  

    

 


 


 
 
 







 
 
 
 
 
 
 
 
 
 
 
 


  
The mission of STINT, The Swedish Foundation for International Cooperation
in Research and Higher Education, is to internationalise Swedish higher education
and research. STINT promotes knowledge and competence development within
internationalisation and invests in internationalisation projects proposed by re-
searchers, educators and leaderships at Swedish universities.
Artificial Intelligence (AI) receives much attention worldwide, as AI is expected to
revolutionise technological development and business. However, several ethical
aspects must be considered.
Many countries are investing in AI and developments are rapid. It is therefore im-
portant that Swedish research environments are part of developments and stay
connected to relevant foreign research environments in this area. Furthermore, in-
ternational research cooperation contributes to quality and international networks
give access to involvement in the development of AI technology, implementation,
standards, and ethical frameworks.
The intention of this study is to provide an overview of current networks in AI
that involve Swedish and international researchers, in order to better understand
the appropriate initiatives needed to position Sweden as a research and innovation
leader in the field.
The project leader of this study is Dr Niklas Z Kviselius, Representative in North
America, STINT. Co-authors of this report are Dr Erik Forsberg, Representative
in APAC, STINT, and Dr Hans Pohl, Programme Director, STINT. We would
like to present our sincere thanks to Prof. Amy Loutfi, Örebro University, who has
read and provided valuable comments on draft versions of the report.

Dr Andreas Göthenberg
Executive Director, STINT

Stockholm, September 2022


 
Developments in AI affect a wide range of research areas – from medical break-
throughs in cancer research to cutting-edge climate change research. AI will have
a substantial impact on the way people live and work in the coming decades. By
now most large economies have drafted and released some sort of national AI pol-
icy or strategy.
This report focuses on the development of AI research in four countries, Canada,
China, Sweden, and the United States, as well as trends in their international re-
search collaborations in the field. International collaboration patterns are mapped
via publication analysis based on Scopus publications and Elsevier’s definition of
AI publications.
AI enjoys increasing interest from researchers, with China dominating the pro-
duction of AI publications followed by the United States. China and the United
States are the largest competitors and collaborators in AI. Sweden has in recent
years shown a very high growth in AI publications, only topped by China. Overall,
AI research is less international than average research but the high numbers of AI
publications from China and India contribute to this low overall value. In almost
all other listed countries, including Sweden, AI publications are slightly more in-
ternational than the average. The share of academic – corporate co-publications is
higher within AI research. Sweden has the highest share of publications including
at least one academic and one corporate actor.
Observations from the Swedish research system includes KTH Royal Institute of
Technology as the largest producer of AI publications in Sweden, Chalmers Uni-
versity of Technology showing the strongest positive trend, and Luleå University
of Technology having both the highest share of AI publications and the most in-
ternational AI research. Chalmers University of Technology has the greatest share
of co-publications with the corporate sector in Sweden. However, collaboration
patterns between institutions do not indicate that Swedish universities have estab-
lished strategic international collaborations within AI with leading actors abroad.
One clear observation on the global arena of AI research is that the United States
and China, however fiercely they compete with each other, clearly are advancing
fast together and leaving other countries behind, both in terms of research and the
pace of innovation. A combination of strong technology sectors and high govern-
ment interest and support in these two countries will further widen this gap.
The report concludes in chapter Observations with some commentary on what
lies ahead for Sweden on how to best position itself to ensure access to and imple-

 mentation of the leading edge of global AI advancements. This would include fos-
tering an AI research and innovation ecosystem that is sufficiently strong to be
able to attract leading scientists, innovators, and companies in the AI field to col-
laborate with, and/or work in Sweden.



Artificial intelligence (AI) is crucial to our future because AI forms the very foun-
dation of computer learning. Developments in AI affect a wide range of research
areas – from medical breakthroughs in cancer research to cutting-edge climate
change research. AI will have a substantial impact on the way people live and
work in the coming decades.
These promises also come with a complex and far-reaching set of risks which
regulators must consider. Some of the top ethical issues in AI include unemploy-
ment and labour considerations when jobs are automated, diversity and inclusion
when AI is designed with unwanted biases, and increased security vulnerabilities
when society becomes dependent on ever more complex and autonomous sys-
tems acting without human intervention.
The biggest research nations, as well as many of the largest corporations, are well
aware of such risks and all try to back up their research community and industries
with national policies dedicated to AI, research funding, and the building of at-
tractive ecosystems where cutting-edge research can be translated into societal
applications. Nations compete as well as cooperate on AI, and the scarce resource
is often human talent. Smaller research nations, such as Sweden, should be aware
of this new space race, and ought to try to leverage existing and international
collaborations to stay competitive.
The European Commission defines AI as systems that display intelligent behav-
iour by analysing their environment and taking actions –with some degree of au-
tonomy– to achieve specific goals. In short, AI leverages computers and machines
to mimic the problem-solving and decision-making capabilities of the human
mind. Applications include self-driving cars, self-learning manufacturing robots,
and various types of decision-making systems in healthcare.
Given the wide range of applications for AI and the tendency to add AI as a key-
word to increase the appeal of a project, clearly defining the scope of AI-related
research is challenging. The publication analysis in this report is based on work
done by Elsevier, see Figure 1.


   

   
  
 



  
  
  
  
  
   
  
  
 
 
 
 
 





    

    

Based on keywords from teaching, research, industry and media, Elsevier com-
piled a shortlist of 800 keywords, which were used to extract 6 million documents
from Scopus. These documents were scrutinised by experts and a gold set of
1,500 documents was selected for training the search engine. In the end, the
query developed with AI support returned 600,000 documents representing dif-
ferent aspects of AI (Elsevier 2020).
This report focuses on the development of AI research in four countries, Canada,
China, Sweden, and the United States, as well as trends in their international re-
search collaborations in the field. A brief overview of national policies in Singa-
pore and the EU is also included. These countries have been selected partly
because they are important in the AI field, and partly because STINT has good
access to data given its regional representatives in North America and China.
China and the United States are at the forefront of funding AI – by sheer research
volumes but also because of the prioritised status AI enjoys in their national poli-
cies. Canada was perhaps the earliest country to develop a national AI strategy
and has built upon decades of continuous funding of the field. Sweden is of ob-
vious interest for STINT.


  
       
  
Using Scopus publications and Elsevier’s definition of AI publications (see above),
it is clear that AI enjoys increased interest from researchers. Figure 2 indicates the
development of the total volume of such publications as well as the share of AI
publications, according to Elsevier’s method (2020), among all Scopus publica-
tions in the world irrespective of scientific field. This share rose from 0.6% in
2000 to 2.9% in 2020.

      

 
















 




 

One interesting detail is that AI publications to a very high extent (~99%) have
ten or fewer co-authors. Typically, hyper-authored publications require a special
approach when for example collaborations are studied. However, since hyper-
authored AI publications are so rare, this is not deemed necessary.
Based on the content of their titles, keywords and abstracts, Elsevier generates a
fingerprint of the publications. In Figure 3, a fingerprint of all AI publications
for 2016 –2020 is presented. The dominance of the green colour reflects rapid
growth and the size of the keyphrases their relevance. Deep learning and neural
networks appear to have been the most relevant keyphrases during this period.

   

During the period 2016 – 2020, China dominated the production of AI publi-
cations with almost 100,000 articles, conference papers, reviews, books and book
chapters, see Table 1. China is followed by the United States. The table lists a se-
lection of countries and these are used for comparisons below. The selection was
guided by the ambition to include some of the main collaborators with Sweden
and the other countries specifically addressed in this report. Sweden’s share of
global AI publications was 0.7% during this period. In comparison, Sweden’s
overall share of all publications was 1.4% for the same period.

    

   



   

   
   
   
   
   
   
   
   
   
   
   


In Table 2, some other important indicators for the same countries are listed.
The trend figure in the third column indicates the average annual growth in AI
publications over the five-year period. It reveals that Sweden has shown a very
high growth, only topped by China. The share of AI publications relates to the
national volume. India has the highest share, which means that 4.1% of all pub-
lications with at least one Indian (co-)author is within the AI field.
The field-weighted citation impact (FWCI) is a normalised indicator of how
often the publications are cited. A value above 1.00 means that the publications
are more cited than the world average. For all countries, the AI publications has
a much higher FWCI than the national average. Singapore has the highest FWCI
for the period and the very high value for the United States should also be noted,
given the large overall volume of publications in the country.
   


   
   
    

  
            
           
           
           
           
           
           
           
           
           
           
           
           
           

The share of publications in the first percentile is another indicator related to

the quality of the publications. This share covariates with the citation impact
and Singapore’s very high share of such publications indicates that AI research
in Singapore is highly cited by other researchers.
International co-publications normalised for the scientific field and other aspects
are indicated in the following columns. Overall, AI research is less international

 than average research, as 0.55 for all publications is clearly lower than 1.00. The
high numbers of AI publications from China and India contribute to this low
overall value. In almost all other listed countries, including Sweden, AI publica-
tions are slightly more international than the average.
Finally, the share of academic – corporate co-publications is higher within AI re-
search. Sweden has the highest share which means that 13.1% of the publications
include at least one academic and one corporate actor.
A closer look at the proportion of each scientific field in the total publication vol-
ume reveals differences between the priorities in each country’s AI research. In Fig-
ure 4, the height of the bars indicates the size of the share of all AI publications in
the country related to each scientific field, using the OECD definitions. The sum
of all six bars per country is 100%. Iran, for example, invests much more in AI re-
search related to Agricultural Sciences than the other countries and the world, see
Figure 4.

 










          
           
   

  

    
    
 

 
   
 












           
           
           
      

  


 


 










           
           
           
       

 

 
 











        
          
  

         
      
 
   
 
 

  








         
       
 
  

          
      
 
     
 



   






         
           

       
    

  
 
    
 
 


Whereas Engineering and Technologies (Figure 5) and the Natural Sciences (Figure
8) could be considered at the core of AI research, investments in fields such as
Medicine (Figure 7) might be interpreted as more applied research. Overall, Swe-
den’s AI research profile is similar to the global profile but with a slightly stronger
emphasis on Medicine.
One interesting, but difficult to study, dimension is dual-use AI research with
both civil and military applications. Even though it is possible to break down
the scientific fields further, this dimension is probably impossible to identify in
normal studies of publication data.

  

  
  
Moving to the institutional level, some indicators were calculated for the five in-
stitutions with the largest number of AI publications in each of the countries
specifically addressed in this report. In addition, five corporate actors were also
included following the same selection criterion. Four of them are based in the
United States and one in China.
The University of Toronto has the highest number of AI publications in Canada
and the most rapid increase in volume, see Table 3. The trend, as mentioned
above, is calculated as the annual growth rate between 2016 and 2020, using lin-
ear regression to calculate the endpoints of the interval. The University of
Toronto’s AI publications have a very high citation impact and include corporate
actors to a rather high extent, in comparison with the other universities. However,

    


  
  
   

 
          
          
          

          
          
          

this is also the university with the highest number of publications overall; the
four other universities are all more focused on AI, with the University of Waterloo
having the highest share of AI publications at 3.6%.

  


  
  
   

 
          

          
          

          
          

          

The Chinese universities are large producers of publications, see Table 4. The
fifth Chinese university produced more publications than the leading ones in all
of the other countries. Tsinghua University has the highest share of AI publica-
tions and the highest citation impact. Its publications are also the most interna-
tional, and they include corporate actors to a high extent. The top five producers
of AI publications in China are also the largest producers of scientific publications
overall. However, below the top five producers of AI publications in China one
finds several of the ‘Seven Sons of National Defence’ (the universities in China

 with the closest ties to the military) higher up in the list than in the overall rank-
ing. While the traditionally top Chinese universities dominate AI research in the
country, several new highprofile AI institutes have been established and can be
expected to make significant contributions. These include the Zhejiang Lab, the
Beijing Academy of Artificial Intelligence, the Shanghai Artificial Intelligence
Lab, and the Peng Cheng Laboratory.

  


  
  
   

 
          

          

          

          
          
          

As can be seen in Table 5, KTH Royal Institute of Technology is clearly the largest
producer of AI publications in Sweden, but it is still smaller than all top five in-
stitutions in the other countries. Chalmers University of Technology shows the
strongest positive trend and Luleå University of Technology has the highest share of
AI publications. It also has the most international AI research. Finally, Chalmers
University of Technology has the greatest share of co-publications with the cor-
porate sector in Sweden, also in comparison with all academic institutions in this
section. A longer list of Swedish institutions is available in the Appendix.
Massachusetts Institute of Technology is the largest AI publication producer in
the United States whereas Carnegie Mellon University has the highest share of
such publications and shows a relatively modest positive trend, see Table 6. The
citation impact is very high for all five. All academic institutions in the United
States collaborate extensively with corporate actors within AI, some of which are
listed below.
Three main US competence centres stand out: Stanford, powered by Google, lit-
erally created the modern AI research field and the ecosystem around Stanford
is vibrant; MIT; the University of Washington, powered by Microsoft.

   


  
  
   

 
          
 
          
 
          
          
          

          

IBM has the highest number of AI publications, whereas Tencent’s publication

volume grew 80% annually during these five years, which is a substantially higher
rate than that of the other corporate actors, see Table 7. Tencent has also the

   


  
  
   

 
          
          
          
          
          
          

highest share of AI publications. The citation impact of all corporate actors is

very high.
Tencent has the highest share of international AI co-publications as well as publi-
cations including academic actors. A high share indicates extensive collaboration
with the academy, but it could also mean that the actor has a somewhat limited
capacity or willingness to carry out in-house research leading to publications.


        
    
Scientific co-publications between countries or actors provide a good indicator
of their collaborations. In Figure 10 – Figure 13, the development over five years
for the five largest collaborations for each of the countries in focus is presented.
To facilitate this comparison, the volumes have been normalised using the total
number of international AI publications each year. For example, this means that
in 2020, approximately 30% of the international Canadian AI publications in-
cluded at least one US co-author.
     







 












    

Most international collaborations are bilateral. Every third international AI pub-

lication with an author with a Canadian affiliation also includes an author with
a Chinese or US affiliation, see Figure 10. However, only 225 of the 1,774 co-
publications with China during the period included a US author. France, ranked
eleventh in the world in terms of AI publications, shares a language with Canada,
whereas Iran, ranked thirteenth, does not.
China has a relatively low share of international AI co-publications at 25%.
Among them, the United States clearly dominates, even though the share has
dropped slightly since the peak in 2017, see Figure 11. Given the rapid growth
in AI publications, the number of co-publications including China and the

    










 



 






    

United States has increased in absolute terms. Australia is ranked ninth in AI publication
volumes and Hong Kong 22nd.

  






 


 





    

 Whereas Sweden has relatively similar shares of co-publications with four coun-
tries, see Figure 12, the other three countries’ collaborations are dominated by
the United States and China. AI collaboration with China has varied considerably
in intensity over the years and overall. Given that the share of international AI
co-publications has been relatively stable at 60% during this period, Sweden’s
collaboration appears to have become slightly more dominated by some of the
largest AI countries in the world.
   









 



 






    

From a US perspective, AI collaboration with China has increased over this pe-
riod, potentially peaking in 2019, see Figure 13. Four of ten international AI co-
publications with a US author also include a Chinese one. Thereafter follow four
other large AI countries with considerably lower shares.
A closer look at the collaborations at the five largest AI institutions in Sweden
reveals further details as to the nature of these collaborations. Do they collaborate
with the leading universities and countries? KTH Royal Institute of Technology
had the highest number of publications and in Figure 14, its top ten collaborators
are listed. The figure at each bar indicates the number of co-publications during
2016 –2020 and the length of the bar relates this number to all AI publications
at KTH Royal Institute of Technology. Blue bars indicate national collaboration
and red bars international collaboration.

    


      

 

  

 

 

 

  

 

 

 

  

Interestingly, the top collaborator is a medical university, with 25 co-publications.

However, this only represents about 5% of all AI publications at KTH Royal In-
stitute of Technology. The second and the final one in the list are probably both
related to AI in automotive applications.

   



     

 

 

 

 

 

 

  

 

 

 


 Like KTH Royal Institute of Technology, Chalmers University of Technology
has a main collaborator of a complementary type in the same city, see Figure 15.
The University of Gothenburg participates in almost every fifth AI publication.
International partners are at the lower end of the list. AI for automotive applica-
tions is also prominent for Chalmers University of Technology, with two auto-
motive corporates among the top ten collaborators. In Scopus, Volvo includes
both AB Volvo and Volvo Cars.
Luleå University of Technology features a clearly different set of top collaborators
within AI, see Figure 16. Countries with emerging academic research are repre-
sented at the top, including Vietnam and Bangladesh. All collaborators are in-
ternational and some of them represent a relatively large share of all AI-related
publications.

  



      
 

 

 

 

  

 

 

 

 

 

Linköping University also has a very international set of top collaborators, but
they represent only a very small share of the total volume, see Figure 17. The In-
ception Institute of Artificial Intelligence is based in Abu Dhabi and the Com-
puter Vision Centre in Barcelona, Spain.


   


         

 


 

 

 

 

  

 

 

 

  

Finally, the AI co-publications at Uppsala University comprise Swedish and

European institutions, each with a relatively low share of the total volume, see
Figure 18. As is the case for KTH Royal Institute of Technology, Karolinska
Institutet is the largest collaborator.

    



      

 

 

 

 

 

 

 

 

 

 


 Overall, their collaboration patterns do not indicate that Swedish universities
have established strategic international collaborations within AI with leading ac-
tors abroad. A report by Sweden’s innovation agency (Vinnova) and others rec-
ommended the establishment and development of internationally leading
environments for collaboration (Vinnova, 2018).
Geopolitically, it is interesting that the controversies between China, Canada and
the United States are not more visible in the development of research collabora-
tion. As previously mentioned, the number of co-authors in AI publications is
relatively low; almost all have a maximum of ten co-authors, which means that
the collaborations are probably fairly balanced with active participation in
both/all countries. The personal relationships on which these collaborations are
based can most likely explain the continuation of collaborative work despite pan-
demic travel restrictions and increased technological competition between the
United States and China. When interviewed for this report, one of the authors
of China’s national AI strategy noted that research exchange between China and
the United States in the field was still “business as usual”, but also expressed con-
cern that Chinese AI students overall no longer were being granted US student
visas. Conjecturing how the currently shifting geopolitical landscape will affect
future collaborations is therefore a challenge.
Canada and France are working with the international community to create the
International Panel on Artificial Intelligence (IPAI) to support and guide the re-
sponsible development of AI that is grounded in human rights, inclusion, diver-
sity, innovation, and economic growth. The Panel will be an organisation
dedicated to AI and will bring together many of the leading global AI experts. It
will form a global reference point on AI, fostering international collaboration
and coordination on AI policy development. Members include Australia, Canada,
France, Germany, India, Italy, Japan, Mexico, New Zealand, the Republic of
Korea, Singapore, Slovenia, the United Kingdom, and the United States. The
European Union is also in negotiations to enter the partnership.
US AI policy also recognises the importance of international cooperation on AI.
The American AI Initiative recognises partnerships with US allies; partners rep-
resent a key “source of strategic competitive advantage,” and the Initiative iden-
tifies the need to “engage internationally to promote a global environment that
supports American AI research and innovation and opens markets for American
AI industries.”


China also emphasises the importance of international cooperation in the AI
field. There are several reasons for this. On the one hand, it is for the good of
the overall development of the AI field, e.g., in relation to ethical issues and the
safe development of AI as well as the utilisation of the complementary strengths
of different nations. On the other hand, it is an effort to ensure that China has
access to aspects of AI development where it considers itself to still be lagging
(behind the United States), mainly in advanced chip technology, basic research,
and education.


    
By now most large economies have drafted and released some sort of national AI
policy or strategy. Since Canada published the world’s first national AI strategy
in 2017, more than 60 other countries and regions have published similar doc-
uments. As in all strategies, there are trade-offs and priorities. PwC (2019) men-
tions three main types of trade-offs, or rather degrees of emphasis in these
national policies: innovation versus regulation, the individual versus the state,
and transparency versus vulnerability. Here we summarise the key aspects of the
national AI strategies (or equivalent) of the four countries studied above, as well
as that of Singapore, which is of interest as it is a small, scientifically advanced
nation with good relations to both the United States and China, and that of the
European Union, which is of obvious importance to Sweden.

  
Canada’s long-term investments in AI have positioned the country as one of the
leading national innovation systems in the area. Radical Ventures (2021) describes
Canada’s AI pedigree as based on decades of research funding. Research funding
agencies such as the Canadian Institute for Advanced Research (CIFAR) and the
Natural Sciences and Engineering Research Council of Canada (NSERC) pro-
vided early funding for the work of AI pioneers Geoffrey Hinton, Yoshua Bengio,
and Richard Sutton. These individuals’ laboratory breakthroughs and subsequent
collaborations with private companies, including translation into real-world ap-
plications, were the direct precedents of the vibrant innovation ecosystems which
gravitated towards Canada’s AI luminaries.
Focused government investment, paired with commitments from the private sec-
tor, anchored this community in Canada, establishing centres of excellence with
the shared mission of attracting world-leading research talent. Today, the Vector
Institute for Artificial Intelligence (Vector Institute), the Quebec Artificial Intel-
ligence Institute (Mila) and the Alberta Machine Intelligence Institute (Amii)
are major hubs in the hyper-competitive global AI talent landscape.
Canada had, and still has, a choice between either building an AI ecosystem or
keeping buying AI technology from the United States. The government promotes
and funds the ecosystem approach. The Canadian government has managed to
recruit and retain a fair number of AI experts across its departments, which is re-
garded positively by researchers and industry alike. In recent years, the Canadian
national strategy involves emphasising innovation via AI ‘superclusters’ to attract
private funding and talent, with great hopes that knowledge transfer from the ac-

ademic sector to these superclusters will speed up the rate of innovation. In 2017,
the Canadian government created the Innovation Superclusters Initiative (ISI).
The combined funding from private sector partners (US$ 700 million) and the
federal (US$ 230 million) and provincial (US$ 60 million) governments was ap-
proximately US$ 990 million. Three central positive results from government
policies and strategies have been facilitating the speedy immigration of top inter-
national talent, unlocking valuable public data, and incentivising private sector
investments in AI technologies and startups. In Canada’s 2021 budget, commer-
cialisation was added as a new element for all AI funding.
One interesting aspect of the Canadian AI policy landscape involves ethics. Canada
is currently involved in leading a global online consultation on AI ethics spearheaded
by Mila and the Algora Lab (Université de Montréal) in partnership with UN-
ESCO. This builds on the 2017 Montreal Declaration for a Responsible Develop-
ment of Artificial Intelligence which articulated ten fundamental values to underpin
ethics principles for AI. Canada also co-leads the Global Partnership on AI (GPAI)
with France. This is the first global initiative bringing together 14 countries and
the European Union to guide the responsible use and development of AI.
The Canadian government has in recent years focused funding into three research
institutes that in turn determine which universities to fund. This is seen as a sig-
nificant step away from the spray-and-pray mentality of the past. The Vector In-
stitute, Mila and Amii are major hubs in the hyper-competitive global AI talent
landscape. Mila has some 50 professors and in total about 700 academics and is
looking to develop pre-competitive stage consortia.
Despite Canada’s successful positioning in AI, global competition is ramping up
quickly. Radical Ventures (2021) highlights the competition via recent results
from the Tortoise AI Index. In the 2020 updates of this international bench-
marking ranking state capacity for AI, Canada held the fourth place on the global
stage, but the country has significantly shifted down the ranks in multiple cate-
gories across implementation, innovation and investment. Most notably, Canada’s
operating environment fell from fifth to thirty-second place. Additionally,
Canada slipped nine places in terms of development, a category requiring indus-
try collaboration to support contributions to new models, techniques and prod-
ucts. The report also cites dwindling incentives to seek patents as a contributing
factor in slipping down the rankings. Industry experts claim that larger Canadian
companies are paying inadequate attention to AI – as far as AI is concerned, these
companies are quite immature and unprepared for data and skills.


 
While several policies and white papers relevant to AI exist, the key national AI
policy document is the Next Generation Artificial Intelligence Development Plan
(NGAI), which was launched in 2017 by the Chinese State Council. The plan
was developed in a project initiated a year earlier by the Chinese Academy of En-
gineering (CAE). A large group of stakeholders from academia, industry, and
government led by the Ministry of Science and Technology (MoST) contributed
to the drafting of NGAI.
Some of the challenges China faces in the development of AI identified in the
NGAI include basic AI research, high-end chips, a well-developed ecosystem for
AI development, and talent supply, i.e., AI education. In parallel to technology
development aims, the NAGI emphasises the importance of the ethical develop-
ment of AI and that China is to become a leader in the development of ethical
norms and standards for AI.
Despite being launched in 2017, the NGAI plan covers the period 2015–2030
and outlines three consecutive steps which all have specific target outcomes:
– By 2020, to have closed the technological gap with world-leading countries
and for China’s AI industry to compete at the highest international level.
– By 2025, to achieve significant breakthroughs in fundamental AI research, to
have developed some world-leading AI technologies and applications, and for
AI to have become the main driving force for industrial development and eco-
nomic transformation in China.
– By 2030, to have become one of the world’s leading AI innovation centres and
to have developed comprehensive regulatory, legal, and ethical principles for
AI development.
As part of the implementation of the plan, a range of national AI platforms has
been established. These include fifteen application-driven National Open Inno-
vation Platforms supported by MoST. Each platform focuses on a specific appli-
cation area and is led by a leading company in that area, e.g., Urban Cognition
led by Alibaba, Image Perception led by Megvii and Autonomous driving led by
Baidu. By 2023, twenty National NGAI Development Experimental Zones sup-
ported by MoST will have been established, as well as several Pilot Zones for In-
novation Application of AI supported by the Ministry of Industry and
Information Technology. The aim of the National NGAI Development Experi-
mental Zones is to push cooperation with local governments to implement AI
technologies in practice, both to demonstrate its usability as well as to experiment
with AI policy development.

In addition to the national AI development plan, local governments in China
have developed, or are developing, their own AI development plans, which play
an important role in delivering on the goals of the NGAI in practice. While co-
ordinated by MoST through the AI Plan Promotion Office, the implementation
of the NGAI can be characterised as quite distributed; a broad range of stake-
holders act largely autonomously to meet the targets of the plan.

 
Prior to the launch of Singapore’s National AI Strategy, in 2017 the government
established AI Singapore, a programme funded by the National Research Founda-
tion, which supports AI development through a range of initiatives and aims to
pool the resources of Singapore’s research institutions and private sector. The Na-
tional AI Strategy was launched in 2019 with the vision to, by 2030, have estab-
lished Singapore as a “leader in developing and deploying scalable, impactful AI
solutions, in key sectors of high value and relevance to our citizens and businesses.”
Through the plan Singapore aims to become a global hub for AI development and
deployment with the purpose of generating economic gains and improving lives.
Part of the plan aims to ensure that Singaporeans understand AI, its uses and ben-
efits as well as to upskill the workforce to enable its participation in the AI economy.
An initial set of five national AI projects was established at the launch of the Na-
tional AI Strategy, focusing on areas that were expected to deliver strong social
and/or economic impact. The approach is for the national projects to guide in-
vestment in AI research, generate lead demand and develop necessary human re-
source and capabilities. The number of national AI projects has since expanded
to seven. These are in the fields of government, healthcare, smart estates, educa-
tion, border security, logistics, and finance.
In parallel, the National AI Strategy aims to develop and support the AI ecosys-
tem, which is done by focusing on five key ecosystem enablers: triple helix part-
nerships between the research community, industry, and government; talent and
education; data architecture (to en-ables quick and secure access to high-quality
datasets); establishing a progressive and trusted environment for test-bedding,
developing, and deploying AI solutions; and international collaborations.
Singapore’s National AI Strategy is implemented through a range of programmes
aiming at supporting businesses’ AI transformations, investments in AI develop-
ment as well as AI edu-cation. The website of the Prime Minister’s Office also
has a form through which anyone can submit proposals to the existing national
projects or propose a new one.

  
The White House sees AI as a prioritised strategic area. The Trump administra-
tion stated that each agency should put out an AI strategy and there is also strong
continuity between the Trump and Biden administrations on AI. The National
Artificial Intelligence Initiative Act 2021 established the White House National
Artificial Intelligence Initiative Office charged with coordinating the national AI
strategy. The National Security Commission on Artificial Intelligence estimates
that the United States needs to spend US$ 32 billion over the next few years to
win the AI race with China, among other rivals.
The 2021 National Defense Authorization Act further develops AI policy in the
defence and non-defence sectors. The National Security Commission on Artificial
Intelligence Final Report, published in 2021, outlines several concrete actions
for the president including defining and prioritising the key emerging technolo-
gies such as quantum computing in which US leadership is essential. It also de-
tails organisational changes in the responsibilities for AI within government
agencies, and states explicitly how China’s influence is to be curbed.
A major part of the research funding strategy for AI in the United States rests on
the US National Science Foundation (NSF). In 2021, the NSF announced the
establishment of eleven new NSF National Artificial Intelligence Research Insti-
tutes, building on the first round of seven institutes funded in 2020. The com-
bined investment of US$ 220 million, building upon the first selection of seven
institutes and US$ 140 million in allocated funding in 2020, expands the reach
of these institutes to include a total of 40 states and the District of Columbia.
Through the NSF-led AI Research Institutes, as well as a range of ongoing pro-
grammes, the NSF supports fundamental research, education and workforce de-
velopment, and advanced, scalable computing resources that collectively enhance
fundamental research in AI.
According to technology consultant Deltek (2021), identifiable federal spending
on AI rose to nearly US$ 1 billion in the 2020 financial year, up 50% from 2018,
making it one of the fastest growing emerging technology investment areas. The
Department of Defense spends nearly twice the amount the civilian sector spends
on AI.


In May 2018, the Swedish government adopted a national focus on AI, which
for example emphasises the importance of disseminating and using AI to


strengthen competitiveness. This is not a strategy but a document that sets out
the direction of Swedish AI efforts and lays the foundation for future priorities.
The policy goals outlined are similar to the digitisation goals and the Swedish
government believes that the document complements the digitisation strategy.
The policy approach has not been to single out AI as a separate strategy, but
rather to regard this area as part of a broader need to digitalise considerable parts
of society. The Swedish government identifies four areas that are assumed to be
particularly important to utilise AI: education, research, innovation and use, as
well as frameworks and infrastructure.
Due to the involvement of various stakeholders in national AI policy, the situa-
tion in Sweden is complex. The ministries concerned are mainly the Ministry of
Infrastructure, the Ministry of Trade and Industry and the Ministry of Education.
Implementing authorities include Vinnova and the Swedish Digital Administra-
tion Authority (DIGG). Vinnova has launched a national AI hub called AI Swe-
den. The Research Institutes of Sweden (RISE) has gathered the players in the
Swedish AI ecosystem to develop an AI agenda for Sweden. Research on AI is
mainly conducted at six universities: the University of Gothenburg, KTH Royal
Institute of Technology, Linköping University, Lund University, Umeå University,
Örebro University and Chalmers University of Technology.
The largest Swedish funder of AI research is not the state, but the private Wal-
lenberg Artificial Intelligence, Autonomous Systems and Software Program
(WASP). The Knut and Alice Wallenberg Foundation will have invested SEK
4.2 billion in AI research by 2030. Together with co-funding from universities
and companies, the entire investment covers SEK 5.5 billion. The programme
funds basic research on AI and autonomous systems. WASP aims to bring to-
gether various higher education institutions through financial support to build
internationally competitive AI research environments. WASP was established in
2015 and has since 2017 been supplemented with an expanded AI investment
in machine and deep learning and next-generation AI, as well as basic mathe-
matical principles behind AI and robotics and visual recognition. Funding has
been allocated for basic research at Chalmers and KTH Royal Institute of Tech-
nology as well as Linköping, Lund and Umeå universities. WASP also attracts
global AI expertise to Sweden through key recruitments. The goal is to create
100 AI research groups. Nine international AI professors have been recruited. In
collaboration with industry, the programme has started a graduate school, opened
research arenas and launched test beds. WASP has also recruited 337 doctoral
students to the doctoral school, 101 of whom are industrial doctoral students.


  
Sweden’s national strategy must be seen in the light of policy development in the
European Union. The European Union approach to AI focuses on two areas: ex-
cellence in AI and trustworthy AI. The main underlying philosophy is that AI
improvements should be based on rules that safeguard the functioning of markets
and the public sector, and people’s safety and fundamental rights.
The effort to develop an AI-strategy started in earnest in 2017. In May 2017,
the European Commission published its mid-term review of the Digital Single
Market strategy. It then highlighted the importance of building on Europe’s sci-
entific and industrial strengths, as well as on its innovative start-ups, to be in a
leading position in the development of AI technologies, platforms, and applica-
tions. Later in 2017, policy statements were made on the sense of urgency in ad-
dressing emerging trends such as AI while at the same time ensuring a high level
of data protection, digital rights and ethical standards. The Commission was in-
vited to present a European approach to AI. The European Parliament has made
wide-ranging recommendations on civil law rules on robotics and the European
Economic and Social Committee has also issued an opinion on the topic.
In April 2021, the European Commission published its AI Package, proposing
new rules and actions to turn Europe into the global hub for trustworthy AI.
This package consists of:
– a Communication on Fostering a European Approach to Artificial Intelligence;
– the Coordinated Plan with Member States: 2021 update;
– a proposal for an AI Regulation laying down harmonised rules for the European
Union (Artificial Intelligence Act).


 
One clear observation on the global arena of AI research is that the United States
and China, however fiercely they compete with each other, clearly are advancing
fast together and leaving other countries behind, both in terms of research and the
pace of innovation. A combination of strong technology sectors and high govern-
ment interest and support in these two countries will further widen this gap.
EU policy and opinion makers fear that European countries are losing ground
to the United States and China in the field of AI. This is one of the rationales
behind the European Union’s policies on AI. The US government teams up with
large software companies to set the tone in its AI development, and while the
Chinese government is working to drive the AI development in China it too
turns to private enterprises for implementation, e.g., through the National Open
Innovation Platforms. In the United States, large corporations control the data,
while in China the state ultimately controls the data. In China and in the United
States, as well as in other nations, there exist strong interest from the military to
utilize AI, though the amount of military AI research spending in China noto-
riously difficult to estimate (CSET 2019).
The US approach is that other national policies – China’s above all – seek to erect
barriers to the free and open development of AI, appropriating the benefits for
their national champions and applying AI as a geopolitical lever. Such policies
could distort the development and benefits of AI for humanity, making the world
less secure for the United States and its allies, and markets less receptive to US
products and services. The second perceived challenge for the United States comes
from other governments whose AI policies could lead to prescriptive regulation
that may stifle AI innovation and discriminate against US technology firms. West-
ern thinktanks such as the Brookings Institute point out that AI will continue to
exert both immense stress on the US – China relationship as well as create oppor-
tunities for potential collaboration. There will be a higher degree of scrutinising
US – Chinese academic exchanges from both these countries and their allies.
China certainly desires to become a world leader in the AI field and looks to de-
velop and support national champions (notwithstanding recent crackdowns on
major domestic technology behemoths by the Chinese government). Authors of
the Chinese AI strategy prefer to point out that, while it focuses on the develop-
ment of AI in China and the path to the international forefront, the United
States’ AI strategies seem more focused on countering Chinese developments. It
could be argued that as the country catching up it is natural that China focuses


 less on specific US achievements and rather concentrates on further developing
AI. However, there is of course an intense focus in China on the United States’
increased restrictions on access to pertinent technologies, with a strong mobili-
sation to develop these areas domestically to achieve technological independence.
Within this context it would be interesting if countries such as India, Canada,
Sweden and the United Kingdom could offer a counterbalance to the US or Chi-
nese models. While striking partnerships with the leading United States and
China is the most attractive, smaller specialised labs in other countries can offer
a complementary way forward. Smaller, less developed countries must invest with
the aim of becoming valuable niche players in the global arena, as well as of de-
veloping strategies on the best ways to navigate the intense technology rivalry
between the United States and China.


 
Research on AI is growing fast. AI research, as measured by the number of scientific
publications, is growing rapidly in absolute as well as relative terms. In a compar-
ison of several countries with substantial AI research, countries such as China,
India, Singapore and Iran have a high share of AI publications in their portfolio
whereas Sweden and Denmark have a relatively low share. Globally, about 57% of
AI publications are within the Natural Sciences, followed by Engineering and Tech-
nologies (27%).
China and the United States are at the research frontier and are both the largest com-
petitors and collaborators. At an institutional level, the top five Chinese universities
have larger volumes of AI publications than all institutions in the other studied
countries (Canada, Sweden, the United States). The largest corporate producers
of AI publications are in the United States. China and the United States dominate
as partners in the international collaboration for all four countries, but less so for
Sweden, which has almost as high shares of international co-publications with the
United Kingdom and Germany.
Dedicated national AI strategies are prevalent in most advanced economies. Over 60
countries have adopted some form of national AI policy, indicating the perceived
policy importance of AI in society. Canada’s long-term investments in AI have po-
sitioned the country as one of the leading national innovation systems in the area.
Canada had, and still has, a choice between either building an ecosystem for AI or
continue buying AI technology from the United States. The Canadian government
promotes and funds the ecosystem approach and is building a critical mass of
home-grown and recruited talent in the field. In China, the State Council has been
driving a national strategy for AI with ambitious goals to become a worldleader
both in research and industrial applications. Singapore’s national AI strategy aims
to develop and support an AI ecosystem, focusing on triple helix partnerships be-
tween the research community, industry, and government. In the United States,
the White House sees AI as a prioritised strategic area and is set to increase funding
in what resembles a new space or arms race, with both multinational corporations
and governments striving for global leadership.
Sweden has strengthened research on AI, but international collaboration is still weak.
Sweden produces AI research with a high citation impact in international com-
parison and the research volume exhibits strong growth. The five largest producers
of AI publications in Sweden (KTH Royal Institute of Technology, Chalmers Uni-
versity of Technology, Luleå University of Technology, Linköping University and
Uppsala University) exhibit different patterns of co-publications, with local/re-

 gional partners dominating in three cases and international in two. Sweden’s sci-
entific profile is similar to the global one, but with a slightly higher share of AI
publications in Medicine. The international collaboration pattern indicates that
these institutions have so far not established strong international partnerships
within the area of AI.
Sweden’s national AI policy focuses on adoption rather than research excellence. The
Swedish government has chosen to focus not so much on singling out AI as a sep-
arate policy field, but rather on including AI in policies to advance digitalisation
in industry, academia, and government in general. Sweden has the ambition of
maximising the potential of AI in areas such as education, research and innovation,
as well as overall adoption in society. The emphasis is on the quick adoption of AI
in society rather than on creating the capabilities to develop world-class AI systems.
As neither a large producer of AI research nor a scientific leader, Sweden should
have a strategy on how to best position itself to ensure access to and implementa-
tion of the leading edge of global AI advancements. This would include fostering
an AI research and innovation ecosystem that is sufficiently strong to be able to
attract leading scientists, innovators, and companies in the AI field to collaborate
with, and/or work in Sweden. With rapidly increasing tensions and competition
between the United States and China, Sweden needs to support its scientists in-
terested in working with both countries in navigating that rivalry.




Amiraux, Valérie, Vice-rectrice aux partenariats communautaires et interna-
tionaux, Universite de Montreal (May 6, 2021)
Ema, Arisa, Associate Professor, University of Tokyo, Japan (April 19, 2021)
Gianchandani, Erwin, Deputy Assistant Director, Computer and Information
Science and Engineering, National Science Foundation, USA (August 11, 2021)
Joanette, Yves,Vice-recteur adjoint Recherche, découverte, création et innova-
tion, Universite de Montreal, Canada (May 6, 2021)
Létourneau, Stéphane, Vice-président exécutif, MILA, Canada (May 17, 2021)
Lindsjö, Göran, Consultant, Governo, Sweden (May 17, 2021)
Meserole, Chris, Ph.D. Research and Policy Director, AI and Emerging Tech
Initiative, Fellow, Foreign Policy Program, The Brookings Institution, USA
(June 5, 2021)
Rentzhog, Magnus, Desk Officer, Government Offices of Sweden, Ministry of
Enterprise and Innovation, Sweden (November 18, 2021)
Veerman, Alan, Chief Operations Officer, Vector Institute, Canada (April 22,
2021)
Wu, Fei, Professor; Vice Dean, College of Computer Science; and Director,
Institute of Artificial Intelligence; Zhejiang University, China (April 4, 2021)


CSET (2019) Chinese Public AI R&D Spending: Provisional Findings, see
https://cset.georgetown.edu/publication/chinese-public-ai-rd-spending-
provisional-findings/
Deltek (2021) Federal Artificial Intelligence Landscape 2022, see
https://info.deltek.com/federal-artificial-intelligence-2022
Elsevier (2020) Artificial Intelligence: How knowledge is created, transferred,
and used Trends in China, Europe, and the United States, see
https://www.elsevier.com/?a=823654


 PWC (2019) Gaining Competitive Advantage through Artificial Intelligence
(AI): Policy Making & National AI Strategies, see
https://www.pwc.lu/en/advisory/digital-tech-impact/technology/gaining-
national-competitive-advantage-through-ai.html
Radical Ventures (2021) 2021 Primer: Canada’s AI Research Ecosystem, see
https://radical.vc/2021-primer-canadas-ai-research-ecosystem/
Tillväxtanalys (2021) AI-politik för konkurrenskraft, see
https://www.tillvaxtanalys.se/publikationer/pm/pm/2021-07-06-ai-politik-for-
konkurrenskraft.html
Vinnova (2018) Artificiell intelligens i svenskt näringsliv och samhälle: Analys av
utveckling och potential, Vinnova VR 2018:08



Swedish institutions with AI publications according to Elsevier’s method (2020).


  

  

  
  
  
  
  
  
  
  
  
  
  

37










