AI has been the talk of the town for over a year now, with tools like ChatGPT, Google Bard, and Microsoft Copilot threatening to upheave much of digital society as we know it. With those great risks and the potential for deepfakes, automatically generated content, and more, the European Union feels that AI legislation is necessary to safeguard its development without harming users. This legislation was first put forward three years ago, where it aimed to set a global standard for the technology.

On February 2nd, 2024, EU member states endorsed legislation aimed at fostering the safe development of Artificial Intelligence (AI), with a focus on mitigating "unacceptable risks" associated with AI technologies. This legislation faced opposition, particularly from France, home to Mistral AI—a company emerging as a formidable rival to OpenAI. France's resistance is attributed to concerns over maintaining competitiveness in the AI sector. Similarly, Germany and Italy expressed reservations about the legislation, advocating instead for "voluntary codes of conduct" to enhance AI literacy among individuals involved in AI's development, operation, and application. This approach suggests a preference for more flexible guidelines over stringent regulations in certain areas.

All countries initially opposing the legislation have now consented, following negotiations and compromises. France, for instance, withdrew its objections after receiving assurances that the legislation would balance transparency with the protection of business secrets, benefiting companies like Mistral AI and Aleph Alpha in Germany. This adjustment allows these firms to compete effectively, such as Mistral AI's competition with OpenAI, without significant regulatory obstacles from European authorities.

Prohibited artificial intelligence uses in the EU

These are all of the things AI won't be allowed do in the EU

There are very few things that are outright prohibited by the European Union when it comes to AI, and they primarily pertain to large corporations and authorities, rather than individual, private users. The EU says that they're to ensure ethical use and protect individuals from harm. It bans AI systems that use subliminal techniques to alter behavior, exploit vulnerabilities of specific groups, and the misuse of real-time biometric identification in public spaces by law enforcement, except under strictly necessary conditions. Certain uses will also require court approval.

Article 5 of the Artificial Intelligence Act specifically prohibits:

  • Manipulative AI Systems: AI that employs subliminal manipulation causing harm or exploiting vulnerabilities of individuals.
  • Exploitative AI: AI targeting vulnerable groups, including children and disabled individuals, to materially distort behavior.
  • Real-time Biometric Identification: The use of AI for real-time biometric identification in public spaces by authorities, barring exceptions for specific, substantial threats.
  • Social Scoring: AI systems that allow governments or other entities to score individuals based on their behavior or personal traits.

Certain other types of AI uses that are deemed "high-risk" will need to be registered in an EU database. These include:

  • Management and operation of critical infrastructure
  • Education and vocational training
  • Employment, worker management and access to self-employment
  • Access to and enjoyment of essential private services and public services and benefits
  • Law enforcement
  • Migration, asylum and border control management
  • Assistance in legal interpretation and application of the law.

These will need to be assessed before being put on the market and will also need to be regularly assessed throughout their lifecycle.

Generative AI

ChatGPT, Microsoft Copilot, and Google Bard all fall under this

The European Union is applying a light touch to generative AI, requiring that it complies with transparency requirements. These include:

  • Disclosing that the content was generated by AI
  • Designing the model to prevent it from generating illegal content
  • Publishing summaries of copyrighted data used for training

A notable aspect of the regulation is the mandate for comprehensive evaluations of "advanced" AI systems, including GPT-4, and the obligation to report serious incidents to the European Commission. This is because "high-impact general-purpose AI models" can pose "systemic risk." There's nothing particularly prohibiting you from running a model locally on your PC, though.

What happens now?

The future of the AI Act in the European Union

The AI Act allows EU nations to shape its enforcement through about 20 secondary legislative acts by the Commission. The AI Office will oversee AI models, staffed by experts from member states. Key milestones include the European Parliament committees' adoption of the AI rulebook on February 13 and a tentative plenary vote on April 10-11. Following ministerial approval, the Act becomes effective 20 days post-official journal publication. Prohibitions activate in six months, AI model requirements in a year, other regulations in two years, with third-party AI system assessments deferred an additional year.

With AI under a microscope currently, it's hard to expect it not to be regulated. Regulations can help companies too, as it provides clear guidelines of "no-go" zones that they can avoid, though without global cooperation, can result in companies in certain regions advancing far further than competitors. With AGI likely being the final frontier of artificial intelligence too, it's unlikely that there will be significant regulation that can prevent its development. This is particularly true given that its actual capabilities, should it be possible, are unknown at this time.