The dawn of generative artificial intelligence (AI) systems has captured the world by storm, as a technology that often appears – at least at first glance – efficient, sophisticated and capable of carrying out complex human tasks. Companies have been particularly successful at galvanizing fascination and enthusiastic investment in such technologies.
Behind the veneer of efficiency, sophistication and complexity, however, hides a reality of human rights-violating design principles, which are akin to those found in many of the most problematic AI tools predating generative AI.
Generative AI systems are machine learning tools that have been trained on massive datasets – often without the knowledge and consent of those from whom the data originates, such as social media users and artists – to create “new” data or content through mimicking or approximating the data they have been fed. This data often comes in the form of text, images, video, music and other multimodal outputs, and could not exist without the input, processing and transformation of existing personal, creative and behavioral data. Generative AI systems are “prompted” to generate such outputs by a user requesting outputs within particular parameters.
This briefing examines how standalone generative AI systems, based on unlawful web scraping, are in conflict with international human rights law (IHRL) and standards through their design, development and deployment. While these technologies promise sophisticated automation and efficiency, they rely on data collection and model training practices that abuse privacy rights, enable discrimination, and threaten freedom of expression and thought.
Read “Violations in the Shell: Exposing the Human Rights Costs of Generative AI.”