Llega una herramienta que previene el uso de obras de arte en la capacitación de modelos de IA

The article highlights a critical ethical conflict in the development of generative artificial intelligence, specifically concerning image generation models. These systems are frequently trained on copyrighted artworks without consent or compensation, creating a significant imbalance of power. This context is vital for open data discussions because it challenges the assumption that publicly available data can be freely harvested for commercial AI training, urging a reevaluation of data provenance and intellectual property rights in the digital ecosystem. To counter this, new defensive technologies like Kin.art are emerging to "poison" training datasets. Unlike previous solutions, this tool obscures image details and alters metadata to prevent unauthorized ingestion, acting as a preventive measure rather than a remedial one. By making the training process less effective or more costly for developers, artists can protect their work. This shifts the narrative from passive exposure to active data sovereignty, demonstrating how technical interventions can enforce ethical boundaries in open data usage. The relevance to open data lies in the potential for these tools to become standard services for platforms of all sizes. As developers aim to offer protection against unlicensed AI scraping, the definition of open data may evolve to include opt-out mechanisms and data integrity standards. This underscores the urgent need for transparent, consensual frameworks in AI development, ensuring that the openness of data does not infringe upon the fundamental rights of content creators and owners.

Source: wwwhatsnew.com
Published on 2024-01-24

Related news

« Mark Tapscott

Leaning In On Making Consumer Tech A Force For Good, Speech By Nikhil Rathi Chief Executive UK Financial Conduct Authority, Delivered At The Imperial College London Business School »