How the EU is trying to create a common legal framework for artificial intelligence that respects human rights and fosters trust and competitiveness
Artificial intelligence (AI) is transforming the world in unprecedented ways. From healthcare to education, from entertainment to security, AI is enhancing our lives and solving our problems. But also poses many challenges and risks, such as data privacy, bias, fairness, accountability, transparency, quality, safety, and security. How can we ensure that AI is developed and used in a way that respects our values, rights and safety, while also supporting innovation and competitiveness?
This is the question that the European Union (EU) is trying to answer with its proposed regulation on AI, known as the EU AI Act.
The EU AI Act is a landmark regulation that aims to introduce a common legal framework for the development, marketing and use of AI in the EU. The main goal of the regulation is to ensure that AI systems respect the values, rights, and safety of the people and the environment, while also supporting innovation and competitiveness. The regulation also seeks to establish the EU as a global leader in AI innovation and regulation, by setting high standards and ethical principles for AI.
The EU AI Act is based on a risk-based classification of AI systems, which determines the level of regulatory oversight and requirements for each system. The regulation identifies four categories of risk: unacceptable, high, limited, and minimal. Unacceptable risk AI systems are those that violate fundamental rights or pose a clear threat to the safety or security of people, such as social scoring or mass surveillance. These systems are banned or prohibited in the EU. High risk AI systems are those that have a significant impact on the life, health, safety, or rights of people or the environment, such as biometric identification, self-driving cars, medical devices, and High-risk or recruitment tools. These systems are subject to strict obligations, such as data quality, transparency, human oversight and conformity assessment. Limited-risk AI systems are those that pose some risks to the rights or expectations of people, such as chatbots, recommender systems, or deepfakes. These systems are subject to transparency and information requirements, such as disclosing the use of AI or the source and nature of the generated content. Minimal-risk AI systems are those that pose no or negligible risks to the people or the environment, such as video games, spam filters, or email assistants. These systems are subject to no specific obligations but are encouraged to follow voluntary codes of conduct and best practices.
One of the most contentious issues in the EU AI Act is the regulation of foundation models, which are powerful AI models that can generate realistic content for various tasks and domains. Foundation models have been widely used and developed by big tech companies, such as Google, Facebook, and Amazon, as well as research institutions, such as OpenAI and DeepMind. Foundation models have also raised many ethical, social, and legal concerns, such as data privacy, bias, fairness, accountability, transparency, quality, safety, and security.
The EU AI Act initially proposed to classify foundation models as high-risk AI systems, subject to the same obligations as other high-risk systems, such as data quality, transparency, human oversight, and conformity assessment. However, this proposal faced strong opposition from the industry, the research community, and some member states, who argued that foundation models are essential for innovation and competitiveness and that the regulation would stifle their development and use in the EU.
They also claimed that foundation models are not inherently risky, but rather depend on the data, the task, and the context in which they are used. They advocated for a more flexible and nuanced approach, based on the actual impact and harm of each application of foundation models, rather than a blanket regulation.
The EU AI Act is currently under negotiation among the EU institutions, namely the Commission, the Parliament, and the Council. The Spanish presidency of the Council, which represents the member states, has made a last mediation attempt on the issue of foundation models, proposing a compromise solution that would allow some exceptions for the use of foundation models under certain conditions.
These conditions include the use of foundation models for research and innovation purposes; the use of foundation models for public interest or common good purposes; the use of foundation models that have been certified by an independent body; and the use of foundation models that comply with specific technical and ethical standards. The proposal also requires that the use of foundation models is transparent, traceable, auditable, and subject to human oversight and intervention. The proposal has received mixed reactions from the different stakeholders, with some welcoming it as a balanced and pragmatic approach, and others criticizing it as a loophole or a concession to the industry.
Let's Wrap it Up
The EU AI Act is a groundbreaking regulation that could significantly impact the future of AI in the EU and beyond. The regulation aims to balance AI's benefits and risks while fostering trust and innovation. The regulation also faces many challenges and controversies, especially regarding the regulation of foundation models, which are at the forefront of AI research and development. The outcome of the negotiation process will determine the final shape and scope of the EU AI Act and its implications for the AI ecosystem and society.
Thank
you for reading.
Best,
Nexa-Hub
