In December 2022, ChatGPT suddenly became known around the world. In only the second month since its launch, ChatGPT is already reported to have had 100 million active users in January 2023 alone. For most users, it became clear that this could boost their productivity enormously: from delivering document outlines or suggesting useful data sources to replacing call centres and accounting activities.
From a development perspective, this raises the question of who will ultimately benefit from this technology, and what this means for workers in the Global South. Concerns about how technology shapes work are of course not new. In 1930, economist John Maynard Keynes speculated that the twentieth century would see widespread unemployment as machines replaced workers. This did not turn out to be true. Rather, specific tasks have been replaced by machines and humans have found new (more productive) tasks in which to specialize – think about software engineers and influencers1. With ChatGPT, for example, workers may spend less time accumulating information and more time analysing or deriving meaning from it.2
Like many technologies, AI is likely to have a bifurcating effect on the wage distribution. Workers performing these new, more complex AI-assisted tasks will likely strengthen their position in the global wage distribution as they become more productive. These workers are typically higher skilled and are already engaged in more complex activities. Those that perform relatively simple or routine tasks, on the other hand, have a higher chance of being replaced by AI (or other technologies) and may have to move into lower wage segments.3 At the global scale, countries with a well-trained, highly skilled population have a higher chance of adopting these technologies and benefiting from the opportunities they offer, while others fall further behind in the global wage distribution.
ChatGPT is different to many other AI and industry 4.0 technologies (like robots) that have been empirically studied.4 ChatGPT is not tied to the factory floor – that is, it does not require large investments (by the end user) and its uses are not limited to manufacturing. It is potentially even more useful for (new) services, and it is (as of February 16, 2023) free to use. The only barrier to entry is access to the Internet, and it can enhance the productivity of anyone who works with a computer. ChatGPT therefore provides an interesting case for exploring the use of AI technologies that are widely accessible across a large set of countries.
Google Trends provides close to real-time data on who is using ChatGPT as it measures the relative popularity of a search term in a given location. Data on the term “ChatGPT” was retrieved on 26 January, reflecting the first “hype” period since its launch. Although this does not necessarily reflect the actual number of users, it is likely to be a good proxy as a high share of first-time users probably looked for it on Google.5
Even though ChatGPT was freely and widely accessible, it turns out that there is indeed a positive correlation between these search trends and country characteristics that are indicative of skills in the population. "ChatGPT" searches tend to represent a greater proportion of all search terms in countries that score higher on the World Bank’s Human Capital Index (see figure above), and in countries with a higher number of STEM (science, technology, engineering and mathematics) articles per inhabitant (see figure below). The latter, in particular, is strongly correlated with the Google Trends, and can be understood as a proxy for the population’s scientific and innovation capacity.6 This also implies that people in lower-income countries search less for ChatGPT, as human capital indicators are strongly correlated to gross domestic product (GDP) per capita.7
While more research is needed to establish the causality behind these patterns, these correlations indeed suggest that workers in lower-income countries make less use of AI technologies even if they are freely and widely accessible. It may be that these workers cannot make use of them because they do not have the skills to do so, or because their tasks cannot be enhanced by these technologies. It may also be that workers could profitably make use of these technologies if only they were aware of them and knew how they could be deployed.
Looking ahead, there is also a risk that workers in the Global South may soon face much higher barriers to using these technologies. ChatGPT has already announced the launch of a premium version that offers preferential server access for $20 a month (to start recovering the high costs of running it). Non-premium users will have no access to ChatGPT when its server capacities are at the maximum. Pricing will constitute a relatively higher barrier in the Global South due to lower average incomes. As these technologies become proprietary, countries may well start developing their own AI infrastructure to enhance workers’ productivity. The Global North has an advantage here too because of its financial capacities and already existing digital infrastructure, as illustrated by its substantial server capacities. Not surprisingly, high server capacity per inhabitant is concentrated in countries like the USA, Germany and the Netherlands (see map below).
What can be done?
Addressing these international disparities relies on closing both the financial and physical capital deficit and the skills gap. First, international financial and technical cooperation are crucial for helping countries in the Global South overcome the cost- and (digital) infrastructure barriers identified above. Second, more must be done in the field of skills development and awareness-raising. There are some promising models for closing the skills gap through public–private development partnerships: the public sector partner (host country) identifies target populations, private sector partners ensure that training curricula and technologies meet local industry standards, and the development partner ensures social inclusion.
The results presented here re-emphasize the need for effective approaches to be developed, rolled out and rigorously tested, to identify best practices for closing the international digital divide.
Disclaimer: The views expressed in this article are those of the authors based on their experience and on prior research and do not necessarily reflect the views of UNIDO (read more).
- Acemoglu, D., & Restrepo, P. (2018). The race between man and machine: Implications of technology for growth, factor shares, and employment. American economic review, 108(6), 1488-1542.
- See Acemoglu and Johnson (2023) for a more critical take on the potential benefits of ChatGPT, arguing that these technologies are currently mainly developed for cutting labor rather than enhancing it, see here: https://www.project-syndicate.org/commentary/chatgpt-ai-big-tech-corporate-america-investing-in-eliminating-workers-by-daron-acemoglu-and-simon-johnson-2023-02?barrier=accesspaylog
- Goos, M., Manning, A., & Salomons, A. (2014). Explaining job polarization: Routine-biased technological change and offshoring. American economic review, 104(8), 2509-2526.
- Artuc, E., Bastos, P., & Rijkers, B. (2020). Robots, tasks, and trade. CEPR Discussion Paper DP14487.
- The cross-country differences are correctly portrayed if the share of people directly going to the site and/or using a different search engine relative to those using Google to find ChatGPT is the same across countries.
- In exploratory multivariate regressions, STEM articles appear to be the main correlate with the Google Trends data. In the shown figure, about 10% of the variation in STEM articles is explained by the Google Trends data (R2=0.13) and, roughly speaking, as relative search interest doubles, so does the number of STEM articles.
- There is also a statistically significant correlation between GDP per capita and Google Trends data.