The application of artificial intelligence in the automated design of labeling systems: An analysis of experiment results

Abstract

Thesis/aim: The aim of the paper is to investigate the potential of using artificial intelligence (AI) to automatically generate labels in the form of headers and options for labeling systems. Research methods: Two experiments were conducted, in which five chatbots - ChatGPT 4.0, ChatGPT Plus, Gemini Free, Gemini Advanced, and Claude Sonnet - were asked to generate sets of labels according to a given prompt and four sets of input data. The evaluation of chatbots' capabilities was carried out using a comparative analysis method. The analysis included assessing the length, consistency, specificity, mutual exclusivity, and descriptiveness of the generated labels. Results: The chatbots generated similar or very different labels depending on the input data. In terms of generating navigation system options, they scored the following points out of 40 possible: ChatGPT 4.0 - 34 pts, ChatGPT Plus - 38 pts, Gemini Free - 29 pts, Gemini Advanced - 25 pts, Claude Sonnet - 35 pts. They performed slightly differently in generating headings, receiving ChatGPT 4.0 - 36 pts, ChatGPT Plus - 35 pts, Gemini Free - 29 pts, Gemini Advanced - 36 pts, and Claude Sonnet - 34 pts. ChatGPT Plus performed best in the experiments, while Gemini Free performed the weakest. Conclusions: Both paid and free versions of chatbots perform well in generating labels. Each of the tested chatbots can be a source of inspiration for an information architect, but final design decisions should be made after conducting appropriate user tests.