Artificial Intelligence (AI) stands at the forefront of technological innovation, reshaping industries across the globe. As AI becomes more integrated into our daily lives, the line between innovation and privacy invasion blurs, underscoring the urgent need for comprehensive regulatory intervention that protects all stakeholders. Effective regulation ensures that technological advancements yield positive societal outcomes, incentivizing responsible behaviours and establishing safeguards. Crafting such an approach requires contextually balancing the benefits and risks of AI rather than assessing them in absolute terms. In response to these growing concerns, UNESCO has released a ‘Consultation Paper on AI Regulation: Emerging Approaches Across the World’, outlining various strategies for regulating AI. The document tackles issues such as governance, ethics, and the broader societal impacts of AI, with the primary goal of guiding the development and regulation of AI systems to ensure they remain ethical, safe, and responsible.

Addressing the Lack of Specificity and Need for Independent Oversight

The consultation paper makes a commendable effort to address governance, ethics, and the societal impact of AI. One notable critique is the lack of specificity. The document emphasizes broad principles such as ethics, transparency, and accountability and provides concrete steps to enforce these. For instance, despite its significance, the term “accountability” in AI is frequently used too broadly, referring to a wide range of principles, procedures, and metrics. This is brought about by the complex relationship between accountability and context, the underlying ambiguity of political processes, and the sociotechnical structure of AI. The consultation paper can benefit from more precise guidelines on accountability when AI systems cause harm or make erroneous decisions as liability which are crucial for enforcing regulations and ensuring that responsible parties can be held accountable, whether they are developers, operators, or users of AI systems are not adequately defined. To be accountable, relevant AI manufacturers/deployers must be able to assure stakeholders, including end users, that the AI systems they are developing or deploying are worthy of trust. Accountability will also require consequences and modalities to ensure all stakeholders comply with stipulated ethical standards. Though accountability is not a new concept or principle, such clarity in an AI context is needed, mainly when the regulation of AI is in scope.

Furthermore, the regulation can also benefit from clear criteria for evaluating AI systems. The current approach leaves developers and organizations without detailed guidelines to align their AI models with regulatory standards. This ambiguity may impede the implementation of the proposed regulations, as stakeholders could need help to convert these general principles into actionable practices.

The consultation paper places considerable responsibility on AI developers and companies to self-regulate their systems, ensuring adherence to ethical guidelines. However, it underestimates the necessity for strong external oversight. While the document assumes that companies will act in good faith, pursuing profit often conflicts with ethical responsibilities. Relying solely on self-regulation is risky and insufficient, particularly in high-stakes sectors like healthcare, law enforcement, and employment, where the societal impacts of AI errors can be profound.

A significant area for improvement in the proposed regulatory framework is the need for more emphasis on independent oversight, such as third-party audits. An effective regulatory regime should take advantage of external audits of AI systems and their societal impacts, especially in high-risk applications. These audits ensure compliance with ethical standards and transparency in developing and deploying AI technologies. Independent oversight is critical for mitigating the risks of bias, discrimination, and unjust outcomes expected in unchecked algorithms.

In addition, the framework risks being overly restrictive, potentially stifling innovation and hindering competition, particularly for smaller companies and startups. Ethical considerations are essential, but more regulation with clear guidelines can create a compliance burden that discourages innovation. The document must provide a more nuanced strategy for balancing ethical compliance with technological progress.

Global Collaboration

AI is a global technology with supply chains and deployment across borders, making international cooperation essential for effective regulation. However, though the consultation paper fails to offer concrete strategies for achieving global collaboration, one can imagine that the existing cooperation channels in specific organizations, such as UNESCO, might be helpful in these cases. While the consultation paper highlights the importance of aligning regulatory frameworks across varying nations, this cannot be overemphasized, as it ensures that diverse legal systems, cultural values, and economic priorities can coordinate effectively.

This gap is particularly evident in data privacy and cross-border AI operations, where global cooperation is crucial. AI development frequently involves data exchange across borders, and managing these flows in compliance with varying international privacy laws requires more comprehensive strategies. The document recognizes the importance of data privacy, but it fails to outline a concrete plan for addressing the challenges posed by differing privacy regulations worldwide.

Several bilateral, regional, and multilateral initiatives have emerged to promote cooperation on AI regulation in recent years. Notable examples include the Bletchley Declaration, the Global Partnership on AI, and the US/UK partnership on the Science of AI Safety. Despite these efforts, countries continue to adopt divergent approaches to AI regulation, influenced by their respective levels of technological development and resources.

For instance, the EU AI Act—which sets a benchmark for AI policy in the EU (with potential global impact)—has triggered the so-called “Brussels Effect,” wherein EU regulations influence other countries’ approaches to AI. However, this dynamic poses challenges for developing countries, which may feel pressured to adopt parts of the EU AI Act to attract investment, even if these regulations contradict their cultural and economic contexts.

The imposition of regulations designed for developed nations on developing countries risks stifling innovation, limiting local economic growth, and creating barriers to AI adoption. Many developing countries need more resources and technological infrastructure to comply with stringent regulations tailored for advanced economies. To foster international interoperability, a more flexible, context-sensitive approach must account for different regions’ unique challenges.

One potential solution is to use international technical standards for AI development, deployment, and use. Such standards provide a common framework to ensure that AI systems adhere to ethical and safety guidelines regardless of geographic origin. However, the broader reform agenda must also address the need for safeguards and guardrails, particularly in high-risk sectors such as healthcare and law enforcement, where the consequences of AI errors can be severe.

Technologically advanced nations have already adopted varied regulatory approaches, contributing to a global patchwork of AI regulations. For example, China has targeted specific AI techniques like generative AI and deepfakes, while the United States has focused on regulating applications such as facial recognition technology. In contrast, the UK has taken an innovation-focused, sector-led approach, relying on existing regulators to govern AI within their sectors rather than implementing new AI-specific laws.

These differing approaches reflect local cultural values and regulatory priorities. The EU AI Act is deeply rooted in protecting fundamental rights, such as privacy and dignity. At the same time, China’s generative AI regulations focus on ensuring AI-generated content aligns with its core values. Such disparities underscore the challenge of creating a universal regulatory framework for AI.

Given the rapid pace of AI development, static regulations risk becoming obsolete. The consultation paper may need to adequately address the need for rules that can be continuously reviewed and updated to keep pace with technological advancements. Mechanisms like regulatory sandboxes, where AI technologies can be tested in controlled environments, could help ensure that regulations evolve alongside the technology.

Protection for Vulnerable Populations

While the document acknowledges the risks AI poses to vulnerable groups, such as those in healthcare or low-income communities, it does not go far enough in specifying how these populations will be protected. There is insufficient focus on algorithmic bias and the potential for AI to exacerbate existing inequalities. Without proper oversight, AI systems could lead to discriminatory outcomes, particularly in healthcare, education, and employment areas where vulnerable populations are most at risk. The regulatory framework should incorporate more robust recommendations for protecting these groups.

Why, when, and how to regulate

The conditions for determining why, when, and how to regulate AI differ between the global North and global South countries for varying reasons, including availability of expertise and funding. While the propositions of the consultation paper are commendable, additional nuanced considerations are necessary. For instance, Step three of Figure 5 of the consultation paper considers the reliance on “other policy tools that are more effective, efficient or more equitable than regulating.” This presupposes that countries have existing structures that can support the regulation of AI or even regulate it temporarily. This might not be the case, especially in developing countries, typically affected by a lack of adequate regulatory structure(s).

Furthermore, countries with weaker laws will likely experience lower AI governance standards, which will likely see AI developers/deployers carry out in these countries, actions that they will typically not carry out in countries with stricter laws. This dynamic, therefore, means that countries, likely developing countries, that do not have the infrastructure, whether in terms of economic power or personnel, should be supported by creating some fund and/or support structure for them. Failure to provide such a structure might result in the slower development of AI regulation and governance in countries with weaker regulatory regimes (likely developing countries). The implication of this possible outcome is that it will hinder global AI development with broader implications for attaining sustainable development goals.

In conclusion, the consultation paper provides an important starting point for AI regulation, but it must evolve to address the complexities of global AI governance. It requires greater specificity, adaptability, context-sensitivity, and robust oversight mechanisms to meet the challenges posed by AI’s rapid development. With these improvements, the framework can protect stakeholders, promote fairness, and foster innovation, creating a sustainable and ethical AI landscape for the future.