The EU AI Act introduces new rules for generative AI developers. It is the first major regulatory law on AI systems and its main objective is to protect the citizens from the potential harms associated with their use.
The EU AI Act include new regulations regarding copyrighted material and stipulates that any copyrighted content used to train the foundational AI models must be publicly disclosed.
The AI developers must obtain permission from copyright holders before using their material, and are also required to declare the usage of any copyrighted content in their application when submitting it for approval.
In addition to disclosing the copyrighted content, developers must ensure that these models are designed in compliance with EU laws and fundamental rights. The penalty for non-compliance is unclear but could entail retraining the model.
The EU AI Act was initiated in April 2021, as a means of governing the application of AI technology.
The clause concerning copyright was drafted in the last weeks with a primary focus on ChatGPT and generative AI, in response to recent announcements from major corporations such as Microsoft, Google, and Salesforce, regarding their extensive implementation of generative AI tools. In addition, Meta has confirmed its intentions to incorporate generative AI tools into several of its platforms, including Facebook, Instagram, and WhatsApp.
At present, various EU member states are considering the possibility of banning certain chatbots. The Italian data protection authority Garante has imposed a temporary ban on ChatGPT due to concerns about data protection breaches. Garante has asked OpenAI, the developers of ChatGPT, to provide more transparency regarding data processing, permission for personal data usage, and restricted access to minors. Similarly, Spain and France have expressed concerns about unregulated AI and its potential to cause discrimination, bias, and other harmful outcomes.
The EU experts reached a final draft agreement on the EU AI Act which is considered to be a compromise between ignoring the copyrighted content and banning its use in training AI models. The European Parliament’s Committee on Legal Affairs (JURI) will hold a vote on the EU AI Act on May 11, 2023. If approved, EU member states would need to agree to the legislation, and it is unlikely to take effect until early 2025.
EU vs US regulations
The EU and US have taken different approaches to regulating AI.
- The EU has a more comprehensive and prescriptive approach compared to the US more sector-specific and flexible approach.
- The EU places more emphasis on safeguarding fundamental rights and values, while the US prioritizes promoting innovation and competitiveness.
- The EU has a more centralized and coordinated governance structure, while the US has a more decentralized and fragmented governance structure.
How will this law affect the developers, consumers, and users?
The EU AI Act categorizes AI applications into four risk categories: unacceptable, high, limited, and minimal.
For instance, low-risk applications with no impact on individuals will receive minimal regulation, whereas high-risk areas such as biometric surveillance (such as cameras or fingerprint readers) or AI decision-making in health or legal issues will face intense scrutiny.
The act does not prohibit high-risk uses of AI, but requires higher levels of transparency regarding their operation and training.
If you are an AI system developer, you will have to comply with the rules that apply to our AI systems based on their level of risk. For instance, if you design AI systems that pose a high risk, you must ensure the quality of the data, transparency, human oversight, and accountability. Furthermore, you must register your AI systems before releasing them on the market.
If you are an AI system user, you must ensure that you use it according to their intended purpose. You also have to monitor their performance, and report any incidents or malfunctions to the authorities.
If you are an AI system consumer, you will have more rights and protection when interacting with them. You will have the right to be informed when you are exposed to AI-generated content and can refuse certain AI systems, such as targeted ads. Additionally, if you suffer any harm or damage caused by an AI system, you have the right to seek compensation and remediation.
Conclusion
EU AI Act represents an important and necessary effort to regulate the rapidly advancing field of AI, and its impact is likely to be felt not only within the European countries, but worldwide as well.
It’s worth noting that these regulations are still in the proposal stage and have yet to be adopted. They need to go through the approval process of the European Parliament and the Council before they can be implemented as laws.
Learn more:
- The EU AI Act can be found on the official website
- The World Economic Forum’s website (detailed explanation of the law)
