AI driftとは 意味・読み方・使い方

名詞

AI drift (uncountable)

1. The gradual degradation of coherence in extended conversations with large language models due to context compression, leading to forgotten details, hallucinations, or off-topic responses.
   • 2024 December, Karin Rolanda Jongsma, Martin Sand, Megan Milota, “Why we should not mistake accuracy of medical AI for efficiency”, in NPJ Digital Medicine, volume 7, number 1, →DOI, page 57:

     AI systems will make mistakes, malfunction, or even breakdown. Mistakes can include biased outcomes, “hallucinations” and AI drift, which may seriously harm patients and therefore demand measures and increased awareness to counter these unwanted effects.

   • 2025 May, Hamidreza Namazi, Mohammad Mahdi Radfar, “Philosophy of medicine meets AI hallucination and AI drift: moving toward a more gentle medicine”, in Journal of Medical Ethics and History of Medicine, volume 18, page 2:

     AI hallucination and AI drift are two key challenges of large language models (LLMs). Hallucination occurs when an AI generates incorrect or fictitious responses, while drift refers to inconsistencies in answers to the same question when posed by different users or as a result of variations in how the question is asked.

2. The degradation of the accuracy of an LLM over its entire lifespan.
   • 2025 November, Erina Quinn, Christoph I. Lee, “Postdeployment Monitoring of Artificial Intelligence in Radiology: Stop the Drift”, in Journal of the American College of Radiology, volume 22, number 11, →DOI, page 1415:

     One major challenge being faced is the ability to surveil AI drift, or the tendency of deep learning algorithms to experience decreased accuracy over time due to changes in the population served, input data parameters, and technological updates.