The rapid emergence of Generative Artificial Intelligence (Gen AI) and its transformative power, exemplified by the widely utilized transformer architecture, has significantly altered the digital landscape. Tools like OpenAI’s ChatGPT have democratized access to cutting-edge AI capabilities, allowing users to generate high-quality content across various mediums with minimal infrastructure or prior experience. While Gen AI offers numerous benefits, its distribution and consequences are not uniform across the global stage, with some countries reaping the rewards more than others.
Risks in Low-Resourced Countries
The benefits of Gen AI systems have primarily accrued to countries with high-resourced languages, such as English, leaving low-resourced countries at a disadvantage. This discrepancy in access and impact becomes more pronounced in nations characterized by histories of violent conflicts and instability, labeling them as risk-critical. In these regions, the potential risks associated with Gen AI are exacerbated.
Misinformation and Disinformation
Gen AI’s proliferation, especially in countries with less regulated media environments, has given rise to concerns about the spread of misinformation and disinformation. Bad actors in risk-critical countries can exploit Gen AI’s capabilities to manipulate public opinion, akin to what has been observed on social media platforms. The consequences can be dire, as the search for truth becomes increasingly elusive.
Synthetic Media and Deepfakes
The strategic use of synthetic media, including deepfakes, poses a significant threat in risk-critical countries. With limited media control, combating the proliferation of deepfakes becomes an uphill battle, further exacerbated by the use of Gen AI. Recent events, such as the Israel-Hamas conflict, demonstrate how Gen AI introduces new risks even in countries with established media ecosystems.
Gendered Harassment
In societies where chastity is highly valued, gendered harassment is a prevalent means of targeting women and girls. Gen AI amplifies this issue, making it easier for harassers to create and disseminate harmful content. The real-world impact of such cases can be profound, leading to tangible harm to individuals.
Data Bias
Bias within AI models is a well-documented issue. In low-resourced countries, it is challenging for developers to identify and rectify bias without access to context experts. As a result, risk-critical markets face longer response times from Gen AI developers, rendering them more susceptible to the adverse consequences of bias.
Recommendations for Mitigation
Addressing these risks is imperative to ensure Gen AI’s responsible use and mitigate its negative consequences, particularly in risk-critical countries. The following recommendations are essential steps in this direction:
Diverse Assessment Teams
Gen AI models should undergo internal and external assessments by diverse teams of experts. Red teaming, a process involving external experts to evaluate Gen AI models for risks, should be open to participants from diverse geographical locations, avoiding restrictions based on compensation or residency. This will help identify and address ongoing risks effectively.
Content Attribution and Safety Guardrails
To combat disinformation, content generated by Gen AI should be clearly marked with watermarks, indicating its AI origin. Safety guardrails, such as bias evaluations of training data, restrictions on graphic training data, and policies to prohibit sexual content generation, should be prioritized. When implementing content policies, cultural specificity and the involvement of context experts are crucial, with continuous updates to remain relevant.
Multilingual User-Reporting Systems
Large Gen AI companies should make user-reporting systems available in the languages of risk-critical countries. Currently, the availability of such reporting systems is often limited to English, leaving users in these countries with limited means to address issues and concerns.
