The AI Act is a European law proposal on artificial intelligence (AI) presented by the EU Commission on April 21, 2021.
On May 11, the internal Market Committee and the Civil Liberties Committee has approved the draft and, if it is approved also by the Parliament and the EU Council, it will become the world first law on AI.
Goals of the AI Act and risk-based approach
The text of the law proposal aims at ensuring that AI systems are safe, transparent, traceable, indiscriminatory, environmentally friendly and, most of all, kept under human control.
The proposed rules abide by a risk-based approach and provide a series of obligations and bans depending on the level of risk that the AI could generate.
AI systems presenting an unacceptable level of risk for people safety are strictly forbidden, including systems using subliminal or manipulative techniques, exploiting people vulnerabilities or used for social scoring (that is to classify people based on their behavior, socio-economic status and personal characteristics).
In the most recent version of the text, the Euro-deputies have broadened the list of activities considered to be high-risk, forbidding further uses of the AI systems which could be discriminatory or intrusive.
In particular, the following are prohibited:
- “Real-time” remote biometric identification systems in publicly accessible spaces;
- “A posteriori” remote biometric identification systems (with the only exception of law enforcement for the prosecution of serious crimes and only after judicial authorization)
- Biometric categorization systems using data considered sensitive (such as gender, ethnicity, religion or political orientation);
- Predictive policing systems (like the one based on past criminal conduct);
- Emotion recognition systems in the police force, border management, workplace, and educational institutions
The indiscriminate collection of biometric data from social media or CCTV footage to create facial recognition databases has also been prohibited.
For what concerns generative AI systems, such as ChatGPT, the proposed law provides for more transparency obligations, like the one to reveal that the content was generated by an AI. It is imposed to design a generative AI model to prevent it from generating illegal contents and infringing the copyright on the material used for its training.
Next steps in the AI ACT approval process
The AI Act will become law only when the Council and the European Parliament will approve a single version of the text.
Before starting the negotiations with the Council on the final version of the AI Act, the law proposal needs to be approved by the entire Parliament, the vote is expected during the 12-15 June session.
Ilaria Feriti