European Union passes AI Act with comprehensive regulations

After extensive deliberations, the European Union has introduced a comprehensive set of regulations governing the utilization of AI technologies, marking a milestone globally. These regulations are poised to have extensive implications not only within the EU but also potentially setting a precedent for other nations contending with the challenges arising from the rapid evolution of AI.

The AI Act seeks to achieve a delicate equilibrium between fostering innovation in the AI domain and safeguarding fundamental rights, democracy, and environmental sustainability. It adopts a risk-based framework, mandating different levels of obligations contingent upon the potential impact and risks associated with various AI applications. For instance, developers of high-risk AI products are required to conduct thorough risk assessments to ensure compliance with legal standards prior to release.

Provisions

The Act prohibits certain AI practices considered too hazardous or unethical. These include discriminatory biometric categorization systems based on sensitive traits like race or gender, as well as the creation of recognition databases by scraping facial images from public spaces. AI systems are also addressed as being employed for social scoring, behaviour manipulation, and exploitation of vulnerabilities. While biometric identification systems are generally forbidden "in principle" for law enforcement agencies, exceptions are permitted under specific circumstances.

For AI applications categorized as "high-risk," obligations outlined in the Act include undergoing risk assessments, maintaining comprehensive logs, ensuring transparency and accuracy, and importantly, maintaining significant human oversight. 

Additionally, the Act empowers EU citizens to lodge complaints about AI systems and receive explanations for decisions affecting their rights.

The AI Act mandates that general-purpose AI systems and their underlying models adhere to specific transparency requirements, including compliance with EU copyright law and the publication of detailed summaries regarding the training data utilized. Moreover, for the most potent models posing potential "systemic risks," supplementary safeguards such as mandatory model evaluations and incident reporting are prescribed. Additionally, the Act addresses the issue of deepfakes by requiring the labelling of content that has been artificially generated or altered.

When it comes to regulating AI, India has also announced its efforts towards establishing an AI regulation framework, with Minister Rajeev Chandrasekhar indicating a potential release by June or July 2024. 

Future

The EU AI Act is anticipated to formally transition into law around May or June of this year, following the resolution of final obstacles, including approvals from member countries. Subsequently, the stipulations of the Act will gradually come into effect.

Systems employing prohibited AI practices, such as social scoring, will need to be discontinued within six months. Regulations concerning general AI applications like chatbots will take effect one year after the law's implementation. Non-compliant companies may face fines of up to €35 million or 7% of their global revenue for breaching the regulations. By mid-2026, the comprehensive regulatory framework for high-risk AI systems is projected to be fully enforced. Additionally, each EU member state will establish its own AI watchdog tasked with investigating citizen complaints of violations.