Navigating Generative AI Adoption in Developing Nations: The Case of Pakistan

In the context of developing countries like Pakistan, the journey of adopting generative AI brings to light not only its promises but also the hurdles that must be overcome. While much of the AI conversation revolves around advanced economies, it’s equally vital to shed light on how this technology influences developing nations. In this article, we take a deep dive into the challenges that accompany the integration of generative AI in the IT landscape of Pakistan.

Societal Risks and Challenges in Innovation

The primary obstacle faced by AI innovators in low and middle-income countries (LMICs) like Pakistan lies in their structural challenges. Introducing new technology into socio-technical systems like public services, health or education poses a significant difficulty. This is further deepened by infrastructure gaps, including inadequate power supply, limited internet accessibility, insufficient data storage facilities, and low cloud adoption as well as virtually non-existent native cloud computing platform. Consequently, addressing these challenges requires a constant process of refinement both during the product development phase and in the post-development period.

Innovators encounter the obstacle of institutional inefficiency and a prevailing lack of confidence in novel technologies among both governments and broader societal stakeholders. In connection to this, the insufficient allocation of funds by governments towards technological advancements has offered minimal motivation for the proliferation of AI innovation.

Lack or insufficient access to high-quality and inclusive data poses an additional challenge to promoting AI innovation. In many cases, innovators from resource-constrained environments turn to models created by major tech firms located in the US, China, and Europe. However, the data used to train these models might not accurately reflect the realities of low and middle-income contexts like Pakistan. Furthermore, most of these state-of-the-art models aren’t freely accessible, resulting in substantial financial burdens to acquire them by the innovators.

Another significant challenge stems from the scarcity of AI experts and companies providing AI solutions. A recent survey conducted in 2022 by the Pakistan Software Houses Association revealed that merely 14% of businesses in the country are engaged in offering AI solutions. This scarcity is further highlighted by the wage discrepancy. On average, a senior AI engineer in Pakistan earns approximately $10,000 per year, while their counterparts in the United States command an average annual salary of about $120,000 and in India it is around $60,000. These challenges create hurdles in building a robust AI ecosystem within the developing countries like Pakistan. This underlines the pressing need for concerted efforts to stimulate the growth of AI-driven enterprises in Pakistan.

The challenges described so far are infrastructural or technological in nature. Generative AI adoption can have an impact on a much larger scale. These risks can be categorized into three distinct levels:

• At the system level, there are inherent hazards like bias in technology, concerns about data privacy, instances of incorrect answers arising from the model’s overconfidence (known as hallucination), and security vulnerabilities.
• On the structural level, there’s a possibility that Generative AI might heighten existing societal and economic inequalities, leading to outcomes like marginalization and job displacement.
• At societal level, there is a risk regarding AI’s capacity to distort democratic and societal perspectives through misinformation, polarization, and the creation of deepfakes.

Considerations before further deployment

Before proceeding with expanded implementation of Generative AI solutions, it is vital to thoroughly explore the implications, potential risks, and opportunities that these technologies present in environments with limited resources, like Pakistan.

It is essential to cultivate support and agreement from stakeholders spanning diverse domains of AI application. This becomes particularly critical within the public sector, where challenges might arise – for instance, in the integration of AI-driven chatbots for public service assistance, government officials, technology experts, and citizens would need to collaboratively define the scope, capabilities, and ethical boundaries of these chatbots. Adoption of these solutions becomes challenging if citizens perceive this as infringing on their privacy or if government officials resist delegating certain administrative processes to automation.

Conversely, an excess of technology may not invariably resolve issues. Establishing unambiguous directives for Generative AI acquisition within the public sector could aid in discerning instances where its implementation might be redundant or unnecessary.

It is also important  to evaluate the societal ramifications of Generative AI. Policymakers and pioneers in AI should gather the general sentiment surrounding the technology, encompassing aspects such as its appropriate domains of application, methods of utilization, desired extent of implementation, and the overall scale. Such an assessment would facilitate the alignment of governmental objectives with the aspirations of end-users.

There is a necessity for regulatory bodies and creators to join forces in developing systems that authenticate the quality and comprehensiveness of datasets or models employed in the broader implementation of Generative AI. Although, Pakistan’s Ministry of Information Technology & Telecommunication (MoITT) has unveiled a preliminary draft of the country’s inaugural National Artificial Intelligence Policy in May 2023, with the final version to follow. However, to progress, concerned stakeholders should refine the policy to establish a more inclusive, efficient, and effective framework by addressing some key points:

• Promote a cohesive approach to AI research and development by consolidating existing fragmented efforts. It is essential to streamline overlapping research endeavors and harness resources toward a unified national objective, enhancing the impact of AI advancements.
• There is a need for a comprehensive national consensus on AI by actively engaging in inclusive multi-stakeholder dialogues that involve academia and industry.
• Federal and provincial governments should devise policy frameworks to overcome resistance to digital transformation within the public sector by instituting vigilant political oversight. Simultaneously, they should transition away from traditional bureaucratic norms, allowing for a more effective adoption of AI.
• Instead of imposing fixed deadlines, it’s advisable to maintain flexibility when outlining ambitions. This approach prevents the tendency of shortcuts to attain objectives, ensuring a more thorough and effective integration of AI technologies.

Last but not least, there should be an emphasis on tailoring policies to the distinct applications of Generative AI within specific sectors. This approach enables policymakers, in collaboration with the technology community, to determine which models carry excessive risk for deployment in certain sectors while posing minimal harm in others. Crucially, AI advancement should incorporate its own human moderation mechanisms (for instance, generative media platforms should establish transparent provenance protocols prior to integration into newsrooms).

Conclusion

In the evolving landscape of technology, embracing generative AI in developing nations like Pakistan is a path filled with both opportunities and hurdles. By addressing challenges such as limited resources, stakeholder alignment, and ethical considerations, Pakistan can pave the way for a future where Generative AI benefits society at large. Through collaborative efforts, informed policies, and tailored approaches, the country can navigate this transformative journey, ensuring that the promises of generative AI are realized while mitigating its potential risks.

About the writer: Dr. Usman Zia is an Assistant Professor at the School of Interdisciplinary Engineering and Sciences, National University of Sciences and Technology (NUST), Pakistan. His research interests are Natural Language Processing and Machine Learning. He has authored numerous publications on language generation and machine learning. As an AI enthusiast, he is actively involved in a number of projects related to generative AI and NLP.