Needed: Intelligent act to regulate AI

By Sapni G K

The 41st General Conference of the United Nations Educational, Scientific and Cultural Organization (UNESCO) concluded on 24 November 2021 with a major step on the global development of norms on the use and regulation of Artificial Intelligence (AI).

193 member states of UNESCO signed and adopted the draft AI Ethics Recommendation. It can be touted as the first globally accepted normative standard-setting instrument in the realm of AI. The voluntary, non-binding commitment is a major point of cooperation between States and leaders in identifying principles of ethics in the regulation AI systems that find wide application in today’s world.

This is one step further in conversations that presume the inevitability of AI systems involving in decision making. Consequentially, it could be a first in outlining the foundations of global AI governance. Presumptions around business practices and current use cases may not hold good in the future. There is ample scope to re-evaluate these presumptions as more concrete policy is laid down and reconsider the perspectives on development and use of AI. Given the encouragement for multilateral and intercultural effort, these recommendations appear to be a good starting point for engaging further in the discussions for rulemaking and governance. The structure of the document, dividing the recommendations into values and policy actions facilitates the same. While far from perfect, it is noteworthy in its exploration of the need for a values-based mechanism to govern AI systems.

Values for Ethical AI

The Recommendation omits a clear definition of AI. Instead, it is aimed at AI systems that can demonstrate intelligent behaviour such as reasoning, learning, prediction, etc. This differentiation eliminates systems based on databases and simple programming from its ambit. It extends the standards established by other international instruments such as the UN Charter into the spheres where AI systems operate.

It establishes certain ground norms that are vital in a world where the technological divide affects different States and populations differently. The emphasis on benefit sharing and adequate attention to the cultural impacts of the use of AI systems is noteworthy. Acknowledging the dangers posed by AI systems through reinforcing pre-existing disparities is an important precursor to well-informed and adaptive governance models. It also addresses the issue of liability for actions of AI systems. Here, the signatories have agreed that the ultimate responsibility will lie on the human being who decided to cede control to an AI system in a particular instance of decision making. This is a blow to the bubbling interest in establishing independent liability on the AI system itself. International and domestic legal regulation will still have to solve the contentious issue of implicating liability on the manufacturer or the user or the owner of a system that takes flawed decisions.

Policy rooted in Ethics

In addition to furthering UNESCO’s longstanding values, the Recommendation also suggests policy actions to translate these principles into globally accepted normative standards. It calls for establishing compliance mechanisms to ensure human rights and rule of law are upheld, in consonance with the Constitutional provisions of the signatories. It urges the signatories to incorporate Ethical Impact Assessments as part of operationalising AI systems. It nudges States to include a transversal gender impact assessment as part of such Ethical Impact Assessments, which is a welcome suggestion that is seldom raised in conventional conversations around AI ethics. Research on AI systems has repeatedly shown the gendered nature of discriminatory decisions made by AI systems. Systems favour men over women and nonbinary individuals as many models inherit bias from training data or falsely identify gendered patterns in assessing merit.

Interestingly, it suggests that signatories tap into the resources at home through multistakeholder forums to establish a network of independent AI ethics officers. These officers are to audit and monitor the AI systems, ensuring adherence to ethical practices. Real-world AI applications are developing and deployed at great speeds, but it remains a niche area of expertise. While this suggestion appears to be well-intentioned, translating it into a policy solution will reveal the critical divide in the talent required to engage in such work in various parts of the world. This can only be remedied by greater international cooperation in education, training, and upskilling. Hence, such mitigating steps must be pioneered by international organisations such as UNESCO. Opportunities for India As the first concrete international consensus on normative standards, this can fundamentally impact the conversations that shape the regulatio