Key features of European Union's 'historic' AI legislation

The proposed law defines four levels of risk in AI: Unacceptable risk, High risk, Limited risk, and Minimal or no risk.

Last week, European Union (EU) lawmakers reached a provisional deal on the world's first comprehensive set of rules regulating the use of artificial intelligence (AI), placing the European bloc EU ahead of the US, China, and the UK in the race to regulate this emerging technology.

The proposed legislation prioritises enhancing regulations concerning data quality, transparency, human oversight, and accountability.

The European Parliament is set to vote on the proposed AI Act early next year, with the legislation expected to take effect by 2025.

Now, let's delve into the key aspects of these proposed rules:

A risk-based approach

The main idea is to regulate AI based on the latter’s capacity to cause harm to society following a ‘risk-based’ approach: the higher the risk, the stricter the rules, the EU said in a statement.

The proposed law defines four levels of risk in AI: Unacceptable risk, High risk, Limited risk, and Minimal or no risk.

Unacceptable risk: All AI systems considered a clear threat to people's safety, livelihoods and rights will be banned.

The provisional agreement prohibits cognitive behavioural manipulation, indiscriminate harvesting of facial images from the internet or CCTV footage, emotion recognition in workplaces and educational institutions, social scoring, biometric categorisation to infer sensitive data like sexual orientation or religious beliefs, and certain cases of predictive policing for individuals.

High risk: These systems are deemed to have significant potential to harm health, safety, fundamental rights, democracy, elections, and the rule of law.

These systems will be allowed but will be subject to a set of requirements and obligations to gain access to the EU market.

Limited risk: AI systems classified as "limited risk" require specific transparency measures.

In applications like chatbots, users should be aware that they are interacting with a machine so they can make an informed decision to continue or opt-out.

Minimal or no risk: The proposal allows for the free use of minimal-risk AI, such as AI-enabled video games and spam filters. These types of AI pose little to no threat to the rights or safety of citizens, so they are exempt from regulations and can be used without restrictions.

Also read: AI Global Regulations: What’s needed?

Law enforcement exceptions

Remote biometric identification is considered high risk and subject to strict requirements.

It is prohibited in publicly accessible spaces for law enforcement purposes, except in cases, such as searching for a missing child, preventing a terrorist threat, or identifying a serious criminal.

General-purpose AI systems and foundation models

New provisions have been introduced to account for scenarios involving the use of AI systems for diverse purposes, particularly in the case of general-purpose AI (GPAI) technology being integrated into high-risk systems. The provisional agreement also addresses specific considerations for GPAI systems.

Distinct regulations have also been established for foundation models—large systems capable of proficiently executing a broad spectrum of tasks, such as generating text, images, and videos, engaging in natural language conversations, performing computations, or creating computer code. The provisional agreement mandates that foundation models must adhere to transparency requirements before entering the market.

Penalties for non-compliance

The AI Act will impose fines on non-compliant companies based on a percentage of their global annual turnover or a predetermined amount, whichever is higher.

These fines range from 7.5 million euros for supplying incorrect information to 35 million euros for violating the Act's prohibitions on specific AI applications.

The EU will also establish the EU AI Office, tasked with monitoring and fining violators of the law.

Event alert: CNBC TV18 & Moneycontrol will be hosting the most definitive summit on Artificial Intelligence with top global and Indian voices on December 16th in Bengaluru. Click here to register and get access to the 'Palo Alto Networks presents CNBC-TV18 & Moneycontrol Global AI Conclave'.