Machine unlearning: Researchers make AI models ‘forget’ data

A new method developed by Tokyo University of Science researchers allows large-scale AI models to selectively "forget" specific data classes without accessing their internal architecture. This "black-box forgetting" technique could make AI models more efficient, resource-friendly, and privacy-compliant. In tests, the method made the CLIP vision-language model forget about 40% of target classes. The innovation has potential applications in specialized AI tasks, image generation, and addressing privacy concerns in industries like healthcare and finance.