Comparison of National Artificial Intelligence Strategies to Promote
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Imprint Editor: Konrad-Adenauer-Stiftung e. V. 2019, Berlin Authors: Dr. Olaf J. Groth, CEO & Managing Partner Dr. Mark Nitzberg, Principal & Chief Scientist Dan Zehr, Editor-in-Chief Tobias Straube, Project Manager and Senior-Analyst Toni Kaatz-Dubberke, Senior-Analyst Franziska Frische, Analyst Maximilien Meilleur, Analyst Suhail Shersad, Analyst Cambrian LLC, 2381 Eunice Street, Berkeley CA 94708-1644, United States https://cambrian.ai, Twitter: @AICambrian Editorial team and contact at the Konrad-Adenauer-Stiftung e. V. Sebastian Weise Consultant for Global Innovation Policy Post: 10907 Berlin Office: Klingelhöferstraße 23 T +49 30 / 269 96-3516 10785 Berlin F +49 30 / 269 96-3551 [Note: This English-language version was translated from the original publication in German.] Cover image: © sarah5/Mlenny (istockphoto by Getty Images) Images: © p. 11: Matthew Henry, p.16: Manuel Cosentino, p. 23: Dan Gold, p. 30: David Rodrigo, p. 42: Victor Garcia (unsplash); p. 36: byheaven (istockphoto by Getty Images) Design and typesetting: yellow too Pasiek Horntrich GbR The print edition was produced in a carbon-neutral way at Druckerei Kern GmbH, Bexbach, and printed on FSC-certified paper. Printed in Germany. Printed with the financial support of the Federal Republic of Germany. This publication is licenced under the terms of “Creative Commons Attribution-Share Alike 4.0 International”, CC BY-SA 4.0 (available at: https://creativecommons.org/licenses/ by-sa/4.0/) legalcode.de). ISBN 978-3-95721-553-6
Comparison of National Strategies to Promote Artificial Intelligence Part 2 Dr. Olaf J. Groth, CEO & Managing Partner Dr. Mark Nitzberg, Principal & Chief Scientist Dan Zehr, Editor-in-Chief Tobias Straube, Project Manager and Senior-Analyst Toni Kaatz-Dubberke, Senior-Analyst Franziska Frische, Analyst Maximilien Meilleur, Analyst Suhail Shersad, Analyst
Table of Contents Preface 4 Background and Definitions 5 Summary 6 Cambrian AI Index © 10 Cambrian AI Index © of countries from part 2 of the study 10 Cambrian AI Index © of countries from part 1 and 2 of the study 10 Canada 11 I.) Introduction 11 II.) Requirements for AI 11 III.) Institutional framework 12 IV.) Research and Development 12 V.) Commercialisation 13 Japan 16 I.) Introduction 16 II.) Requirements for AI 17 III.) Institutional framework 17 IV.) Research and Development 18 V.) Commercialisation 20 Israel 23 I.) Introduction 23 II.) Requirements for AI 24 III.) Institutional framework 24 IV.) Research and Development 24 V.) Commercialisation 25 United Arab Emirates 30 I.) Introduction 30 II.) Requirements for AI 31 III.) Institutional framework 31 IV.) Research and Development 31 V.) Commercialisation 32
India 36 I.) Introduction 36 II.) Requirements for AI 36 III.) Institutional framework 37 IV.) Research and Development 37 V.) Commercialisation 39 Singapore 42 I.) Introduction 42 II.) Requirements for AI 43 III.) Institutional framework 43 IV.) Research and Development 43 V.) Commercialisation 44 Methodology of the Cambrian AI Index © 49 Annexes 58 Annex 1: Overview of resources and research areas of AIST, RIKEN and NICT (Japan) 58 Annex 2: Overview of the research areas of the Fundamental Research Programme of the AI Singapore Initiative 60 Bibliography 61 Acknowledgements 77 The Authors 78
Preface Technologies can rarely be reduced to their mere commercial added value. The history of the Industrial Revolution teaches us that nation states have always endeavored to build or maintain political supremacy through pioneering economic achievements. In the age of digital upheavals, multiple disruptions and immense acceleration, this dictum still applies. The topic of “artificial intelligence” plays a special role here – a technology that is currently being discussed worldwide and increasingly applied. As with any new tech- nology, both the Cassandrian pessimists (such as Steven Hawking or Elon Musk) and the progress optimists (Mark Zuckerberg, Eric Schmidt or Bill Gates) contribute their theses on the future development of humanity. They widely vary from dark dystopia to paradisiacal future prospects. Let us hope that the recently appointed Enquete Commission of the German Bundes tag will counter an agitated and possibly overheated debate with sober stocktaking. Germany’s AI strategy, which is expected by the end of the year, must define quanti- fiable goals and concrete measures, which will then be vigorously implemented. The political crash barriers necessary for the use of automated machine learning need to be established. Other countries are well ahead in this respect. They have long since defined AI strate- gies, developed business models and subjected ground-breaking applications to initial practical tests. Therefore, it is worth taking a close look at how other economies are dealing with the digital revolution: What regulatory framework conditions have they defined? How do they implement policy strategies and programs to create new indus- trial policy facts? With this two-part publication, the Konrad Adenauer Foundation intends to give a com- parative overview of the AI strategies of major national economies in order to provide food for thought and inspire the German debate. We believe: “Tech is politics” – and pol- itics and civil society should give this more attention and discuss this more vigorously. I hope you find this publication an inspiring read. Yours Dr. Gerhard Wahlers Dr. Gerhard Wahlers is Deputy Secretary General and Head of the Principal Department for European and International Cooperation of the Konrad-Adenauer-Stiftung. 4
Background and Definitions Artificial intelligence: This overview, however, focuses on the analysis of rapid developments and designs of the AI framework conceptss in six countries and how initial conceptual AI frameworks. they deal with the revolutionary potential of Artifi- cial Intelligence. In July 2018, the German federal government published a key issues paper on the German The following definition is used as a basis strategy with regard to artificial intelligence (AI) for the terminology: and there they acknowledge: “Artificial intelli- “In the broadest sense, artificial intelli- gence has reached a new stage of maturity in gence is the ability of machines to learn, recent years and is becoming a driver of digitiza- to think, to plan and to perceive; i. e. the tion and autonomous systems in all spheres of primary qualities that we identify with life.”1 Therefore, the state, society, the economy, human cognition. This ability is achieved the government, and science are urged to con- by digital technologies or digital-phys- sider artificial intelligence in depth and to deal ical hybrid technologies, which imitate with its chances and risks. the cognitive and physical functions of humans. For that purpose, AI systems A comprehensive German AI strategy was pre- do not only process data, they recognize sented at the Digital Summit in December 2018. patterns, draw conclusions, and become The aim is to prominently embed the topic of AI more intelligent over time. Their ability to in the digital policy of the federal government. adopt and refine newly developed skills In this way, Germany is catching up with a large has improved significantly since the turn number of countries which in recent years have of the century. This also means that what seen extensive initiatives for AI strategy finding is referred to as AI changes with each processes.2 major technological breakthrough, and the definition must therefore be periodi- These strategies are motivated by partly spec- cally adjusted.” tacular progress in research and application of AI systems, based on techniques of Machine Learning (ML) as well as its subdiscipline of Deep 1 Cf. German Federal Government, Key Points of the Federal Learning (DL) and its large varieties of neuronal Government for an Artificial Intelligence Strategy (July 2018), https://www.bmbf.de/files/180718%20 Eckpunkte_ networks. KI-Strategie%20final%20Layout.pdf. 2 To this end, see the OECD overviews, for example, The global relevance of AI technologies is dem- http://www.oecd.org/going-digital/ai/initiatives-worldwide/, onstrated by their prominent representation on Future of Life Institute, https://futureoflife.org/ai-policy/, the Smart Data Forum, https://smartdataforum.de/ en/ this year’s international agenda – from the Munich services/international-networking/international-ai- Security Conference in February, to the presenta- strategies/, Charlotte Stix, https://www.charlottestix.com/ tion of the EU Commission’s3 AI paper in April, to ai-policy-resources, und Tim Dutton, https://medium. com/politics-ai/an-overview-of-national-ai-strategies- the joint AI declaration of the G7 states in Canada 2a70ec6edfd, all last retrieved on 17.9.2018. in June (“Charlevoix Common Vision for the Future 3 Cf. European Commission, Artificial Intelligence for of Artificial Intelligence”).4 Europe (April 2018), http://ec.europa.eu/newsroom/dae/ document.cfm?doc_id=51625. The role of Artificial Intelligence as a potential key 4 See Canadian G7 Presidency, Charlevoix Common Vision for the Future of Artificial Intelligence (June technology of dystopian future concepts, social 2018), https://g7.gc.ca/wp-content/uploads/2018/06/ control, and autocratic world power fantasies is FutureArtificialIntelligence.pdf. also increasingly finding its way into public debate. 5
Summary The following summary describes the insights and assessments of the status quo of national strate- gies for the promotion of Artificial Intelligence (AI). The analysis of the six countries from part 1 of the study and the findings from the countries examined in this part (Canada, Japan, Israel, United Arab Emirates, India and Singapore) serve as a basis. Against the background of the results of the first two parts of the study, there is also a third part in preparation which focuses exclusively on the analysis of the recently published German AI strategy and the recommendations for action that follow it. Policy fields for AI: In the European countries, become more and more evident. In Japan and the economic potential of AI is considered above South Korea, it is also the globally positioned all, while in Japan AI is seen as part of the inevita- hardware-oriented conglomerates that are driving ble next stage in human development, which will AI forward. In the countries of continental Europe encompass all areas of life: Society 5.0. In Israel, (including Germany), which focus on science and the security policy dimension plays a central role; the protection of individuals, there is currently a an aspect that other countries selectively or only lack of global technology companies. This is based rudimentarily address in their strategies. The in part on a general skepticism towards digital USA, and China in particular, benefit from mutu- technologies and the resulting protective attitude ally open civil and military sectors. In India, the AI of politics and society. focus is on the promotion of social and societal aspects. AI superpowers vs. “fortress countries”: In com- parison to the USA, where cooperation between Ethics and human orientation as strategic business and science has grown over decades, strengths of the Europeans: Although ethical Europeans, but also countries such as Canada, issues are discussed in different social forums in Singapore, Japan and India, have so far only been China and the USA, the topic is not given any high able to achieve this permeability to a very limited priority by the respective governments. This has extent. Excellent research results remain in the left leeway for France’s Emanuel Macron to take ivory tower and its walls of debilitating regula- the lead on the topic of ethical AI. Although Lon- tion. Connecting AI research with the needs of don took the initiative faster than Paris to develop the industry thus represents a major challenge in an international AI governance architecture, it was these countries. weakened by the Brexit in its attempts to project and implement such initiative. Other countries, Ecosystems as a strategic asset: In order to such as Finland, also lack the international signifi- develop better solutions and introduce them into cance for this. the international dialogue, in addition to research- ers, 2. talented developers, 3. data pools, 4. com- The private sector as driver in AI development: puting capacities and 5. strategic entrepreneurs, In the leading AI countries, i. e. the USA and China, experienced investors (6.) and an agile (7.) are also the development dynamics are largely determined required. While in the USA and China as well as by the private sector, in particular by young com- partly in Israel these factors provide the breeding panies and globally operating Internet groups. ground for the successful commercialisation of AI, These dynamics are enhanced by the deregulation within Europe this is only possible to some extent tendencies that are common in the USA. In China, in the United Kingdom. In France or Finland, the on the other hand, trends towards increased small number of AI startups bears witness to the state control of the large technology companies fact that such ecosystems are not yet actively 6
Summary thought through and supported to the necessary The countries of the second part of the extent. In Japan and South Korea, the majority study at a glance: of these factors is concentrated in large corpo- rations, while local startup ecosystems remain Canada: Canada has developed a comprehensive small. strategy that involves all levels of economy, sci- ence and industry in order to strategically posi- Lack of computing capacities as a strategic tion the relevant players and support each other weakness: While availability of data and training across disciplines. The aim is to strengthen the of specialists are promoted by many strategies as global presence and improve national prospects. a prerequisite for AI research and commercialisa- The country also strives to take on an interna- tion, only in a few cases is there also a focus on tional leadership role in research and ethical prac- the expansion of domestic computing capacities tice. Canada’s early involvement in AI research (except: South Korea, Japan and China). The cur- has made it possible for provincial, regional and rent global trade conflicts reveal that availability of federal governments to provide targeted support powerful chips or access to cloud-based comput- for educational and research institutions since the ing power is a strategic necessity. Despite a tech- 1980s. The institutions are now well positioned, nologically highly developed economy, there is especially in terms of research and development. only a small globally competitive semiconductors The government continues to play an important industry in Europe. At the moment, US companies role in the development of advanced technologies dominate this sector. by supporting communities with high-level strate- gic investments and transfers. The state collabo- Vague and inconsistent AI definitions: In the rates closely with the private sector to improve strategies compared, there are very different defi- financing, access and commercial applications. nitions of AI, or in some cases no definitions at all. At the same time, the value of ethics is empha- The only thing that the strategies have in com- sized. By supporting academic and commercial mon is that they consider AI a driving force in the recruiting and loyalty tactics through three major digital revolution, which harbours both potentials institutions, Canada aims to offer its economy a and risks in terms of social, economic and secu- sustainable future and an open approach for the rity policy. Accordingly, they often leave aside any exchange of knowledge. existing sector or digital strategies. This is the the- oretical foundation and basis for the divergence Japan: No government in any of the reviewed and convergence of different national AI strate- countries links the future of their country as gies. Such divergence and convergence is both closely to AI as Shinzō Abe in Japan. One of his positive and problematic, since it leads to appli- two top goals is the realization of Society 5.0. In cation diversity on the one hand, but at the same this next stage of human evolution, the physi- time increases the cost and time expenditure for cal and cyber worlds merge. Numerous govern- political transition and implementation of global ment initiatives and an interministerial AI strategy approaches for cooperation and regulation. have already resulted from the implementation work. The focus of research funding is on three Lack of target systems: The strategies are pre- institutes that perform research on specific AI dominantly formulated in general terms. Their applications for increased productivity, mobility, partially vague objectives relate to different levels medicine and safety. So far, the private sector has of impact. For example, China measures the eco- been the main contributor to research spending. nomic strength of the AI industry, among others, Public university research is underfunded and while Great Britain and South Korea also set tar- internationally isolated. The degree of permeabil- gets for the number of future PhD students. The ity and exchange between research and indus- Japanese government has established clear goals try is also rather low. In order to change this, the and time frames for AI applications. cross-appointment system allows researchers and developers from science and the private sector, 7
Summary respectively to work part-time in the other area. United Arab Emirates: Small data pools, no Universities also receive incentives to do more supercomputers, very low scientific performance, research together with the private sector. Large and practically no AI patents. In contrast to that, companies such as Toyota, NEC or Toshiba have there is an urgent need to reduce dependence recognised that the future of machines lies in on oil revenues and to diversify the economy, in their intelligence. They can scale it globally if they combination with the government’s high politi- catch up in software development and adopt the cal capacity to act. As part of a cabinet reshuffle concept of open innovation. So far, there are only in October 2017, the government systematically a few AI startups. With deregulation in sandboxes focused on the vision of making the UAE the “best and geographically defined Strategic Special Zones, country in the world” by 2071. Technology, and AI policy is linked to the development of struc- particularly AI and innovation, plays a central role turally weak regions. The government wants to in this. These ambitions are reflected in the newly achieve global integration by adapting data pro- established State Ministry for AI and in many tection legislation to the EU GDPR, establishing ambitious and concisely formulated technology international standards for robotics and guide- strategies. Their goal is clear: a leading position lines for ethical AI. for the UAE in the application of AI. For implemen- tation, the country relies above all on the creative Israel: In the last two decades, the “startup nation” power and demands of the authorities and on has created a dynamic high-tech sector with at partnerships with countries such as India. It is not least 360 AI start-ups through incentives and sup- yet clear which concrete capacities are to be built port programmes. In a highly developed network up and how they will fit into the global AI land- approach, national and international business and scape. science cooperate closely with the state. Virtually all the major tech companies and corporations in India: India has a unique focus on the applica- the world operate research centres in Israel. They tion of AI to improve welfare and social problems generate many jobs, but often transfer the added and seeks leadership in this area. Before this value of intellectual property abroad. Through the can be achieved, however, India must overcome Technology Transfer Companies (TTC), the domestic a multitude of challenges. There is a particular universities and research clinics have established lack of institutional and coordinating capacities, an effective instrument for the commercialisation which the Indian AI Strategy of 2018 aims to rem- of research. Startups can rely on funding run- edy with institutions at two levels: to promote ning into billions by the innovation authority and the basic AI research and to advance AI commer- Venture Capital, particularly for establishment and cialisation. Other obstacles include a low level of development of marketable technology products. data processing capabilities, a weak intellectual The military considers AI to be “the key to survival property system and relatively low Internet pen- in the modern world” and acts as a driving force etration. India’s approach will bear fruit only if of innovation through its selection and training and when these factors show progress. However, programmes: Israel – a knight of the cognitive era important findings on institution establishment Although there is still no independent AI s trategy, can be derived from the Indian AI strategy, such yet, Prime Minister Benjamin Netanyahu sees as the strategically positioned two-stage institu- great export potential for AI in the fields of cyber tions at the level of basic AI research and AI com- security, digital medicine and mobility. However, mercialisation policy. Both aim to build partner- the transformative force of AI has so far been little ships with relevant stakeholders to found and reflected in society of the country. expand networks and accelerate AI development. 8
Summary Singapore: The country is seeking a co-operative AI ecosystem between industry and science and intends to provide an environment with appropri- ate resources to foster innovative development. In the absence of mineral resources and geopo- litical relevance, the promotion of digitisation and AI has become one of the government’s priori- ties, with the aim of becoming known worldwide as the AI hub. To this end, the five-year initiative AI Singapore was developed in 2017. Within the 100Experiments scheme, companies can submit problems for which there is no standardised AI solution yet, but for which a solution could easily be worked out. These companies and their prob- lems are then placed with AI developers. Through Fundamental Research, Singapore supports basic research and addresses research gaps, with a view to both technical and ethical societal issues. Understanding and acceptance of AI among the population is promoted by the AI for Everyone scheme, which offers specific learning formats on AI. The Accreditation@ SGD initiative supports young companies in their early growth phases by means of customised consulting services, par- ticularly in the areas of technical applications. In addition to its efforts to create an AI-friendly envi- ronment among relevant players and in terms of legislation, the government also intends to pro- mote, through various initiatives, the responsible use of data and the ethical use of AI solutions. 9
Cambrian AI Index © The analysis assessed the countries on the basis was developed. The index is limited by proxy indicators that incorporate the countries’ pre- measurements for which reliable and compara- conditions, the research and development situ- ble data are available at this early stage of the AI ation and the degree of commercialisation of AI. application. There was no weighting of the data. To integrate these indicators and determine the The reference country for the index is the United AI position of a country, the Cambrian AI Index © States, the world’s leading AI nation. Cambrian AI Index © of countries from part 2 of the study 0,7 0,6 0,7 0,5 0,7 0,6 0,4 0,6 0,5 0,3 0,5 0,4 0,2 0,4 0,3 0,1 0,3 0,2 0,0 0,2 0,1 re iriara bb ira b da ia an y l 0,1 ae Em AAraraEm Ara an tes Ind po na Jap Isr rm 0,0 ga Ca d 0,0 ite Ge Sin rere ddaa iaia an Un yy ll aeae an an tetses InIndd ppoo nnaa ap an IsIrsr JaJp rm gaga CCaa rm dd itiete 1,2 GGee SiSnin Em UUnn Cambrian AI Index © of countries from part 1 and 2 of the study 1,0 1,2 1,2 0,8 1,0 1,0 0,6 0,8 0,8 0,4 0,6 0,6 0,2 0,4 0,4 0,0 0,2 0,2 da an ira b ia ereiri s s eri s re ina y e l d gd dd gd d ae KKoo oof f Ko of AAmm tete Am tate Em AAraraEm Ara an tes nc ca lan KKinin nnitieteKin nite Ind om po na Jap rea Ch Isr c Fra rm 0,0 bli Fin ga Ca S d 0,0 U ite Ge d pu Sin ite ddaa an b iaia rere Un ininaa yy ee ll dd Re aeae an b an tetses nnc c oof f StSataof caca an Un InIndd oomm ppoo nnaa ap reraea an laln CChh IsIrsr cc JaJp FrFara rm iriara bbli li FiFnin gaga CCaa rm dd UU gd itiete GGee d ppuu SiSnin d itiete Em UUnn RRee UUnn General requirements Research and Development Commercialisation Cambrian AI Index General General requirements Research requirements Research and and Development Development Commercialisation Cambrian Commercialisation Cambrian AI AI Index Index 10
Canada AI research location with a long history ›› AI research for decades ›› A pan-Canadian network of university laboratories focusing on AI core research ›› Location of AI research labs of many tech giants including Microsoft, Google, Facebook and Samsung ›› Favourable immigration policies, grants and tax incentives; as well as special regulatory zones to encourage local and international companies to commercialise AI I.) Introduction achieve these goals, the various administrative and governmental levels in the country have allo- In November 2017, Canada’s Prime Minister Jus- cated a total of approximately 430 million Euros6 tin Trudeau stood on stage with Eric Schmidt, provided by various schemes, either directly or CEO of Alphabet, discussing the importance of very closely linked to R&D in AI, talent delivery AI as a driver of economic growth.5 Looking at and commercialisation. the text of the Canadian AI strategy paper, it is striking that social sectors, such as healthcare and social welfare are not explicitly mentioned – II.) Requirements for AI unlike many other national AI-related strategies. Rather, the declared objectives are: 1) to prevent Canada has about 36 million inhabitants (about AI talent from migrating to the USA, 2) to create one tenth of the United States), 32 million of favourable conditions for the commercialisation whom used the Internet in 2016.7 The source of and economic growth of the country. In order to potential data for AI is correspondingly small. 11
Canada The public sector, on the other hand (also a rel- this end, the AI & Society team runs nation- evant data source for AI) makes its data freely wide workshops, conducts surveys with available in high quality (among the countries experts and cooperates with the UK and compared, the availability and quality of public France. In addition to this “top down”-Initia- sector data is better only in the UK).8 In relation tive, the Montreal AI players have compiled to Canada’s small population, about as many the Montreal Declaration for Responsible Master’s students are trained in AI-related fields AI, which aims to stimulate the discussion as in the USA. This puts Canada ahead of Great between the public, the private sector and Britain and France and behind Singapore and the state.12 It proposes a framework and a Israel. In absolute figures, however, it is esti- set of values, such as well-being, autonomy, mated that there are currently only 860 Master’s justice and privacy, which must be assessed students, which means that the country still lags and observed in the development or imple- significantly behind the USA.9 In addition, Canada mentation of AI technologies. only operates six of the world’s 500 best super- computers.10 IV.) Research and Development III.) Institutional framework In Canada, total R&D spending in 2016 was at Canada has a strong institutional framework to around 22 billion Euro.13 Around 200 million Euro implement AI R&D funds for a network of estab- have been allocated for AI research; 83 million lished top research institutions. A central pil- via the Pan-Canadian AI Strategy14 and 118 million lar in this framework is the Canadian Institute For are forwarded to the universities of Montreal and Advancement and Research (CIFAR), founded in McGill through the Canada First Research Excel- 1982, which currently runs twelve different R&D lence Fund.15 In addition the provincial govern- programmes in 16 provinces, which in turn pro- ments of Quebec and Ontario recently allocated vide orientation for the Pan-Canadian AI Strategy, a further 100 million Euro exclusively for R&D in and whigh governs the recently established Fed- AI.16 According to the CSRanking, about 120 teach- eral Fund with 83 million Euro. CIFAR is currently ers (about 1,060 in the USA) have been engaged forwarding the majority of this fund to the three in research on AI in the country since 2016,17 main AI R&D centres in Canada: the Montreal Insti- which represents a similar relation relative to the tute for Learning Algorithms (MILA) in Montreal, the overall population as in the United States.18 These Vector Institute in Toronto and the Alberta Machine teachers are estimated to supervise 370 doctoral Intelligence Institute (AMII) in Edmonton. All three students per year.19 The AI-relevant scientific pro- centres are associated with universities that strictly duction is 1,200 citable documents,20 which puts focus on AI R&D and not on commercialisation. Canada in eighth place among the twelve coun- tries compared. Interestingly, however, Canada Ethics – “top down” and “bottom up”: The ranks fourth in the H index, which measures the discourse on AI ethics and the impact of AI influence of these publications (after the US, UK progress on society in Canada takes place at and China),21 which emphasises the high quality several levels On the one side, CIFAR man- of Canadian AI publications. ages an independent program (with its own fund and team) called “AI & Society”. It is Research areas and instruments focused on future national economic, ethi- The most important instrument for AI R&D is the cal and legal policies which take into account Pan-Canadian AI Strategy managed by CIFAR (see the concerns about the impact of AI on the above), which coordinates research at the three labour market or on the health sector.11 To major AI hubs. These three centres, located in Montreal, Toronto and Edmonton, are all associ- 12
Canada ated to local universities and focus almost exclu- V.) Commercialisation sively on AI research and development. These centres, each led by one of the three famous AI Commercialisation of AI research is the weak researchers Dr. Yoshua Bengio, Geoffrey Hinton point of the country. While a study by Asgard and and Richard Sutton, are pioneers in the develop- Roland Berger in 2018 recorded 131 AI startups ment of Deep Learning, neural networks and rein- (just under 1,400 in the USA),28 another mapping forcement learning. It is worth noting that they of the Canadian startup scene revealed 650 AI- continued to operate during the last “AI hiberna- focused startups scattered across the country in tion” of the 1990s and 2000s, financed by a sister 2018.29 In 2016, however, only 2.7 percent of all AI programme also managed by CIFAR (“Learn- global investments in AI startups were made in ings in Machines & Brains”).22 Today, with renewed Canada (USA 62 percent in the same period).30 interest in AI research, all three centres benefit Also, only a small proportion of relevant AI pat- from the basic funding provided by the federal ents originate in the country (between 2015 and programmes. The Vector Institute hub in Toronto 2017, an average of 1.43 percent of all interna- is currently the best funded. In addition to the tionally enforceable AI patents worldwide).31 At the 14 million Euro it received from CIFAR, a further level of automation of the economy, measured 86 million Euro came from the provincial govern- by the number of robots per 10,000 employees in ment and the private sector, in particular Goog- the manufacturing industry, the country is ranked le.23 It’s not uncommon in Canada for private in the back midfield (145 robots in comparison investments to be made following public funds. to 631 in South Korea and 71 in Great Britain).32 With the additional funding from the private sec- Among the countries compared, the public sec- tor and the provincial governments, all three tor exerts the least influence as a possible driver centres have expanded their basic research and of innovation through its demand33 and does not are now actively participating in the development play a role in the strategy as a possible user of AI, of the following key technologies: Deep learning, either. neural networks, reinforcement learning, Pattern Recognition, Computer Vision, unattended learn- Regulation: The Canadian federal govern- ing, Natural language processing, Deep Networks, ment sets the guidelines for provincial regula- learning theory and optimisation of Deep learn- tion. Of the ten provinces, the following are the ing, statistical theory and algorithmic gaming most economically important: Ontario, Que- theory. All in all, the centres directly or indirectly bec, Alberta and British Columbia. The existing support a large network of researchers, with rules and regulations are not sufficient to pro- the MILA Montreal Centre taking the lead and mote the use and application of AI, in particular directly supporting 234 AI researchers.24 with regard to intellectual property, copyright and ownership of both inputs (data) and out- Canadian companies are also offered 15 to puts (decisions or actions of robots) and think- 35 percent tax benefit for basic and applied ing machines that use AI).34 CIFAR, with its “AI & research in the field of science and technology Society”programme, is working hard tochange through the Scientific Research and Expert Develop- this. They focus on cross-company and cross- ment (SR&ED) program.25 In addition, the Global industry collaboration, data privacy and ethics. Talent Stream program26 simplifies immigration PIPEDA, the law on the protection of personal requirements for AI researchers, and other spe- data and electronic documents, which in many cialists and talents. According to Navdeep Bains, respects is similar to the EU GDPR, regulates the Canada’s Minister of Innovation, Science and Eco- use of data (the input) as well as the actions in nomic Development it only takes “two weeks” until case of data protection violations, as stipulated a work permit for these researchers is applied for by the latest amendment (November 2018).35 The and issued.27 Other countries can learn from this regulations also include how Canadian startups as the battle for AI talent intensifies. and companies can use data in their AI models. 13
Canada The Canadian Securities Authority (CSA), the gov- At the beginning of 2018, the federal govern- erning body of the Canadian capital market, has ment announced a new commercialisation instru- established a special regulatory zone (sandbox) ment, the “Supercluster” initiative, which will be to explore new business models with innova- financed with around 635 million Euro over five tive products in the capital markets, such as the years.41 The Superclusters were created specifi- use of AI for trading.36 Similar special zones are cally to promote economic growth and to encour- also being developed by Transport Canada for the age the private sector to collaborate with educa- development of self-propelled vehicles, after the tion and research institutions to create regional institution has specified “Guidelines for Testing of innovation ecosystems comparable to Silicon Highly Automated Vehicles”.37 Health Canada also Valley. Three of these superclusters are specifi- promotes the research and testing of technologi- cally oriented towards applied research and the cally advanced (and possibly AI-controlled) medi- use of AI technologies for commercialisation. The cal devices.38 Advanced Manufacturing Supercluster in Ontario focuses on the next generation of manufacturing Generally speaking, while Canada’s current regula- and robotics.42 With 270 participants, the largest tions remain insufficient and overall cautious with “Digital Technology Supercluster” in British Colum- regard to AI applications, they are not too strict to bia concentrates on the use of larger data records discourage companies. On the contrary: Compa- and machine learning to improve service delivery nies and startups are familiar with what is allowed in the fields of natural resources, precision health and what is not (e. g. through PIPEDA), they are care and manufacturing.43 The “AI-Powered Supply given the opportunity to participate in the discus- Chains Supercluster (SCALE.AI)”, headquartered in sions on future regulations (e. g. through CIFAR Quebec, focuses on harmonising the sectors of and the CSA Special Regulatory Zone) and have manufacturing, transport, IT and retail as well as a good understanding of which industries will be on developing intelligent supply chains using AI better regulated next. and robotics.44 Promotion of startups and companies: The Cana- dian federal government is focusing its direct sup- port of commercialisation of technology and inno- vation on larger projects and companies. However, smaller companies and startups, are not left alone and well supported by private venture capital. The Strategic Innovation Fund of Canada (SIF), a program designed to accelerate technology trans- fer and the commercialisation of innovative prod- ucts, is intended to provide grants for companies and their projects, which will cover up to 50 per- cent of the expenditure.39 This fund is also availa- ble to foreign companies intending to do business in Canada. With the 2018 budget, this fund was replenished with 840 million Euro for a period of five years.40 Although the SIF is more traditional in its operation (i. e. ‘grant application’), it is quite flexible and often changes its parameters. In Feb- ruary 2018, for example, the fund announced that it would adapt to the economic situation and gen- eral access to capital by providing 6.7 million Euro to support larger projects. 14
Canada 5 The New York Times, 2017. 23 Shead, 2017. 6 For standardisation purposes, all amounts in foreign 24 MILA, k. D. currencies were converted into Euro and rounded at the 25 Ernst & Young, 2018: 45. exchange rate of November 12, 2018. 26 Government of Canada, k. D.b. 7 World Bank, 2016 (cf. Methodology of the Cambrian 27 Smith, 2018. AI Index). 28 Asgard Human Venture Capital/Roland Berger, 2018 8 Open Data Barometer, 2016 (cf. Methodology (cf. Methodology of the Cambrian AI Index). of the Cambrian AI Index). 29 Gagné, 2018. 9 CS Ranking, 2018 (cf. Methodology of the Cambrian 30 CB Insights, 2017a (cf. Methodology of the Cambrian AI Index). AI Index). 10 Top500.org, 2018 (cf. Methodology of the Cambrian 31 M-Cam, 2018 (cf. Methodology of the Cambrian AI Index). AI Index). 32 IFR, 2017 (cf. Methodology of the Cambrian AI Index). 11 CIFAR, k. D. 33 World Economic Forum, 2017: 82–83 (cf. Methodology 12 Université de Montréal, 2017a. of the Cambrian AI Index). 13 UNESCO, k. D. (cf. Methodology of the Cambrian AI Index). 34 Aubin, Freedin, 2017. 14 CIFAR, 2017. 35 CanLII, 2015. 15 Government of Canada, k. D.a. 36 CSA, 2017. 16 Université de Montréal, 2017b; Government of Ontario, 37 Government of Canada, 2018a. 2017. 38 Government of Canada, 2017. 17 CS Ranking, 2018 (cf. Methodology of the Cambrian AI Index). 39 Government of Canada, k. D.c. 18 Ibidem. 40 Ibidem. 19 Ibidem. 41 Government of Canada, 2018b. 20 SJR, 2017 (cf. Methodology of the Cambrian AI Index). 42 Ibidem. 21 Ibidem. 43 Ibidem. 22 CIFAR, k. D. 44 Ibidem. 15
Japan On the way to “Society 5.0” with AI ›› Interministerial AI strategy and numerous government initiatives for attainment of Society 5.0 ›› Three research institutes researching specific AI application areas ›› Cross-appointment system and reforms for increased permeability between research and industry ›› Adapted copyright law for text and data mining and adaptation to EU GDPR ›› Deregulation in special regulatory zones (sandboxes) and geographically defined strategic special zones ›› But: Hardware focus of large corporations (IoT) and only a few AI startups I.) Introduction that machines are increasingly taking over peo- ple’s jobs is less cause for concern in Japan than “Abenomics”, Premier Shinzō Abe’s ambitious for hope. For the population is aging and shrink- economic programme, has two key objectives. ing, especially in rural areas. Since immigration One is sustainable growth. The other is nothing is very limited for various reasons (less than two less than mankind’s evolutionary move towards percent of foreigners in 2016),47 AI is intended to the “super-smart society 5.0.”45 “Society 5.0 is a help compensate for a shrinking working popu- human-centred society that reconciles economic lation. Consequently, the focus is on practical progress with the solution of social problems fields of application which fundamental research through a system that integrates cyberspace and is intended to serve: 1) productivity, 2) health, physical space to a very high degree.”46 The fact care and wellbeing, 3) mobility and 4) security.48 16
Japan I-supported robots are to be used primarily in A turn, investments of 17 million Euro are planned geriatric care,49 which is also based on a cultural for research into quantum computers and chips, openness to non-human forms of intelligence.50 and around 15 million Euro for the establishment AI technology should be understood as a service of an “open platform for nanotechnology and (AIaaS) based on large amounts of data from the materials research”.58 Internet of Things (IoT), as the strategy presented in March 2017 explains.51 The talent pool for AI is valued at 210 Master stu- dents who graduate annually at computer science For the 2018 financial year alone, the govern- institutes, and rather small compared to South ment has planned total investments of around Korea.59 580 million Euro in the AI sector, which repre- sents an increase of 30 percent over the previous year, but which the Japanese press still considers III.) Institutional framework to be too low.52 Shinzō Abe and his party have a comfortable The globally operating Japanese corporations majority in both houses of parliament with regard are strong in the hardware sector, but lack soft- to implementation of Society 5.0. The strategies ware innovations. They try to buy these abroad, of most ministries include references to AI, such as there are hardly any AI startups operating in as the Integrated Innovation Strategy and the Japan the country at the moment. There is also a lack Revitalization Roadmap.60 of training for AI talents, but research is interna- tionally isolated. If, however, AI research support The framework conditions for research and devel- is successful, the country catches up in terms of opment of innovations as well as the allocation of software development and the state regulation budget and personnel are set by the “Council for measures become effective, Japan has the poten- Science and Technology and Innovation” (CSTI). tial to become a world leader in AI in the areas of The CSTI is chaired by the Prime Minister and the industrial production, medicine and mobility. individual sector ministries follow the guidelines of the CSTI.61 Already in April 2016, the “Strate- gic Council for AI Technology” was founded on II.) Requirements for AI the instructions of Shinzō Abe, which presented an AI strategy in March 2017. The Council con- With around 118 million, Japan has comparatively siders itself a “control tower”62 that coordinates many Internet users (fourth place after China, the efforts of various ministries. The coordinated India and the USA),53 however, due to the continu- ministries are: 1) Ministry of Internal Affairs and ing negative demographic trend, the population Communications, 2) Ministry of Education, Cul- is ageing and shrinking. The country ranges more ture, Sport, Science and Technology, 3) Ministry in the midfield in terms of access to and qual- of Economy, Trade and Industry. Coordination ity of public data (sixth among the twelve coun- includes the research centres subordinated to the tries surveyed).54 Of the 500 strongest computers ministries as well as the “Japan Science and Tech- worldwide, 36 are in Japan (third place after China nology Agency” (JST) and the “New Energy and and the USA). One of them (ABCI) currently ranks Industrial Technology Development Organization” fifth in the top 500.55 In the strategically important (NEDO).63 The CSTI also includes representatives of semiconductor industry, four companies56 gener- the Ministries of Health and Labour, Land, Infra- ated sales of around 30 billion Euro in 2017, which structure, Transport and Tourism, as well as Agri- represented around 17 percent of the sales of US culture and Fisheries, which have large amounts companies.57 The 2018 budget therefore foresees of data at their disposal,64 as well as representa- investments for “R&D in AI chips of high efficiency tives of universities and the industry association and speed through industry-government-science Keidanren. The Ministry of Defence has so far only cooperation” at an amount of 77 million Euro. In played a minor role. 17
Japan In addition to promoting R&D, the Council coordi- at the University of Tokyo as the only Japanese nates with the industries that use AI (exit indus- university AI institute in the country.70 Since 2016, tries) and promotes the social fields of application 30 faculty members have been actively research- for AI within an industry coordination body. The ing in AI areas at this institute71 and supervising AI strategy also calls for an ambitious industrial approximately 90 doctoral candidates in this field strategy (Industrialization Roadmap), which com- every year. bines the “wisdom of industry, science and gov- ernment” to arrive at “consistent approaches” in The three non-university AI research centres terms of research, commercialisation and social AIST, RIKEN and NICT are “integrated administra- implementation of AI.65 tive institutes” subordinated to various ministries. They employ numerous researchers, but do not Numerous proposals and initiatives for offer any training or educational courses. Moreo- ethical AI with positive basic tenor: Already ver, there are more than 70 robotics laborato- in May 2016, the Advisory Council for AI and ries nationwide,72 some of which also deal with Human Society was established on the initia- AI applications.73 Elderly gentlemen dominate the tive of the CSTI. It consists of twelve members management of these research labs, which in from the fields of engineering, philosophy, turn hardly cooperate with each other. Moreo- law, economics and social sciences. Its report ver, the higher education system is underfunded of March 2017 takes a critical look at ethical, and becomes less and less efficient.74 In 2017, legal, social and economic considerations, researchers in various fields have been able to as well as the influence of AI on education issue 2.800 citable AI publications (fourth place and research, but comes to the conclusion among the twelve countries surveyed). However, that the positive aspects clearly predomi- their overall influence was rather small (sixth nate. New forms of interaction between man place out of twelve).75 Only publications on Com- and machine are an opportunity to rethink puter Vision and Pattern Recognition (fourth place the concept of humanity.66 The Council also worldwide) and Human-Computer Interaction proposes that an Institutional Review Board (fifth place worldwide) had a noteworthy world- (IRB) on AI be established at all universities.67 wide influence.76 International rules and standards for AI were adopted by the Ministry of Internal Affairs In order to implement Society 5.0, Prime Minister and Communications (MIC) in 2016 at the Abe’s government intends to increase total R&D G7 meeting of ICT ministers.68 In addition, spending to one percent of GDP,77 to fund three the Japan Society for Artificial Intelligence has research centres for concrete AI applications in developed principles for ethical AI.69 The lack four areas of activity, train AI talents, increase per- of diversity (gender, age, origin, language) meability between business and science and over- within the research and development teams come the international isolation of the research in Japan could become a problem for devel- sector. opment of neutral (unbiased) algorithms. Research areas and instruments In the centre of state funding of AI research are three research centers,78 that work interdisci- IV.) Research and Development plinary and collaborate with each other. These research centres are also planned to be used as In absolute figures, total R&D spending in 2016 “research hubs” for “open innovation” created by amounted to around 140 billion Euro (third place cooperation between industry, government and among the twelve countries surveyed). The pub- science (by other universities).79 According to the lic sector contributed only around 22 percent of Industrialization Road Map, there are clear plan- this sum, the lowest of all the countries surveyed. ning guidelines for roles, timelines and expected CSRankings rates the Computer Science Institute results of these centers.80 They will focus on 18
Japan research in four strategic application fields (pro- related to AI. A laboratory for interlinking of vari- ductivity, mobility, health, safety). This involves ous AI projects both of the research institutes and approaches that must be consistently pursued the universities (AIP Network Lab, PRISM)88 is a par- from fundamental research through to social ticularly noteworthy project in this context. implementation. On the other hand, there are approaches for which a short-term monetarisa- The National Institute of Information and Commu- tion and thus a commitment of the private sec- nication Technology (NICT) reports to the Ministry tor is not expected, or approaches in cooperative of the Internal Affairs and Communications (MIC) areas, such as international standardisation and and was endowed with around 17 million Euro common infrastructure technology.81 From 2020, in 2016. On the one hand, research is being car- the results of the individual centres are to be com- ried out there on the development of “universal bined in cooperation with companies and minis- communication technology”.89 On the other hand, tries to form “integrated systems”. CiNet is also located there and engages in applica- tions of nerosciences in computer science, inter- The Artificial Intelligence Research Center (AIRC) at faces between brain and machine, and robotics the AIST Institute considers itself a “central contact for medical applications, which “helps us under- point for the promotion of large-scale research”82 stand how humans and robots can best coexist in and has at its disposal 123 million Euro (2016), future.”90 the largest budget among the three centres men- tioned above. This is expected to almost dou- Counteracting the lack of AI talents is one of the ble in 2019 to around 212 million Euro.83 On the most urgent challenges because by 2020 the METI basis of Deep learning and neural networks twelve Ministry of Economics expects a shortfall of about teams are performing research on the interaction 48,000 people in “leading IT human resources”.91 of AI and IoT, among others, as well as pattern These are divided into three groups: those that and image recognition (for medicine or security). 1) solve basic AI problems (especially in the fields According to AIST information, the world’s largest of information technology, robotics, Natural Lan- open AI computing infrastructure (ABCI) is avail- guage Recognition, neurosciences), 2) translate AI able to AIST.84 AIST also maintains a joint AI labora- fundamentals (e. g. in algorithms, database archi- tory with NEC and partnerships with foreign insti- tectures and programs), 3) are able to apply and tutes such as the DFKI in Germany. use AI practically in industries and services.92 The Center for AI Development (AIP) is part of the The first two groups are to be expanded through RIKEN Institute for Physics and Chemistry gov- attractive salaries, favourable research condi- erned by the Ministry of Education (MEXT). More tions and contents for domestic and foreign than 50 teams do research on fundamentals, researchers. Young AI researchers, in particular, “goal-oriented technology research” and “AI in are intended to benefit from this and to access society”.85 The AIP will have an important role to funds, for example through JST programmes.93 The play, especially in the development of AI-support- third group needs AI further training and overall ing “all-purpose infrastructure technology” for the improved training and education in mathematics purpose of “revolutionising industrial production”, and IT subjects. The exact needs and strategies for medical diagnostics and damage limitation in for this are to be clarified in a discourse between the event of natural disasters.86 In 2016, around industry and science under the auspices of the 56 million Euro were available for these areas.87 METI Ministry of Economics.94 The AIP cooperates with NEC, Toshiba, Fujitsu and Fujifilm through separate, respective AI centres. To date, the degree of cooperation between uni- Also under the authority of the Ministry of Educa- versities and industry has been rather low (sixth tion (MEXT) is the Japan Science and Technology place among the twelve countries surveyed).95 Agency (JST), which in turn runs programmes for An important instrument to improve this is the the promotion of “strategic fundamental research” cross appointment system,96 which allows research- 19
Japan ers from universities to also work part-time at sufficiently adopted by the corporations.107 At the national research institutes or in the private sec- same time, only few impulses are generated by tor. In return, researchers from the private sec- the domestic innovation ecosystem. Asgard and tor are allowed to work part-time at universities. Roland Berger only count 113 AI startups (USA: New subsidy mechanisms are also intended to 1,393, China: 383),108 of which CB Insights consid- benefit universities that implement “manage- ers at least two to be among the most influential ment reforms” and seek funding from the private AI startups in the world, i. e. “Preferred Networks” sector. At the same time, companies are to be and “LeapMind”.109 Similar to South Korean con- encouraged to invest more in long-term research glomerates, Japanese corporations buy innovation projects at and in cooperation with universities97 abroad, such as Toyota in Silicon Valley. and establish so-called Moonshot initiatives.98 The Abe government supports AI commercialisa- Moreover, in 2016, the Ministry of Economics tion alongside its commitment to global indus- (METI) stated that Japan was isolated from the try standards for industrial robots under the New global flow of researchers and research funds.99 Robot Strategy,110 through adjustments in copy- Therefore, the immigration requirements for right law, an EU-compatible data protection law, highly qualified foreigners and the underly- accelerated or suspended approval procedures, ing points system were revised and updated in startup support and tax relief for modern IT pro- 2017.100 Now, researchers are permitted to take curement. on multiple employment, obtain permanent stay permits much faster, initiate family reunion and Regulation: From January 2019111 an already obtain work permits for their partners. In the resolved extension of copyright law will apply, field of AI, research partnerships have also been which through three new articles permits the agreed, for example with Israel101 and Germany102 use and further processing (text and data min- ing) of protected contents by commercial and non-commercial AIs even without the consent of V.) Commercialisation the authors (fair-use principle).112 Access to large amounts of data is important to improve the self- In addition to cooperation with the aforemen- learning capabilities of machines. A similar regula- tioned centres, many of the globally operating tion is currently initiated in Singapore and Can- corporations (Toshiba, NEC, Hitachi, Sony, Mit- ada. The legal framework is also to be extended subishi Electrics, Fujitsu, Canon, etc.) conduct their for copyright of products created by AI (e. g. texts, own application-oriented research programmes music).113 on AI.103 According to government information, the private sector invests around 4.5 billion Euro in In May 2017, the Act on the Protection of Personal technology per year.104 On average, between 2015 Information was also revised in order to promote and 2017, 5.17 percent of internationally enforce- the collection and sharing of data while at the able patents came from Japan,105 which puts the same time decoupling them from personal data country in second place in the world. Despite this and processing them anonymously. The revised ranking, the gap to the USA, who are at first place, version has many similarities with the EU GDPR. is still enormous, since around three quarters of Against this background, the EU has put Japan the mentionend AI patents in the same period on a “white list” of countries with data protection came from the USA. standards comparable to the high standards of the EU.114 This promotes global integration of the Although Japanese companies continue to be country, as Japan, unlike China, allows its AI prod- leading in the production and export of robots, ucts to be exported to the EU. this position is threatened because manufacturers have invested too little in software development.106 A law adopted in June 2018 allows companies to The concept of open innovation has not yet been develop innovative pilot projects within previously 20
You can also read