Artificial Intelligence could be classified as a general-purpose technology (GPT), one that has many applications across a wide spectrum of uses. The growth of globalisation in this sub-field of technological research, therefore, forms the backbone of long-term economic growth and development. It’s also hardly surprising that in the age of the fourth industrial revolution, it is technological globalisation that dominates all other forms of globalisation, such as economic, financial, cultural, or political. 

A free exchange of people and ideas forms the bedrock of globalisation. Globally, there is a rapid shift to be noticed from nationalised systems innovation to globally connected systems of knowledge creation. Such a play of scientific globalisation is especially relevant to the field of AI which is largely propelled by open data sharing, open source software and other open science principles.

In a recent paper co-authored by Saurabh Mishra and Kuansan Wang, the researchers have examined the gaps that exist in scientific research, in general, and AI research, in particular, between the developing and developed economies. The paper titled “Convergence and Inequality in Research Globalization” attempts to find answers to many important questions related to scientific globalisation. Is there convergence in AI research? Is there hope for developing countries to catch up with the advanced economies in research output and impact? What is the role of science policy to help the developing countries catch up with rich countries?

Such a comparison is better understood by the contrasting effects of the Matthew Effect and the Catch-up Effect. The catch-up effect supports the idea of convergence between poorer economies and wealthier economies, due to a faster growth rate in the poorer economies. The Matthew Effect, on the other hand, is the concept of divergence which propagates that the rich get richer and the poor get poorer.

The researchers have conducted an in-depth study based on scholarly and patent publications covering STEM research from 218 countries/regions over the past four decades, covering more than 55 million scholarly articles and 1.7 billion citations. The study employs a quantitative and data-driven approach. Unique to this investigation is the simultaneous examination of both the research output and its impact in the same data set, using a novel machine learning based measure, called saliency, to mitigate the intrinsic biases in quantifying the research impact.

What does the research conclude?

Research output is measured by the number of scholarly articles published, and the research impact is measured by the citations that are received. Therefore, convergence is whether the share of output and impact is increasing for developing countries. 

In AI, the sign of a convergence in research impact is modest but encouraging. 

There is a growing list of indicators, including composite indices that are used to guide AI policies of countries and global organisations. Many of them lend support to convergence as a necessary outcome, where it can be observed that the dominant are becoming less so. On the other hand, deep learning’s unanticipated rise appears to have resulted in divergence, where compute divide between large firms and non-elite universities increases concerns of "de-democratization" of knowledge production.

It is also widely observed that inequality, characterized by the Matthew Effect, is deeply entrenched in the global research ecosystem, confirmed again in a recent study that shows top-cited scientists, often affiliated with high-ranking institutions in the western world, are receiving ever increasing citations. 

In terms of output, the lower income countries have produced more research in AI. AI is in the middle field of convergence with upper middle income countries, catching up in research output but not yet in impact. 

A case for collaborative research

Collaboration can help bridge the divide. Developing countries can foster research collaborations with leaders in fields of strategic importance. For example, the US-China have become the largest research collaborations, not just in AI but many other STEM fields helping China converge with rich countries. Newcomers can receive faster and broader recognition by associating themselves with those already highly recognised. 

Data shows that most of the impact gains in China’s AI research in recent decades can be largely attributed to research with global collaborations. The fact that collaborative work increasingly accounts for a higher saliency ratio suggests high impact research is likely a result of global collaboration, an encouraging indication to support a convergence of globalised research. 

In the field of AI, china has surpassed US in output but is still dwarfed in impact. Further, a transpacific collaboration has led to higher output and impact than transatlantic. US is seeing a larger portion of its output and impact coming from its collaboration with China in recent years than with Europe earlier in the decade of 2010s.

• Read the full paper here

Image by Gerd Altmann from Pixabay