欧洲人工智能生态系统地图（英文版）.pdf

Joining the dots A map of Europes AI ecosystemThe battle for AI supremacy has begun. Around the globe, governments, businesses and society have realized the technologys potential to transform the world we live in, whether by automating tasks or providing insights into our health, behavior or livelihoods.But as with past game-changing technologies, AIs arrival will not be without upheaval. There will be major implications: businesses, jobs and living conditions will all be affected, leaving many fearful of the world to come. This means its important to get AI right. At the moment, many players are pursuing the technology enthusiastically, but failing to fully harness its potential.Roland Berger and France Digitale are convinced AI will be a force for good provided it is properly implemented. As such, we feel there is a need to better understand the technology, and also the whole ecosystem behind it. This report goes some way towards that. France Digitale has a wealth of data on AI in Europe, and suggested that Roland Berger make a joint effort to organize and analyze it, to build a picture of the European AI ecosystem. The result is a long-term knowledge tool that consolidates Europes AI expertise, and provides a great base to bolster Europes fight for AI leadership.ForewordCharles-Edouard Boue CEO, Roland BergerFrdric Mazzella Vice President, France DigitaleDamien Gromier Prsident de FranceIsAI by France Digitale2 Joining the dots A map of Europes AI ecosystem1. Results: Volume and dynamism 82. Towards a unified European voice on AI? 163. Recommendation 17Coverillusration: feoris / istockphotoContentsJoining the dots A map of Europes AI ecosystem 3Artificial intelligence (AI) is not yet so intelligent that it happens by itself. The process of automating tasks pre-viously only suited to humans is driven by huge amounts of data, smart algorithms and even smarter people.By the same token, AI is not only about startups devel-oping the technology or using it in self-driving cars or to automate emails. There is a whole ecosystem behind it. nullhile startups are a vital part, this also includes re-search laboratories (including public and private labs, and combinations of the two), all kinds of communities (from informal groups to conferences and publicnullri -vate associations) and investors. nulloland Berger is at the forefront of AI knowledge re-search, having published many important studies in this area. nulle of the most recent (Artificial Intelligence null A strategy for nulluropean startups) found that nulluropenulls AI ecosystem is thriving, but, unlike in the null or nullina, it is fragmented and lacks a clear strategy. This means the continent could get left behind.To help remedy this, and in partnership with nullance nullgitale, we carried out a study to investigate what enullct -ly the nullropean AI ecosystem looks likenullwhonulls involved, where theynulle based and what theynulle doing. It covered null countries and included a detailed look at startups, products, research labs, communities and investors.The result is a comprehensive map of the nulluropean AI ecosystem. It shows a rich and diverse landscape, high-lighting three distinct country clusters. In summary, the ecosystem consists of nullnullnullstartups, nullnulllabs and nullnullnullcommunities spread over nulldominant, null rising star and null follower countries. In this document, we report the studys methodology and results, analyze what they mean and suggest ways in which the currently fragmented European AI ecosystem can find its voice.Management summary4 Joining the dots A map of Europes AI ecosystemRoland Berger and France Digitale have built an original database containing information on three types of European AI players: startups, labs and communities. This was done using desk research and sources such as CB Insight, Crunchbase and Traxcn. A web survey across our networks was also conducted to collect more detailed information. The result is a detailed picture of the European AI ecosystem in 2018, which will be updated and enriched over time.The studys scope covered the 28 European Union members plus Norway and Switzerland. Within each, startups, labs and communities were identified according to specific criteria. For example, startups had to be a producer or user of a recognized AI technology such as machine learning or neural networks; labs had to be in educational institutes or run by public, corporate or public/private backers; and communities had to fall under a list of categories from Facebook groups to summer schools and magazine readerso key indicators were then calculated for each country. The first was mass criteria, or the volume of startups, labs and communities. The second was ecosystem density, or volume divided by the countrys population. Where relevant, other analysis was performed, for example measuring the activity of communities by number of members or planned events. We also had a close look at policy initiatives developed by the different countries, and the European institutions. According to their results, countries were clustered into three categories: dominant, rising star or follower. The follower group is not considered in detail in this report. Methodology: Crunching the numbersJoining the dots A map of Europes AI ecosystem 5FranceIrelandSpainPortugalUnitedKingdomEuropes AI playersWhile leaders are home to the highest volume of AI players, rising stars boast very dense AI ecosystems.67%50%57%Avg. nb of members in meetups between 210-53020%Startups/population ratios 1%23%Avg. nb of members in meetups between 450-620 24%13%26%Avg. nb of members in meetups between 110-53020% of total (#)LeadersStartupsLabsCommunitiesRising starsFollowersSource: Roland Berger6 Joining the dots A map of Europes AI ecosystemFranceBelgiumLuxembourgNetherlandsDenmarkSwedenEstoniaLatviaLithuaniaFinlandNorwayGermanySwitzerlandPolandCzech RepublicAustriaSlovakiaHungaryRomaniaBulgariaGreeceCroatiaSloveniaCyprusMaltaItalyJoining the dots A map of Europes AI ecosystem 7The dominant fournullasured by mass criteria, four countries stood out as dominant players. The null, nullance, nullermany and nullain are clearly driving nullropenulls AI ecosystem, contributing nullnullof startups, labs and communities across the null countries. Broken down, they are home to nullnullnullstart-ups (nullnullof the total), nullnulllabs (nullnull and nullnullnullcommu -nities (nullnull.The four finished top in all three volume categories, with only the order of the top three differing. nullain ranked fourth in each case. The null finished top in the startup (nullnull and communities (nullnull categories, while nullance led in number of labs (null).nullher notable volume figures included the fact that nullnullof the labs within the top four are based in publicly funded universitiesnullchools, and that so-called meetup groups were by far the largest community type, dwarf-ing others such as nullnkedIn groups, conferences or re -search networks. Twitter use was also higher among the top four (see nenull page).The four are also leaders when it comes to technological and sectoral trends. In fact, they are so dominant that they are imposing these trends on the rising stars and follower countries. The distribution of the AI technolo-gy used across nullrope, for enullmple, is almost the same among all three groups, reflecting the four leadersnullpref -erence for machinenulleep learning and natural language processing.The leaders also have a focus on four AI sectors that is replicated by the rising stars and followersnull tech, finan-cial services, entertainmentnulledianullulture and health -carenulliotech make up very similar proportions of the market in each category (between null and nullnull. nullme nullnullof startups in the top four have applications in these sectors.The figures suggest that the successful AI model devel-oped by the leaders is being copied by their neighbors. In short, everyone is working in the same areas, and fol-lowing the trends started by the dominant four. The rising starsnullr cluster of rising stars comprises null countries, main -ly nullordic and Baltic states, which are snapping at the heels of the leaders and becoming increasingly influen-tial in AI. nullile their overall volume of startups, labs and communities is much lower than the top four (about nullnullcompared with nullnull, some rising stars, such as the nulletherlands and nulleden, were close to over -taking nullain in individual categories. 1. Results: Volume and dynamism The leaders have a focus on four AI sectors that is replicated by the rising stars and followers: tech, financial services, entertainment/media/culture and healthcare/biotech.8 Joining the dots A map of Europes AI ecosystemTweet or delete? AI on TwitterTo get a snapshot of AI activity in Europe, we performed an unscientific but nevertheless revealing Twitter analysis. Data was collected for the 30 study countries plus several large cities over 13 months (three for the cities) based on mentions of the key words “artificial intelligence” and “ai”, plus their French and German equivalents. From these, the level of output (number of tweets) and engagement (number of retweets, shares, likes) were calculated.While the results had an obvious bias towards the UK, France and Germany, they gave a good idea of the level of interaction around AI. More than 8.2 million tweets were counted in total, with France (2.6 million tweets 31% of the total) and Paris leading the country and city output rankings. Again, the dominant four topped the country chart, with the UK in second place (1.98 million tweets) and Germany in third. Italy was fifth, with rising stars the Netherlands and Switzerland close behind (both just under 300k). Interestingly, while London finished as the second-placed city, Madrid beat Berlin to third position with 5% of the total city output.When it came to engagement, France and the UK again finished as the top countries with 96,381 and 59,590 retweets/shares/likes, giving average levels of engagements per tweet of 16 and 10.1. But Italy and Switzerland overtook Germany and Spain to secure third and fourth spot. The dominant fours capitals took top places in the city stakes.Source: Roland BergerTwitter use is highest among the leaders.France and United Kingdom host the most dynamic communities on twitter.Leader Rising star FollowerNumber of tweet (millions)FRA ITAGBR NLDDEU CHEESP RUS BEL LUX2.571.970.630.530.470.30 0.300.250.180.1231%24%8%6%8.2TOTALJoining the dots A map of Europes AI ecosystem 9STARTUPS - Volume vs. density.Leaders are home to 67% of AI startups, but Nordic and Baltic countries show density dynamism.Source: Roland BergerLeader Rising star Follower Startups (#)1.41.21.00.80.60.40.20.08508000 450150 65030050 500 550200 700350100 600250 750400Startups / pop. ratio (%)Startups (#)United KingdomFranceGermanySpainIrelandEstoniaFinlandSwedenSwitzerlandDenmarkAustriaLithuaniaNorwayBelgiumCzech Rep.PortugalHungaryBulgariaItalyPolandRomaniaGreeceLuxemburgMaltaCyprusLatviaSloveniaSlovakiaCroatiaNetherlands10 Joining the dots A map of Europes AI ecosystem