Essential Insights on EU's AI Act: Critical Details Explained
The European Union (EU) is set to introduce its first comprehensive law on artificial intelligence (AI), the AI Act, following a series of discussions among member states. The law, still subject to voting by the European Parliament in early 2024, will impose strict regulations on companies that develop and utilize foundation models, such as large language models.
Under the AI Act, companies building foundation models will be required to prepare technical documentation, comply with EU copyright law, and disclose the content used for training. Advanced foundation models that pose 'systemic risks' will face increased scrutiny, including evaluating and mitigating risks, reporting significant incidents, implementing cybersecurity measures, and reporting their energy efficiency.
The AI Act also bans certain AI uses deemed harmful to human rights and democracy, including biometric categorization systems using sensitive characteristics, untargeted facial recognition database creation, emotion recognition in workplaces and educational institutions, social scoring based on behavior or personal characteristics, AI manipulation of human behavior, and AI exploitation of vulnerable individuals.
High-risk AI systems, such as those affecting elections or finance, will be mandated to conduct a fundamental rights impact assessment and demonstrate transparency and accountability to the public. The EU is encouraging robust cybersecurity solutions in various industries, including education, to protect sensitive data and maintain the integrity of systems.
While the AI Act is heralded as a groundbreaking regulatory framework, concerns have been raised about its effectiveness. Adam Satariano of The New York Times writes that some provisions may not take effect for 12 to 24 months, potentially affecting AI development, and that policymakers struggled to balance innovation and safety during negotiations.
Ursula von der Leyen, President of the European Commission, expressed optimism, stating that the AI Act would make a significant contribution to the global development of trustworthy AI. The EU intends to continue working on international AI guidelines and has recently supported the Hiroshima AI process agreement on International Guiding Principles and a voluntary Code of Conduct for Advanced AI systems.
According to Scott Roxborough of The Hollywood Reporter, the EU's transparency demands for large general-purpose programs will primarily apply to the most powerful large-language models, defined as those that use foundational models requiring over 1025 Floating Point Ops (FLOPs) for training. Companies violating the regulations may face fines of up to 7% of their global sales.
The AI Act is also perceived as beneficial for the book publishing industry, as earlier versions of the bill required companies using generative or foundation AI models to provide summaries of any copyrighted works used during training. As businesses integrate AI into their operations, partnering with reliable AI development services to adhere to ethical standards and legal frameworks is advisable.
- The AI Act, a comprehensive law on artificial intelligence in the European Union, will impose strict regulations on companies developing and utilizing foundation models, requiring them to prepare technical documentation, comply with EU copyright law, and disclose the content used for training.
- Advanced foundation models that pose 'systemic risks' under the AI Act will face increased scrutiny, necessitating evaluating and mitigating risks, reporting significant incidents, implementing cybersecurity measures, reporting their energy efficiency, and adhering to regulations that may result in fines of up to 7% of their global sales for non-compliance.
