What is overgeneralisation in AI models?

title: 'Fig. 1: Comparison of the strengths of humans and statistical ML machines, illustrating the complementary ways they generalise in human-AI teaming scenarios. Humans excel at compositionality, common sense, abstraction from a few examples, and robustness. Statistical ML excels at large-scale data and inference efficiency, inference correctness, handling data complexity, and the universality of approximation. Overgeneralisation biases remain challenging for both humans and machines. Collaborative and explainable mechanisms are key to achieving alignment in human-AI teaming. See Table 3 for a complete overview of the properties of machine methods, including instance-based and analytical machines.'

Overgeneralization in AI models refers to a phenomenon where models make incorrect predictions or assertions by applying learned patterns too broadly, ignoring critical differences. The text states, 'models overgeneralise, which means that they over-confidently make false predictions for (known or novel) concepts precisely because critical differences are ignored in prediction.' A specific example mentioned is 'hallucination,' which occurs when models deviate from their source of information, typically the pretraining data for large language models[1].


Related Content You May Like