Exploring Alternative Approaches to Contemporary AI Testing

Abstract

While the first artificial intelligence test proposed by A. Turing has a long-standing tradition, it is no longer sufficient to meet today's challenges. There is now a strong need for more multi-contextual testing and comparison of AI solutions. Among other things, the emergence of the new Chinese language model, DeepSeek—designed to compete with Western solutions such as ChatGPT, Copilot, and Gemini—has caused quite a stir. It has not only influenced financial markets but also ignited a broader discussion about the quality of contemporary AI systems. As a result, there is a growing need to systematically test and compare these tools. The aim of this paper is to present an original attempt to identify and systematize contemporary approaches to AI testing. The discussion is framed by a reference to the Turing Test and its relevance in the modern context. The author seeks to identify common features between software testing, human intelligence testing, and AI testing. Subsequently, based on a critical analysis of relevant literature and other available sources, the paper outlines and organizes the types of tests currently in use. The conducted considerations allowed for the identification of three trends in contemporary AI testing: tests imitating or referring to human intelligence testing, tests analogous to approaches used in software engineering, and tests based on parameters.