As the excitement for the fourth industrial revolution looms, anxiety grips me. With AI positioned to reshape the world, I can’t help but question: Will jobs disappear? Is Nepal ready, or will we be left behind?
The Oxford Insight’s Government and AI Readiness Index 2023 paints a stark picture for Nepal. It ranks Nepal 150th out of 193 countries, far behind the top three: the USA, Singapore, and the UK.
The knowledge gap
Darshan Parajuli, a young Nepali entrepreneur, personifies the same anxiety. After witnessing the transformative power of Generative AI abroad, he returned to find most Nepalis unaware or reluctant to adopt AI. To address this, Darshan co-founded TeachMeAI to offer workshops aimed at bridging Nepal’s AI knowledge gap. He says, “It’s my mission to make AI easy, accessible and exciting for everyone.” He sees AI as having enormous potential to impact students and views it as a great equalizer. He believes that with awareness and upskilling, Nepal’s young generation can compete globally.
Despite his optimism, the current landscape paints a bleak picture. Experts estimate that only 5% of Nepal’s population uses AI regularly, and those figures seem accurate considering Nepal’s economic realities. With a per-capita income of $1,456, spending $240 (the cost of ChatGPT Pro) annually on AI seems inaccessible.
Specialized solutions for local problems
For LDCs like Nepal, the true potential of AI lies not in mass-consumer products (like ChatGPT) but in leveraging data to solve local problems. A study suggests that the needs of all nations are diverse, necessitating specialized AI solutions. For example, AI applications in Agro-Tech could significantly boost India’s agricultural sector.
Prabhat Khadka and Sampanna Tiwari are hoping to create such a specialized solution to address the legal literacy gap in Nepal. Their project is Juriease, a legal chatbot designed to answer legal questions using AI.
However, in developing Juriease, the team is encountering several challenges, particularly among them is the processing of Nepali legal texts. Juriease currently is built upon base LLM models like Gemini and ChatGPT, which support the Nepali language, but their capabilities are limited. Prabhat notes that the results often miss linguistic nuances, producing odd translations and sometimes even producing Hindi as a response. This is because the models are mostly trained in Hindi for the large consumer base of India, and these companies have limited incentive to prioritize or focus on the smaller consumer base like Nepal.
These issues highlight the need for a native LLM model trained specifically in Nepali. Prabhat, too, recognizes that developing such a model will be crucial for Juriease, viewing their current product as but a proof of concept.
However, the development of a native LLM is not easy; Prabhat acknowledged that extensive investment would be required for such a model. “Access to AI infrastructure is not a problem,” Prabhat notes to me. “The issue is access to investment.”
This is because creating powerful language models like ChatGPT demands millions of dollars of investment in computing power. Providers for such infrastructure are limited and charge a premium for access, and big tech companies are willing to pay the price. However, for entrepreneurs in most developing nations, the cost can be an insurmountable burden.
Job market and inequality
AI’s disruptive potential extends beyond the tech industry, affecting jobs across multiple sectors. However, experts say this is natural. A study revealed that more than 60 percent of employment in the United States in 2018 was in job titles that didn’t exist in 1940. Thus, while jobs in other sectors may be reduced, the scope of AI-related jobs is expected to increase. The nature of jobs created will change; for instance, new jobs like data labellers who tag and filter data for training AI are emerging.
However, these new jobs often exacerbate inequality. They primarily benefit highly skilled labor and capital owners. Trends show rising wage premiums for high-skilled workers, while low and middle-skilled workers face sluggish wage growth. AI is expected to intensify these disparities. It is evident that while an AI engineer at the top would be paid a premium, jobs like data labelling will require minimal skills and thus be paid less. This reduction in pay and jobs can widen the income gap, which affects not just individuals within countries but also between nations.
But all is not lost.
Bishal Kharal, a young engineer from Nepal, is working remotely as an AI Engineer at Fagoon AI. Currently, Bishal is working on highly sought-after sectors like LLMs and diffusion models, which qualifies him as a highly skilled tech employee. He exemplifies the participation of individuals from developing countries in the AI revolution.
But maybe even more interestingly, he is a self-taught engineer.
When asked if such upskilling was possible for other students like him, Bishal notes, “With enough hard work and dedication, one can learn the necessary skills regarding AI.” As he points out, “The mathematics course in Grade 12 in Nepal is advanced and really helped in such a transition.”
Bishal, however, observed that in Nepal, the utilization of AI for productive work is minimal. He emphasized that to tackle local problems, development in technologies like native LLMs trained in the Nepali Language would be required. Such technology would require extensive work and collective effort.
Ultimately, these are massive challenges that risk undoing years of efforts put into reducing income inequality and narrowing the gap between developing and developed nations. They can’t be ignored. There’s an urgent need to democratize AI and address the digital divide. But the stories of Darshan Parajuli, Prabhat Khadka and Bishal Kharal offer optimism about Nepal’s participation in this Fourth Industrial Revolution.
Saugat Pokharel is a BA LLB fourth year student at the Kathmandu School of Law. The longer version of the above article was first published in the Global Campus of Human Rights Magazine.
Comment