Another principle underpinning the legislation is a risk-based approach. The law defines four levels of risk.
High risk
These include AI technology used in activities and services that may have a significant impact on people’s safety or fundamental rights. These systems will have to meet stringent requirements before being placed on the market. They could fall into these categories: critical infrastructure (for example transport), which could endanger citizens’ lives and health, education or vocational training, AI applications in robot-assisted surgery or medicine, management of workers in the recruitment process, credit scoring, administration of justice, and so on.
Limited risk
These are AI systems that require specific transparency or information to be provided to users. The AI Act introduces specific obligations to ensure that users know, for example, that they are interacting with a machine, such as a chatbot. Another important matter: providers will have to ensure that AI-generated content is identifiable. Generated texts of public interest will have to be identified as artificially created by a machine, even if they appear to be false.
Minimal or no risk
This category covers AI systems that pose only minimal concern for safety. It includes applications such as AI-enabled video games or spam filters. Most AI applications fall into this category. Such systems can be used freely with few or no restrictions.
Unacceptable risk
This category covers AI applications considered too high a risk to be permitted. These systems are banned or severely restricted because of the potential harm they could cause. They could include systems that can manipulate human behaviour or exploit people’s vulnerabilities, and mass biometric surveillance in public spaces.
Law enforcement is permitted to use facial recognition software in public spaces, biometric data detection and the use of artificial intelligence aimed at enhancing these activities.
Who monitors
For monitoring purposes, the European AI Office was established in February by decision of 24 January 2024 C/2024/1459
The Office will carry out monitoring tasks to ensure compliance with the rules of the AI Act regulation. But that is not all: it will also monitor artificial intelligence strategies and developments worldwide.
By way of example, it will be able to investigate possible infringements of the rules on general-purpose AI models and systems, including by collecting complaints and reports. The Office performs numerous other functions, including at the political level.
Timing
The AI Act will enter into force 20 days after its publication in the Official Journal. It does not become operational immediately and will be fully applicable only after two years. But there are some exceptions:
The bans, for example, will apply after just six months, while the obligations for general-purpose AI models will apply after 12 months. The law provides for 36 months for AI systems — embedded — in regulated products.
With an ‘AI Pact’, the European Commission has sought to invite developers, leaving them entirely free not to do so, to comply with the core obligations of the AI Act ahead of time.
The environment and AI
In this regulation the environment does not go unnoticed. It refers to the opportunity to monitor and rationalise scarce resources, energy and parameters in order to streamline and optimise actions to reduce the harmful effects of climate change. But that is not all; the text speaks of “minimising the impact of AI systems on environmental sustainability, including as regards energy-efficient programming and techniques for the efficient design, training and use of AI”.
The impact of AI on the environment is not negligible. Data management processes and the running of AI algorithms and models require great computing power, and hence a high demand for energy. We must acknowledge that artificial intelligence is extremely energy-hungry.
Considerations
The EU has enacted the world’s first law regulating AI. The ambitious project rightly aims to address the dangers that are growing exponentially with the commercial development of this technology. At the same time, it is also necessary to begin a genuine effort to educate the population about AI.
The educational process is essential to gain greater awareness of this tool. Who has not cursed at their computer because it wasn’t working? Hands up. This human tendency to personify technology could risk becoming stronger if we have to deal with technologies that seem even sentient. Knowledge of artificial intelligence will make it clear to everyone that, despite everything, we are always dealing with machines running software, however advanced.
For a deeper look at the subject of AI, we refer you to the video interview “Artificial intelligence beyond the myths – Interview with Prof. Massimiliano Badino” published on GTM
