AI news recap: While Hollywood strikes, is ChatGPT getting worse?
Recent developments in artificial intelligence highlight critical tensions between technological advancement and human rights, particularly within the creative industries. The SAG-AFTRA strike underscores deep anxieties regarding consent and compensation, as actors fear studios might exploit AI to replicate their likenesses and performances without oversight or payment. This movement reflects a broader ethical crisis where the use of vast amounts of human-generated data to train models is increasingly viewed as exploitative, demanding new legal frameworks to protect intellectual property and creative labor. Simultaneously, the technical integrity of AI systems is facing unforeseen challenges that threaten their reliability. Research indicates that training models on AI-generated content can lead to a degenerative cycle, gradually degrading output quality. Furthermore, efforts to optimize models for efficiency or cost-cutting have resulted in significant performance drops in specific tasks, such as mathematics. These issues suggest that scaling up data and models does not guarantee improvement, and that current methodologies may inadvertently introduce biases or errors that are difficult to detect or reverse. These findings are profoundly relevant to open data because they expose the fragility and ethical risks inherent in unregulated data ecosystems. If open datasets used for training contain synthetic data or reflect systemic biases, they can produce unreliable and harmful AI outcomes. This situation calls for greater transparency, rigorous data governance, and a reevaluation of how open data is curated and shared, ensuring that the drive for larger datasets does not compromise accuracy, fairness, or the rights of data creators.
Source: newscientist.comPublished on 2023-07-29