Rethinking the Risk of Uncertainty: Human-AI Interaction in Household Mycology

BackgroundIn recent years, the rise of AI-mediated technological assistance has impacted applied mycology. Various tools employ AI to analyze images of Macromycetes fruiting bodies for species identification. This trend has sparked widespread interest in online applications despite the potential risks associated with relying on AI-generated advice. MethodsWe conducted a comparative analysis of popular AI-based mushroom identification tools using over 100 original photographs of fungi fruiting bodies from nearly 60 species, taken in real-world conditions. Reference searches were conducted with mushroom names in five languages, including Latin species names. Functional test scores and an overall accuracy score were calculated for twelve selected AI applications to evaluate their general reliability. ResultsThe AI-based applications evaluated in the study were able to recognize only a portion of the provided mushroom images. Even the best-performing tools frequently failed to accurately identify fruiting bodies in real-world conditions. None of the tested applications consistently provided a single, correct species name. Instead, users were often presented with multiple options, among which the correct answer might have been found. ConclusionWhen addressing mycological queries, it is crucial to recognize the inherent risk of relying solely on AI-mediated resources for mushroom identification. Various limitations hinder their effectiveness in real-world environments. These tools should only be viewed as supplementary aids since they are inadequate for making definitive or safety-critical decisions.