Source: https://www.pexels.com/photo/elegant-artistic-globe-with-warm-tones-28667827/ With artificial intelligence (AI) becoming increasingly sophisticated, policymakers are forced to consider regulation and restriction of such technology, which is utilized in life-critical domains such as the healthcare, finance, and infrastructure sectors, as well as in software development. However, regulatory activity could limit the companies interested in investing in and operating such technologies. The EU and the US approach the development and utilization of AI in notably different ways, which in turn affects businesses and enterprises that operate within these economic blocs and the open-source AI community, the members of which collaborate and share developments for the collective good. The EU is concentrated on promoting risk-based regulation of AI, which entails additional requirements for businesses that operate it. Under the EU AI Act, certain activities are outright prohibited, or specific requirements are imposed on systems in a risk category. In particular, systems in the high-risk category include certain applications that pertain to employment, biometric identification, critical infrastructure, or creditworthiness, although certain biometric practices are prohibited outright rather than classified as high-risk. Systems in the high-risk category, depending on the type and purpose of the application, may entail additional documentation, risk management, and governance requirements for the organizations utilizing them. In addition, the Act stipulates transparency requirements for certain systems and specific regulation for broader models, which could include additional requirements, depending on their category. The Act could also apply to companies that are not based in the EU if their activities or the utilization of their technologies fall under its purview. As such, the organizations would have to consider the Act requirements for novel projects, which would create notable differences for companies that operate in similar fields but are not based in the EU. On the other hand, the US approach to regulating AI is to rely on existing laws, specific industry regulations, and voluntary adherence to guidelines, rather than passing a comprehensive federal law similar to the EU AI Act, while allowing market forces to influence the development of novel technologies. According to the Brookings Institution, the current approach represents a mosaic of relevant laws, which cover discrimination, deceptive trade practices, and market manipulation. While the US approach does not cover all AI businesses and operations with a comprehensive set of federal requirements, existing laws could apply to the products and services that utilize AI. As noted in the article, the policy choice allows businesses to innovate, but the environment for compliance is notably complex. As such, organizations that violate existing laws could be subjected to enforcement actions or civil liability, depending on the relevant laws and the circumstances of their AI use. The differences in the approaches to AI regulation on the part of the EU and the US affect open-source AI and its participants in particular ways. The latter term refers to developers who share their code and, in some cases, model weights and other information with others, who can analyze, edit, and utilize these resources to further develop their own projects. Within the EU, organizations that participate in the creation of qualifying free and open-source AI components can utilize certain exemptions under the AI Act, but only to a limited extent. The requirements that apply to such components depend on the category of AI model or system, the purpose of its utilization, and the role of the organization that provides it. As noted earlier, general-purpose AI models that present systemic risk face additional obligations, while systems in the high-risk category are subject to additional requirements depending on how they are utilized. As such, when considering the adoption of such a model, a company should consider which obligations apply to it based on the model's documentation, support, distribution, and intended use. These obligations can vary depending on how the organization uses or distributes the model. In addition, it should consider the potential legal responsibilities of the original model developer. Such complexities are particularly acute for models that are systematically adapted by multiple organizations before being applied in a high-risk field. The differences between the EU and the US approaches to AI regulation affect the distribution and development of such models, which can create significant compliance challenges for their participants. The differences between the approaches to regulating AI on the part of the EU and the US create distinct opportunities and liabilities for the open-source AI community. In particular, companies that utilize such models can save substantially on research and development by implementing such solutions. However, an organization that utilizes an open-source AI component in a system covered by the EU AI Act may have to fulfill the legal requirements that apply to its particular role and use of the system. In addition, the obligations of the original developers depend on the requirements that apply to them; making model weights and documentation available does not automatically fulfill all obligations that may apply to the original developers or to organizations that subsequently use the model. The responsibilities of each organization depend on its role, the model or system involved, and how the technology is used or adapted. Therefore, even if a model was originally designed for general use rather than a high-risk application, organizations that adapt or deploy it may still need to meet the requirements applicable to their particular use. In addition, small-scale businesses or organizations may lack the resources to implement solutions that fulfill such obligations due to their limited compliance capabilities. The EU's more comprehensive approach may create substantial compliance challenges for enterprises, while the US approach may offer greater flexibility through its reliance on existing laws and the absence of a comparable comprehensive federal AI act. However, US businesses can still face legal liabilities depending on how they use AI components. Such complexities require open-source AI component developers to consider their options for distribution, liability, and documentation, as well as inform the organizations using their models of relevant limitations and responsibilities. The regulation of AI on the part of the EU and the US will have lasting implications for such technologies for years to come. The EU approach centers on the implementation of risk-based requirements for AI systems and general-purpose AI models, which can create additional costs for organizations that develop or utilize such technologies, including those that are not based in the EU but seek to sell their products in the European market. The US approach may be more flexible in certain circumstances but can still create legal liabilities depending on the use of a particular AI component. Open-source AI projects are particularly affected by the differences between the EU and the US approaches to AI regulation because developers may have limited control over how other companies modify and use their projects. However, open-source projects are essential to supporting rapid development and innovation; therefore, regulators should aim to assign responsibilities proportionately to the risks involved and the roles of the organizations using or developing such technologies, while allowing them to operate within the boundaries of applicable laws. The ways in which the EU and US balance these considerations in the future will help shape the development of such projects for years to come.
Sources
https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-53 https://ai-act-service-desk.ec.europa.eu/en/ai-act/article-2 https://www.brookings.edu/articles/regulating-general-purpose-ai-areas-of-convergence-and-div ergence-across-the-eu-and-the-us/