Regulating AI – on EU’s draft Artificial Intelligence law

Source: This post is created based on the article “Regulating AI – on Artificial Intelligence” published in Indian Express on 18^th May 2023.

Syllabus Topic: GS Paper 3 – Artificial Intelligence

News: The European Parliament’s committee has approved a draft artificial intelligence (AI) law.

What are the provisions of EU’s draft Artificial Intelligence law?

Scope of the act: The proposed Act aims to establish standards for AI deployment within the European Union (EU). This act applies to any entity serving EU residents, much like the EU’s General Data Protection Regulation (GDPR).

The Act prohibits several types of AI tools that infringe on privacy or enable discrimination.

Risk based Categorisation: AI will be categorized based on risk levels, from minimal to limited, high, and unacceptable. High-risk tools will be allowed under strict oversight, auditing, and transparency requirements.

Banned Systems: It will ban systems such as real-time remote biometric identification systems used in public spaces and post-real-time use of such systems. Certain exceptions are provided for law enforcement in serious crimes following judicial authorization.

Other prohibited systems include biometric categorization based on sensitive attributes, such as race, gender, or political orientation, and predictive policing systems.

The Act also prohibits the widespread scraping of biometric data from social media or CCTV footage to create databases. Also, “emotion recognition” systems used in several contexts to detect discomfort through facial expressions or body language.

High-risk AI will include systems that ae potentially harmful to health, safety, fundamental rights, or the environment. It will include AI used in political campaigns to influence voters and recommendation systems used by large social media platforms.

Generative models, like GPT, would need to comply with robust transparency requirements.

Exemptions: There are provisions for exemptions for research activities under open-source licenses, and the Act supports the establishment of regulatory sandboxes to test AI before deployment.

The Act further introduces legislation to facilitate citizens in filing complaints and requesting explanations for AI-based decisions affecting them.