Regulating Artificial Intelligence: Challenges and Perspectives | Hemant Batra

Growing Importance of AI Regulation

The rapid advancements in artificial intelligence (AI) have led to an increased need for establishing appropriate regulation. The use of AI has resulted in a range of ethical, legal, and social dilemmas, including new challenges in several areas such as data privacy and protection, employment, economic competition, healthcare, intellectual property, and security. At the national and supranational levels, governments are seeking to establish a framework of laws and soft laws to regulate the use of AI in contexts ranging from misinformation and disinformation to the use of AI in warfare and economic competition. Organizations now face an increasing legal responsibility to demonstrate that their AI is used in a way that respects the public's rights and best interests. Furthermore, they need to balance often conflicting needs and expectations towards ensuring fairness, safety, transparency, and privacy.

The lack of appropriate regulation of AI has several social and economic implications. Autonomous AI agents and systems are capable of making decisions with potential lethality. Left unregulated, they may reproduce the human biases and discrimination that are contained in training datasets. In such a scenario, AI agents would become a source of increasing disempowerment of already marginalized communities by denying fair access to areas such as health systems, education, and employment both in the present and in the future. These developments have led to an international movement toward the development of ethical principles that should guide the governance and regulation of AI. A group of experts in computer science, stakeholders from government, industry, and privacy advocates have been working on standardizing, regulating, and certifying the adherence to ethics in the development and use of AI. Regulating AI at the source is an emotional and wise reaction to societal harm.

Current State of AI Regulation Worldwide

The European Union proposed the AI Act, which offers the world’s first comprehensive framework for AI, based on intrinsic European values. Although profoundly innovation-focused, as the Act establishes a system of approvals and post-market audits to allow marketing throughout the continent, the proposal sets important limits on the use of AI for surveillance and swarms, which do not align with special thresholds regarding fundamental rights. The debate on how to use global competition law as an advantage has also engaged academics and lawmakers in the United States. The Statement on Use of Enforcement Authority Regarding Competition in Technology Markets emphasizes the agency’s commitment to enforcing protections to ensure the innovation and growth of startups. A series of national initiatives build a backdrop of robust AI regulation.

As mentioned, the cautionary dimension of AI has also been captivating the attention of legislators. Here, the regulation comes to prevent “worst-case scenarios” (with predictions including mass supervision by a civil or military dictator, mass unemployment because of automating all jobs, and the initiation of an atomic or biological arms race). In Europe and beyond, national laws and regulations on autonomous systems are enacted primarily for military objectives, and they do embed detailed technical requirements, which are analogous to current limitations within civilian settings. For example, the UK Defence and Security Accelerator aims to regulate swarming. Commercial AI regulation is at an early stage everywhere and is often qualitative. The manufacture and operation of autonomous vehicles and drones, in many jurisdictions, is the area most regulated. While the U.S. and the UK started at the federal level, and Israel at the national one, others have opened the regulatory debate at a sub-national level, like Australia, Canada, Germany, Italy, and India, while working on a legal framework for AI and robotics. The European Union does present the most advanced regional proposal.

Challenges in Regulating AI Technologies

Artificial Intelligence (AI) and its underlying technologies are rapidly diffusing into every aspect of contemporary economies, society, and politics. This technological shift is frequently associated with public and policy concerns, especially regarding regulatory insufficiency, given that current legal frameworks are not specifically aimed at AI in all its forms. The first relevant issue is that it is extremely difficult to provide a comprehensive definition of AI to guide regulatory intervention. A second problem lies in the urgency of quick regulatory responses that contrast with the time-consuming characteristics of the legislative process. Accountability and liability concerns are strongly debated in the literature on AI technologies and are also explicitly addressed by the AI Act.

One relevant source of opacity that affects AI systems is the lack of ability to understand their decision-making processes in contrast with older technologies. Relatedly, transparency is an important value that guides the regulation of AI. Regulating AI has the potential to prompt a societal debate on how to make algorithms more transparent and regulate public interests in this field. A fourth challenge is that AI systems have been shown to be affected by different forms of bias. A fifth fundamental challenge to regulatory intervention is the strong financial interests driving the ongoing process of development and deployment of AI technologies. Large corporations that develop AI technologies resist stringent regulation for fear of reducing profits. The interoperability of AI systems across companies and the harmonization of standards are both preconditions to promote the EU as a hub for innovation. AI operates across the globe. Therefore, regulatory interventions should be global. However, regulation aimed at global AI that involves all significant stakeholders and operators is also subject to opposing pressures. Finally, a deep collaboration between, primarily, technologists, ethicists, and legislators is necessary. This is because these thematic areas can be seen as dependent on each other, rather than completely separate and logically distinct categories of interest.

Emerging Frameworks and Approaches for AI Regulation

Equally innovative are the regulatory approaches and algorithms being proposed and implemented around the world. Concepts and models such as benchmarks of good practice, soft law, or so-called sandboxes await study and more general adoption. The fundamental importance of adaptive regulation, when it comes to an unpredictable and quickly evolving ecosystem of smart machines, cannot be stressed enough. At the same time, enforceability remains an open question, with suggestions that compliance audit systems model those implemented in the financial sector, which is noteworthy. Given this pioneering and, consequently, rapidly expanding legal literature, the purpose of this write-up is not to reiterate the models. Rather, the ambition is to situate the discussion in the ongoing debates on AI governance and law-making, through, in particular, sociological, political science, and policy studies literature. Instead of starting this discussion with the state of the law and emerging avenues for future research, this write-up takes a step back to introduce key theoretical considerations around the development of new regulatory frameworks and the distinctive role of AI ethics guidelines in regulatory processes.

Ethical principles for AI systems have been suggested as a starting point and foundation for international regulation. These principles have been developed at the invitation of policymakers, businesses, and researchers. A significant effort has similarly been made to produce guidelines for the development and stewardship of AI. Digital borders and megatrends present and analyze the emerging legal and policy frameworks in the area of AI governance. The diversity of ongoing rule formation processes around the world demonstrates that at the heart of today’s rules and enforcement are new regulatory models or models that are being given effect for the first time at scale. However, today’s 'compliance with what?' remains largely, still, an answer seeking a question. And in order to foster debate on questions, what do we know about the emerging regulatory processes? What are some of the risks or regulatory designs that we know about from ongoing work? And, finally, what do we not know and where might research be directed as a starting point for further work in this area?

Future Directions and Recommendations for Effective AI Regulation

One of the biggest challenges hindering AI regulations is the need to be proactive rather than reactive in the development of such laws. It is important to preempt and focus on prospective AI developments and trends, as well as their far-reaching risks and impacts on our world, international trade, and public growth. Governments and international bodies must collaborate with one another and the private sector to disrupt unwanted developments in the field of AI. This will also lead to improved adherence to health and safety norms and other regulatory requirements. It is crucial for international bodies to work together and establish a cohesive international AI regulatory framework to ensure that AI regulations are easily understood and highly effective across nations. Here are a few future directions and recommendations for regulating the field of AI.

It is essential to establish a clear privacy and data governance framework to ensure that data application through AI is as ethical as possible. The policymaking process for AI must be inclusive and engaging. Policymaking in AI should not embody an elitist view of AI capabilities and how they should be regulated. The process for creating AI regulations must involve various stakeholders, from academia to large consortiums, small startups, local players in the industry, and policymakers, as well as healthcare professionals and public sector employees. It is important to fund research and support public education to make the intricacies of AI understandable to diverse audiences. New studies on AI should be continually assessed as part of a reliable foundation for new regulations. Ensuring accountability and ethics in AI should be the cornerstone of AI regulations.

*Hemant Batra is a lawyer, published author and tv host