Utilizing Publicly Accessible Machine Learning for Training Student Formulators in Personal Care Product Formulation: Specification-Driven and Cost-Conscious Experiments

Abstract

Abstract This work explores the application of the questionable use of machine learning (ML), specifically the ChatGPT 3.5 system, in the training of student formulators. Here, an experiment is undertaken to explore the ability of ML to aid in training students in the role of formulators of a personal care product. The focus is on whether or not the students can successfully rely on ML to guide them through the formulation process of a 10-minute hydrating face mask recipe. While exploring the iterative process of recipe adjustments with ML, it was found that the language model demonstrates the ability to help formulators in training due to its above-average knowledge in chemistry, but when given clear prompts, it performs much better at providing helpful suggestions for ingredient substitutions. However, ML lacks a reliable memory, even within a single extended conversation and struggles with mathematical calculations. ML is not found to be proficient in accurately calculating cost adjustments. Additionally, the contribution of ML may only be marginally helpful in the training of more seasoned formulator. Despite its limitations, ML can quickly and effectively, in the hands of student formulators in-training, provide direction and support to produce and improve upon a base formula resulting in a quality product.