basic canvas/WebGL) that visualizes and allows interactive manipulation of borane structures, including vertex connectivity, polyhedral shapes, and electron-pair-based bonding representations. The project aimed to achieve three primary objectives: (1) enhance students' coding proficiency through iterative AI-assisted development, enabling students to create functional web-based applications without traditional programming instruction; (2) develop advanced prompting skills by refining queries to product accurate, and meaningful code outputs; and (3) critically evaluate common errors and limitations in GenAI predictions of borane geometries according to Wade's rules.
Preliminary findings indicate significant gains in self-reported coding confidence and prompting expertise, with students successfully producing interactive 3D/2D models for clusters like B₆H₆²⁻ (closo octahedral) and B₅H₉ (nido square pyramidal). However, analysis of generated outputs revealed recurring GenAI inaccuracies, such as incorrect assignment of skeletal electron pairs, misclassification of cluster types (e.g., confusing nido and arachno forms), improper polyhedral vertex counting, and structural distortions that violate Wade's electron-counting principles. These errors provided rich opportunities for student-led debugging and deeper conceptual understanding of cluster bonding theory.
This work demonstrates that guiding students to actively critique and correct GenAI-generated chemical models not only builds essential AI literacy but also significantly strengthens their mastery of Wade’s rules and electron-deficient cluster bonding concepts.