Breaking down the differences between the EU's AI Act Proposal and US President Joe Biden's Executive Order on Artificial Intelligence
A short comparative analysis of the two approaches to regulating Artificial Intelligence
The European Union's Proposed AI Act and President Joe Biden's Executive Order on artificial intelligence both aim to regulate the use of AI. Still, they differ in their approaches, contexts, and focus areas. The text of the entire executive order is here.
Below is a comparison between the two:
1. Context and Objectives:
Biden's Executive Order: The order focuses on ensuring the US leads in seizing the promise of AI while managing its risks. It emphasises safety, equity, civil rights, and American leadership domestically and internationally.
EU AI Act: The EU's proposal is designed as a regulatory framework for AI in the European Single Market. It categorizes AI systems based on risk and aims to ensure the development and use of AI respect EU values and fundamental rights.
2. Scope & Flexibility
Scope: Biden's Executive Order is broader, touching various sectors from healthcare to education and national security. It sets foundational directions, allowing for flexibility in implementation. The EU’s AI Act is more prescriptive, with clearly defined rules, especially for high-risk AI systems.
Biden Executive Order on AI:
Breadth: The Executive Order is comprehensive and addresses various sectors, from national security to healthcare to education.
National Emphasis: Its primary focus is on the US and its leadership in AI both domestically and internationally. While it does reference international collaboration, the priority is on ensuring US domestic interests are upheld.
Governmental Involvement: The Executive Order places significant emphasis on the role of federal agencies in AI governance, oversight, and advancement.
EU's AI Act Proposal:
Breadth: The AI Act also offers a broad overview of AI regulations but is more focused on categorising AI systems based on the risks they present and then regulating them accordingly.
Regional Emphasis: Given the nature of the EU, the AI Act is designed to provide a harmonised framework for all member states. This ensures a level playing field across countries in the EU.
Stakeholder Involvement: The EU's proposal includes a more detailed framework for involving various stakeholders, from developers to deployers, in the AI lifecycle.
Flexibility:
Biden Executive Order on AI:
Adaptability: The Executive Order’s directives, while comprehensive, still allow for considerable flexibility in terms of implementation. For instance, it often calls upon specific departments or agencies to develop further guidelines or best practices.
Future-proofing: The Order acknowledges the need for ongoing revisions and the development of future frameworks. For example, it calls for the development of a National Security Memorandum and the monitoring of emerging threats.
Collaboration: The Executive Order frequently references the need for departments to collaborate, indicating an understanding of AI’s cross-sectoral implications.
EU's AI Act Proposal:
Tiered Approach: The EU’s proposal classifies AI applications into different risk categories: Unacceptable, High, Limited, and Minimal risks. This allows for flexibility because not all AI systems are treated the same; the regulations adapt to the level of risk.
Regulatory Sandboxes: The proposal encourages the establishment of "regulatory sandboxes", which would allow for experimentation in a controlled environment. This is a recognition of the fast-paced evolution of AI and the need for regulations to adapt accordingly.
Review Mechanism: The proposal includes provisions for periodic reviews, ensuring that as technology advances, the regulations can be updated.
While the Biden Executive Order and the EU’s AI Act Proposal are comprehensive in their approach to AI, they differ in their primary focus areas and mechanisms for flexibility. The US approach under Biden is more centred on national interests with a broad mandate for federal agencies. In contrast, the EU’s approach is harmonised across member states with a tiered system for risk management. Both, however, demonstrate an understanding of the rapidly evolving nature of AI and the need for adaptable frameworks.
2. AI Safety and Security:
Biden's Executive Order: The order is expansive on AI safety, with requirements for developers of robust AI systems to share safety tests with the US government, development of standards for AI safety and security, protection against the use of AI in creating biological threats, cybersecurity initiatives, and protection against AI-enabled fraud.
EU AI Act: The act categorises AI applications based on their risk to fundamental rights, with prohibitions on specific high-risk uses, strict regulations for high-risk applications, and lighter regulations for lower-risk applications.
3. Privacy:
Privacy is a crucial aspect of AI governance, especially given AI's capabilities in data processing, pattern recognition, and decision-making. I’ll delve into the privacy implications and strategies presented in the Biden Executive Order on AI and the EU's AI Act Proposal.
Biden Executive Order on AI:
Key Directives & Implications on Privacy:
Data Extraction & Exploitation Concerns: The Executive Order acknowledges that AI can intensify the extraction, identification, and exploitation of personal data. This recognition is foundational in the measures that follow.
Call for Legislation: President Biden calls on Congress to enact bipartisan data privacy legislation. This suggests a move towards a more unified, perhaps comprehensive, national data privacy framework.
Privacy-Preserving Techniques: The Order emphasises accelerating the development and use of techniques that let AI systems be trained while preserving the privacy of the training data. This could refer to methods such as differential privacy or federated learning, which aim to utilize data without exposing individual data points.
Guidelines for Privacy-Preserving Techniques: The Executive Order seeks to establish guidelines for federal agencies to assess the efficacy of these privacy-preserving techniques, ensuring that agencies have a clear understanding of how to protect data effectively.
Commercial Data Collection: There's a specific emphasis on evaluating how agencies gather and use commercially available data, especially from data brokers. This underscores concerns about third-party data markets and their potential risks to individual privacy.
2. EU's AI Act Proposal:
Key Directives & Implications on Privacy:
Data Governance: The EU's AI Act Proposal is built on solid data protection rules, especially given the pre-existing General Data Protection Regulation (GDPR). The GDPR already sets rigorous requirements for data consent, access, and erasure.
High-Risk AI Systems: For AI systems considered "high-risk," there are requirements for the quality of datasets, ensuring that they are free from biases which can lead to discrimination. While this is more about fairness, it does touch on data handling and processing, implicitly highlighting the importance of data privacy.
Transparency Requirements: The proposal emphasises that citizens should be informed when interacting with an AI system, which indirectly impacts privacy. Knowing when and how AI processes one's data can empower users to make informed decisions about their data privacy.
Biometric Identification: The AI Act is particularly strict about using remote biometric identification (like facial recognition) in publicly accessible spaces. Such systems are considered high-risk and are subject to stringent requirements, reflecting concerns over both privacy and civil liberties.
Comparative Analysis:
Common Ground: Both the Biden Executive Order and the EU’s AI Act Proposal recognise the intrinsic risks AI poses to personal privacy, especially regarding data collection, processing, and potential misuse.
Legislative Approach: While the Biden Executive Order calls for new legislation, the EU already has a robust framework in the form of the GDPR. The AI Act further builds upon and complements these existing regulations.
Emphasis on Techniques: The US approach places significant emphasis on the development and adoption of privacy-preserving techniques. In contrast, the EU's approach is more regulatory, setting clear boundaries and criteria for AI operations.
Privacy is a paramount concern in both frameworks. The US approach, as laid out in the Biden Executive Order, leans towards promoting technological solutions and overarching legislation. At the same time, the EU's stance is more regulatory, grounded in pre-existing comprehensive data protection laws, and focuses on categorizing and managing risks associated with different AI applications.
4. Equity and Civil Rights:
Equity and civil rights are core considerations in the governance and deployment of AI systems, given AI's potential to amplify and mitigate systemic biases and discrimination. Let's examine the implications and strategies presented in the Biden Executive Order on AI and the EU's AI Act Proposal. Biden's Executive Order addresses potential discrimination and bias arising from AI use in sectors like justice, housing, and healthcare. It guides discriminatory use and promotes fairness throughout the criminal justice system. The EU’s AI Act emphasises the protection of fundamental rights. High-risk AI systems, especially, are subject to requirements ensuring they do not discriminate or violate individual rights.
Biden Executive Order on AI Key Directives & Implications on Equity and Civil Rights:
Addressing Discriminatory Outcomes: The Executive Order explicitly mentions the need to address and mitigate AI-driven outcomes that could reinforce biases and lead to discriminatory decisions. This acknowledges that AI systems can perpetuate or amplify pre-existing systemic biases if not designed and monitored carefully.
Inclusive Stakeholder Engagement: The Order calls for public engagement to ensure diverse perspectives, especially from communities historically marginalized or underrepresented, are taken into account when formulating AI policies.
Equitable Access to Benefits: The Order emphasizes that all Americans, regardless of demographic or socioeconomic factors, should benefit from AI's advancements. This underscores the importance of ensuring that AI-driven solutions are accessible and beneficial across diverse populations.
AI Impact on Employment: Recognizing the potential of AI to reshape the job market, the Order stresses understanding and addressing AI's impact on employment, especially for communities and populations vulnerable to displacement.
EU's AI Act Proposal Key Directives & Implications on Equity and Civil Rights:
Fundamental Rights: The AI Act Proposal is anchored in upholding fundamental European rights, including non-discrimination, personal and data privacy, and the protection of personal data.
High-Risk AI Systems: AI systems deemed "high-risk" must comply with strict requirements. This includes ensuring the quality of datasets, which should be free from biases that could lead to unfair or discriminatory outcomes.
Transparency and Redress: The AI Act emphasises the right of individuals to know when they are interacting with an AI system and to challenge decisions made by such systems. This promotes transparency and provides mechanisms for redress in cases of potential discrimination or bias.
Prohibitions: Certain AI practices, which could violate user rights and lead to unjust outcomes, are deemed unacceptable and are prohibited outright. This reflects a hard stance on applications that could be harmful or discriminatory.
Comparative Analysis:
Shared Emphasis on Non-discrimination: Both frameworks stress the importance of addressing potential biases in AI systems and ensuring that these systems don't perpetuate or amplify discriminatory patterns.
Engagement vs. Regulation: The U.S. approach, as per the Biden Executive Order, emphasizes stakeholder engagement, especially from marginalized communities, to shape AI policy. In contrast, the EU's approach is more prescriptive, setting clear boundaries and requirements for AI operations, especially for high-risk systems.
Economic Impacts: The U.S. Order highlights explicitly the potential economic impacts of AI, focusing on employment and the need for equitable distribution of AI's benefits. The EU's Act, while not explicitly emphasising economic outcomes, lays a foundation that ensures AI applications align with fundamental rights.
The US and EU recognise the profound implications of AI on equity and civil rights. The US places a strong emphasis on inclusive policy formation and understanding economic ramifications. At the same time, the EU's approach is grounded in upholding fundamental rights with clear regulations, especially for high-risk AI systems. Both approaches underscore the need to mitigate biases and ensure AI systems serve the greater good without inadvertently perpetuating discrimination.
5. Innovation and Competition:
Biden's Executive Order promotes innovation and competition by supporting AI research, assisting small developers, promoting a competitive AI ecosystem, and streamlining visa processes for AI experts.
EU AI Act: The act focuses on establishing a balance between regulation and innovation. While strict rules exist for high-risk applications, there is a clear intention not to over-regulate and stifle innovation.
6. International Cooperation:
Biden's Executive Order: It emphasizes working with global allies and partners on AI governance frameworks and advancing American leadership in the international AI space.
EU AI Act: The act has a more regional focus, primarily intended for the European Single Market. However, the EU's influence means its standards often have global repercussions.
7. Government Use of AI:
Biden's Executive Order: The order guides the federal use of AI, from procurement to deployment, and emphasises the need for AI training across government agencies.
EU AI Act: While it does not explicitly focus on government use of AI, the act's regulations apply to all AI applications, including those deployed by governments within the EU.
While the Biden Executive Order and the EU AI Act aim to address the challenges posed by AI, they come from different contexts and have different areas of emphasis. The US approach leans toward safeguarding national interests, promoting innovation, and global leadership. In contrast, the EU's approach is more regulatory, aiming to balance innovation and protect fundamental rights.
