Finally A.I. Goes Crazy: Do You Know What’s New!

Researchers from Rice and Stanford University have discovered that feeding artificial intelligence (AI) models with content generated by the same AIs leads to a drastic decline in result quality. This problem, called MAD (Model Autophagy Disorder), confirms the well-known principle of “garbage in, garbage out”. If the input data is of low quality, the output will be too. The researchers highlighted that generative models need fresh real data created by humans to maintain quality. If an algorithm is trained only on artificial data, the model becomes progressively less accurate and diverse. This has significant implications for people and companies working with AI. For example, Google is developing a new version of its search engine that could present AI-generated texts as results. However, if websites start to integrate generated texts, a problem could occur. Modern algorithms are increasingly searching for new data on the Internet, and the increase in AI-generated content could trigger the MAD mechanism. Although the document has not yet been subjected to peer review, it represents an important indication for those working with AI. In conclusion, while AIs can perform generative tasks, it is necessary to ensure the use of fresh real data to maintain quality and prevent the deterioration of generative models.