Research on artificial intelligence in the Netherlands

Countries around the world, including the EU and the Netherlands, are investing heavily in the development of artificial intelligence (AI). By 2030, the EU wants the combined investment of governments and businesses to have increased to €20 billion annually (European Commission, 2018a). To this end, in the period 2021-2027, it wants to invest €1 billion per year from Horizon Europe and the Digital Europe programme (European Commission, 2021).

In the Netherlands, the importance of AI is also increasing. For example, the government and industry developed the AiNED Nationaal Groeifonds Investeringsprogramma (National Growth Fund Investment Programme), which should lead to an investment of €2.1 billion in AI over the period 2021-2027 (NL AI-Coalition, 2021). In April 2021, the first phase of this plan was awarded €276 million from the Nationaal Groeifonds. In addition, at NWO and RVO we see an increase in both the size and the share of the funding that goes to AI-related projects (see text box below). 

Investments in research quality and capacity are an important part of both the European and Dutch plans. Its high quality research system is seen as one of the strengths of Europe, which has fewer large companies (and related investments) than China and the US. Therefore, in this fact sheet, we look at the scope and focus of AI research in the Netherlands and Europe.

In the first part of this fact sheet, we look at the scope of AI research in the Netherlands. What part of the Dutch research effort and output is focused on AI? We compare this share with the focus on AI research in the ten reference countries and the European (EU-27) and global average. As the figure below shows, the Netherlands is a relatively small actor in the field of AI. Therefore, in the comparison by country, we will use the share of the research performed within that country that can be characterised as AI research. China, the EU-27 and the USA are by far the largest actors in the AI field. Therefore, we will look at them individually.

Between 2013 and 2018, 14 million scientific publications were listed in the Scopus database worldwide, of which 314,212 were AI-related. Dutch researchers contributed to 4,047 AI publications. This is 1.3% of the total number of AI publications worldwide. Looking at all publications, Dutch researchers contributed to 2.1% of the total. 

In the same period, 4.1 million researchers were active worldwide. 66,447 of them were AI researchers. 985 of these AI researchers worked in the Netherlands for a period of time: 1.5% of the worldwide total. This is relatively little. Of all active researchers, 2.9% worked in the Netherlands for a period of time. 

Therefore, the Dutch contribution to AI research is smaller than the average contribution of the Netherlands to research in all disciplines combined. 

When we zoom in on the share of AI in the total Dutch research effort (researchers) and research output (publications), we see a similar pattern. When we compare the Netherlands to the other countries in this analysis, we see that a small part of Dutch research effort and output is focused on AI, compared to the average of the EU-27 and the worldwide average. We made this visible in the figure below by comparing the share of AI researchers in the total research population (Y-axis) to the share of publications that are AI related (X-axis). Both indicators show which part of the total (public) research in a country is AI research. 

Share of AI in research input and output per country

Of the researchers connected to the Netherlands between 1996 and 2019, 0.8% are AI researchers. This is lower than the global average (1.6% of researchers are AI researchers) and the EU27 average (1.2%). 

Of all (partly) Dutch publications from the period 2013-2018, 1.3% were AI-related. This is also lower than the global average (2.2% of publications) and the European average (1.8%).

During this period, the Netherlands greatly increased its focus on AI research: the share of AI-related publications increased by 115% between 2013 and 2018. Only the US and Japan saw a stronger increase in the share of AI publications (by 151% and 135% respectively). Iran and India, which have a high share of AI publications, are no longer growing as fast. 

This increase also comes to the fore in the high quality and international relevance of Dutch AI research. This can be deduced from the high average citation impact scores of AI publications of researchers connected to the Netherlands (see figure below). The citation impact of publications involving scientists from Dutch institutions is very high with an average of 2.08. This means that these publications have a citation impact of more than two-thirds of the total citation impact of the research, meaning that these publications are cited more than twice as often as the worldwide average in their field. Of the ten countries where most AI publications originate, only the US, Canada and the UK have a higher average citation score for AI publications.

China, Europe and the US are the three biggest actors in AI development - each with their own policy focus, strengths and weaknesses (Castro and McLaughlin, 2021; European Commission, JRC, 2018; Mols, 2019). The presence of a number of large companies is often seen as the US's strength, China's strength is its large public investment, and Europe’s greatest asset in developing AI is its high-quality research landscape. 

China spends a relatively large share of its public research on AI, as the second figure in this fact sheet already showed. Of all AI publications from 2013-2018, 26% have a Chinese author, 22% an author from the European Union (excluding the UK) and 18% an American author. By comparison, of the global total of publications in all fields, only 20% have a Chinese author. The European Union and the US play a slightly larger role (27% and 22% respectively). 

China's growing investment in AI has been visible since 2015, as the figure below shows. The number of AI publications with one or more Chinese authors grows from 11,192 in 2015 to 22,926 in 2018. 

The figure also shows that the EU-27 has not intensified AI research as much as China and the US. The EU-27 still produced 8,690 AI publications in 2013 - similar to China and 51% more than the US (5,750). In 2018, the EU-27 was involved in 15,346 AI publications (+77%), the US in 14,899 (+159%) and China in 22,926 (+151%).

The table below shows the changing proportions of China, the US and the European Union in percentages. While the share of China and the US in global AI publishing output increases (by 3.8 and 3 percentage points, respectively), the share of the EU-27 decreases by 4.6 percentage points.

This development of the European share in AI research, which was also visible in analyses by Elsevier itself (Elsevier, 2018), is worrying. After all, other comparisons between the US, China and the European Union show that high-quality academic research and the presence of AI talent are the stronger points of Europe's AI capacity, and thus important to retain (Castro and McLaughlin, 2021: European Commission, 2018b). 

As industry plays an increasingly larger role in AI research, the share of academic AI research may decrease even further. This increasing role of industry is visible in the AI Index 2021 (Zhang et al., 2021). This index cites two studies which show that companies are increasingly present at AI conferences and that more and more PhDs in the United States are choosing a career in business. 
    

Cooperation between sectors

Looking at the authors of publications, 92% of AI publications in the period 2013-2019 involve authors from academia (Scopus data AI index). In the same period, industry is involved in 6% of AI publications. The involvement of companies is highest in the US (19%). Within the European Union and China, industry’s role is much smaller, 7% and 9% respectively. In the Netherlands, the involvement of industry is high with 15%.

However, the role of industry in AI research has grown the most within the European Union. Whereas in 2013, only 3% of AI publications in the EU involved authors from industry, in 2019 this had already risen to 9%. In China, the contribution of industry grew from 3% to 7% over the same period, in the US from 17% to 19%.

Industry publications are often the result of collaboration with higher education researchers. The figure below shows the share of AI publications that are the result of such cooperation. Compared to other countries, the Netherlands has a high percentage of this type of cooperative research. 

In the second part of this fact sheet, we will take a closer look at the focus of AI research. What are the topics and components of AI research in the Netherlands and the European Union? How does this differ from other countries?

Worldwide, most AI publications cover the following research areas: machine learning, computer vision and neural networks (also in combination with other areas). Of all publications worldwide, 51% are linked to machine learning, 45% to neural networks and 36% to computer vision. 

With 11%, Fuzzy systems is the smallest research area. Since 2013, the research areas machine learning, neural networks and computer vision have grown the most. In practice, there is overlap between the various research areas. For example, neural networks are part of machine learning and you need both these techniques for computer vision. In the drop-down box below the following figure you can find more information about the various research areas.
 

Compared to other countries, AI research in the Netherlands has a relatively strong focus on the use of AI for planning and decision-making processes. In this research that includes social applications such as self-driving cars and robotics. 26,3% of the Dutch publications fall entirely or partly within this research area (planning and decision-making). This percentage is above the global average of 14.7% and is higher than in all other countries in this analysis. The research areas neural networks, fuzzy systems and search and optimisation receive relatively less attention in the Netherlands. Neural networks is an AI technique that imitates the human brain. Fuzzy systems is a form of reasoning that helps computers to convert undefined measures such as 'large', 'cold' or 'long' into a number. Search and optimisation is the use of AI to optimise search functions.

Since 2013, the research areas machine learning, neural networks and computer vision in particular have grown considerably, both globally and in the Netherland

This focus on planning and decision-making processes also becomes apparent when we compare the distribution of AI publications across all research areas in the Netherlands with those in the other countries in our analysis. In the first figure, we compare the efforts of the Netherlands with those of China, the US and the EU-27. In the second figure, we compare the Netherlands with the other countries in the European Union.

Share of the seven research areas in all AI publications (by country)

The focus of Dutch AI research is most comparable to that of the US, except for the stronger emphasis on the research area of planning and decision-making (26,3% of Dutch AI publications compared to 17,5% of the US). In the areas of neural networks and machine learning, the relative share of US AI publications is about 6% higher than that of the Netherlands. These research areas focus on systems that can identify patterns in data independently.

China has a different pattern. This country clearly puts less effort into language recognition and planning and decision-making and relatively more effort into fuzzy systems, a form of reasoning that helps computers to convert undefined measures such as 'big', 'cold' or 'long' into a number. With 40% of the publications related to computer vision (image recognition), China, together with India (42%), has the biggest output in this field.

Other research reaches similar conclusions

The EU Joint Research Centre came to (partly) similar conclusions (European Commission, 2018c). The JRC distinguishes between four categories (machine learning, connected and automated vehicles, speech recognition and natural language processing and face recognition). The JRC also concludes that face recognition plays a relatively large role in China, while the US relies more on speech recognition. At the same time, the JRC analysis shows that in the EU (in this case including the UK) the research effort is fairly evenly distributed, but there is clearly less focus on face recognition. It also shows a strong focus on connected and automated vehicles within Chinese AI research. These differences can be caused by a difference in categorisation. In addition, the JRC categorisation includes companies and looks at more types of output, such as patents and company registries.

AI in the European Union

Share of the seven research areas in all AI publications (by country)

Similar to the Netherlands, there is a relatively large focus within the European Union on planning and decision making (20%). The focus on fuzzy systems varies greatly between the various member states included in this study.

Within the European Union, the focus on various research areas differs. For example, Spain and France focus more on the research areas search and optimisation, the use of AI to optimise search functions, and fuzzy systems, which help AI systems to quantify undefined measures such as 'large', 'cold' and 'long'. Of all the countries in this analysis, Germany and France are most focused on natural language processing, speech recognition: 27% of their publications are on this topic.

The data in this fact sheet show that the Netherlands publishes high-quality AI research. Over the period 2013-2018, the focus on AI research has grown significantly. Nevertheless, compared to other countries, AI research occupies a relatively modest place in the Dutch research landscape. Europe is also in danger of increasing rather than decreasing its gap with the US and China.

However, since 2018, both the Netherlands and the European Union have increased their focus on AI. The European Commission's ambition in the 2018 AI Strategy is to grow European AI investment from €4-5 billion in 2017 (European Commission estimate, 2018a), to €20 billion annually by 2030 at the latest. These are investments by the European Union, Member States and private parties together (European Commission, 2018c). It is still too early to report on these investments. With its AI policy, the EU is choosing its own path, investing in AI that both boosts Europe's competitive position and provides added value for European citizens (European Commission, 2020; Mols, 2019; Rathenau Institute, 2020).

The Dutch government presented the Strategisch Actieplan Artificiële Intelligentie (Strategic Action Plan for Artificial Intelligence) (SAPAI) in 2019. AI played an important role in the recent awards for the National Growth Fund and the gravity grants.