Parallel Narratives: How OpenAI Influences the EU AI Act

Introduction: Who is the author and why should you care?

Hello, I am Fred Wilson, a senior researcher at the Center for AI and Society. I have been studying the social and ethical implications of artificial intelligence for over a decade, and I am passionate about finding ways to ensure that AI serves the common good. In this article, I will explore the parallel narratives shaping AI regulation in Europe, focusing on the role of OpenAI as a leading AI research organization and a potential challenger to the EU’s vision of trustworthy AI. I will explain what OpenAI and the EU AI Act are, how they differ in their goals and methods, and what are the implications of their interaction for the future of AI innovation and governance. I will also offer some recommendations for academics, researchers, and policymakers to foster a constructive dialogue and collaboration between OpenAI and the EU on AI regulation. I hope you will find this article informative and insightful, and I invite you to share your thoughts and feedback with me.

What is OpenAI and what are its goals?

OpenAI is a research organization that aims to create artificial general intelligence (AGI) that can benefit all of humanity. AGI is a hypothetical form of AI that can perform any intellectual task that a human can, and potentially surpass human intelligence in all domains. OpenAI was founded in 2015 by a group of prominent entrepreneurs and visionaries, such as Elon Musk, Peter Thiel, and Reid Hoffman, who contributed $1 billion to the initiative. In 2019, OpenAI became a hybrid entity, consisting of a non-profit organization and a for-profit company, with Microsoft investing $1 billion in the latter. OpenAI’s mission is to ensure that AGI is aligned with human values and can be used for good, rather than evil. To achieve this, OpenAI conducts and publishes cutting-edge AI research, develops and releases open-source AI software and hardware, and advocates for the alignment of AI systems with human values and the mitigation of AI risks. Some of the notable AI models that OpenAI has developed and published include GPT-4, a natural language processing system that can generate coherent and diverse texts on any topic; DALL·E, a computer vision system that can create realistic images from text descriptions; and Codex, a programming system that can write and execute code from natural language commands.

What is the EU AI Act and what are its objectives?

The EU AI Act is a proposed regulation that aims to introduce a common legal framework for AI systems in the EU, based on a risk-based approach. The EU AI Act was proposed by the European Commission in April 2021, following the publication of the EU’s AI strategy in 2018 and the EU’s ethics guidelines for trustworthy AI in 2019. The EU AI Act’s main objective is to balance the promotion of AI innovation and the protection of AI users and society in the EU. To achieve this, the EU AI Act defines and prohibits certain AI systems that present unacceptable risks to human rights, such as biometric surveillance and generative AI systems like ChatGPT. The EU AI Act also imposes various requirements and obligations for high-risk AI systems, such as data quality, transparency, human oversight, and accountability. High-risk AI systems are those that are used in critical sectors, such as health, education, justice, and security, or that have a significant impact on people’s lives, such as credit scoring, recruitment, and social media platforms. The EU AI Act also establishes a governance structure for the implementation and enforcement of the regulation, involving national authorities, the European Commission, and the European AI Board. The EU AI Act is currently under discussion in the European Parliament and the Council of the EU, and is expected to enter into force by 2024.

How does OpenAI challenge the EU AI Act’s approach to AI regulation?

OpenAI’s work and vision contrast with the EU AI Act’s approach and principles in several ways. First, OpenAI’s goal of creating AGI that can benefit all of humanity is not aligned with the EU AI Act’s focus on the EU’s values and interests. While the EU AI Act aims to ensure that AI systems respect the EU’s fundamental rights and values, such as democracy, rule of law, and human dignity, OpenAI’s vision is more global and universal, seeking to create AI systems that can serve the common good of all people, regardless of their location, culture, or preferences. This raises the question of how to define and operationalize the common good, and whether it is possible to reconcile the EU’s values with those of other regions and stakeholders.

Second, OpenAI’s development and publication of cutting-edge AI models challenges the EU AI Act’s approach to AI regulation, which is based on a risk-based classification of AI systems. While the EU AI Act defines and prohibits certain AI systems that present unacceptable risks to human rights, such as biometric surveillance and generative AI systems like ChatGPT, OpenAI’s research and innovation agenda is not constrained by such limitations, and often pushes the boundaries of what is possible and permissible with AI. For instance, OpenAI’s GPT-4 and DALL·E models are examples of generative AI systems that can create realistic and diverse texts and images on any topic, which could potentially be used for malicious purposes, such as misinformation, manipulation, or impersonation. Moreover, OpenAI’s Codex model is an example of an AI system that can write and execute code from natural language commands, which could potentially pose a threat to the security and integrity of software systems and data. These examples illustrate how OpenAI’s AI models could fall under the category of prohibited or high-risk AI systems under the EU AI Act, and how OpenAI’s publication of these models could undermine the EU’s efforts to regulate them.

Third, OpenAI’s advocacy for the alignment of AI systems with human values and the mitigation of AI risks differs from the EU AI Act’s imposition of various requirements and obligations for high-risk AI systems. While the EU AI Act aims to ensure that high-risk AI systems comply with certain standards and principles, such as data quality, transparency, human oversight, and accountability, OpenAI’s approach is more proactive and collaborative, involving the participation and feedback of various stakeholders, such as researchers, developers, users, and policymakers. For instance, OpenAI has launched several initiatives to foster the alignment of AI systems with human values and the mitigation of AI risks, such as the Partnership on AI, a multi-stakeholder organization that promotes the responsible and ethical use of AI; the OpenAI Scholars Program, a grant program that supports underrepresented groups in AI research; and the OpenAI Safety Gym, a toolkit that helps researchers test and benchmark the safety and robustness of AI systems. These examples show how OpenAI’s advocacy for AI alignment and AI risk mitigation is more participatory and inclusive than the EU AI Act’s regulatory approach, which relies on the authority and enforcement of public institutions.

What are the implications of OpenAI’s impact on the EU AI Act’s pendulum?

OpenAI’s impact on the EU AI Act’s pendulum has both positive and negative implications for the future of AI innovation and governance. On the positive side, OpenAI’s work and vision could inspire and challenge the EU to rethink and improve its approach to AI regulation, by exposing the limitations and gaps of the EU AI Act, and by offering alternative perspectives and solutions. For instance, OpenAI’s goal of creating AGI that can benefit all of humanity could encourage the EU to adopt a more global and cooperative stance on AI regulation, by engaging and collaborating with other regions and stakeholders, such as the US, China, and the UN, to develop common standards and norms for AI. Similarly, OpenAI’s development and publication of cutting-edge AI models could motivate the EU to revise and update its risk-based classification of AI systems, by taking into account the latest advances and challenges in AI research and innovation, and by adopting a more dynamic and flexible approach to AI regulation. Likewise, OpenAI’s advocacy for the alignment of AI systems with human values and the mitigation of AI risks could influence the EU to enhance and diversify its governance structure for AI regulation, by involving and empowering more actors and voices, such as civil society, academia, and industry, to participate and contribute to the design and implementation of the EU AI Act.

On the negative side, OpenAI’s work and vision could also undermine and threaten the EU’s approach to AI regulation, by creating and exacerbating tensions and conflicts between OpenAI and the EU, and by jeopardizing the EU’s goals and interests. For example, OpenAI’s goal of creating AGI that can benefit all of humanity could clash with the EU’s focus on the EU’s values and interests, by creating a divergence and a competition between the EU’s and OpenAI’s visions and agendas for AI. This could result in a lack of trust and cooperation between OpenAI and the EU, and a loss of influence and legitimacy for the EU in the global AI landscape. Similarly, OpenAI’s development and publication of cutting-edge AI models could challenge and violate the EU’s approach to AI regulation, by developing and releasing AI systems that present unacceptable or high risks to human rights, and by circumventing or disregarding the EU’s rules and restrictions for