An organization's AI ethics policy is a highly consequential matter. Reports of gender bias in employment algorithms or job search results that are widely publicized can harm a company's brand, put it in legal hot water, and potentially result in large fines from the government. Organizations are progressively developing specialized procedures and structures to proactively instill AI ethics as a response to these challenges. A few businesses have taken this a step further and established institutional frameworks for AI ethics, HBR reported.
However, a crucial point that is often overlooked is the fact that ethics vary depending on the cultural setting. First, moral principles from one culture might not apply in another that is essentially different. Second, it's crucial to consider the significant variations in ethical reasoning that may exist at work, such as cultural norms, religious tradition, etc., even in cases where people agree on what is right and wrong. Lastly, laws pertaining to AI and related data are rarely consistent across national borders, which could lead to variations in the compliance-related aspects of AI ethics. Businesses and their clients may suffer serious consequences if these considerations are ignored.
As of right now, the majority of the developing international norms pertaining to AI ethics are Western-centric. One consolidated collection of reports, frameworks, and recommendations, the AI Ethics Guidelines worldwide Inventory (AEGGA), for instance, gathered 173 guidelines by April 2024 and observed that "the overwhelming majority [came] from Europe and the U.S." – not as worldwide as one may assume. However, a lot of businesses just use these standards anywhere they operate.
AI models are also indirectly encoding Western viewpoints. For instance, according to some estimates, the Chinese and Indian diasporas, who together make up one-third of the world's population, are represented in less than 3% of all photos on ImageNet. In general, limited predictive power and bias against underrepresented groups are likely results of low-quality data, and it may even be difficult to design tools for certain populations at all. For example, languages that aren't widely represented on the Internet can't now be educated for LLMs. According to a recent poll of Indian IT companies, the biggest obstacle to ethical AI practices is still a shortage of high-quality data.
An unregulated lack of diversity in ethical considerations might hurt businesses and their clients if artificial intelligence takes hold and begins to dominate company operations.
Companies must create a contextual global AI ethical model that emphasizes cooperation with regional teams and stakeholders and gives those teams decision-making authority in order to solve this issue. This is especially important if their business operates across multiple regions.