Artificial Intelligence is progressing rapidly, putting regulators, governments, and creators in a challenging position to keep pace. In this issue of AI Pulse, we delve deeply into the realm of AI regulation, covering Brazil’s confrontational stance against Meta, the defense of creators’ copyrights, recent alerts from law enforcement, and pondering the implications once autonomous AI starts exhibiting self-awareness.
Governance of AI: Leading the Charge
Upon the emergence of generative AI, critiques surfaced immediately regarding AI companies’ extensive usage of data for model training. Debates persist over the ownership of data and legitimate usage rights, along with the manner and location of its utilization.
The rapid pace and lack of transparency in AI advancements have made it challenging for content holders, regulators, and lawmakers to adapt, but efforts are still underway. In recent weeks and months, authorities have curbed the ambitions of major players while communities have expressed their grievances vociferously—sometimes resorting to strong measures.
This edition of AI Pulse highlights several significant advancements in AI governance and emphasizes the increasing necessity of safeguards with the emergence of ‘agentic’ AI.
Latest Developments in AI Governance
Regulators Take a Stand Against the AI Frenzy
In early July, Brazil dealt a blow to Meta by suspending the company’s initiative to train AI models using Instagram and Facebook posts. As per a report by the BBC, Meta intended to extract public posts, images, and comments for AI training, including content from children and teenagers. Meta asserted that its plans aligned with Brazilian privacy regulations; however, the country’s data protection agency, ANPD (Autoridade Nacional de Proteção de Dados), disagreed.
Following pressure from European governments, Meta had previously withdrawn a similar proposal in the EU and lately announced that it would refrain from deploying its upcoming multimodal AI model in Europe and Brazil due to regulatory ambiguity. (Multimodal AI models encompass a variety of media types, not just text, video, or images on their own.) Whether this decision is linked to the EU’s enactment of the AI Act in July, the commencement of the act signifies that companies must swiftly adhere to the regulations.
Ownership of Data: Debated Territory
AI regulation delves into issues of copyright and data confidentiality governed by pre-existing legal structures. In January 2024, Italy suggested that OpenAI was potentially infringing the EU’s GDPR, an ongoing case.
A major concern is defining ‘publicly available’ data, a term viewed distinctively by AI firms and regulators. As highlighted in a recent Axios feature, ‘publicly available’ and public domain do not align perfectly, with AI developers often accessing data without explicit permission from users or creators, whether intentionally or unintentionally. When a July feature in WIRED suggested that some companies might have breached YouTube policies by utilizing video subtitles for AI training, it was acknowledged that this might have occurred inadvertently since the dataset was acquired by a third party.
About Author
