At Rise of AI 2019 in Berlin, Hans Uszkoreit compares the state of AI research and commercialization in China, the United States and Europe, and challenges Kai-Fu Lee's claim in AI Superpowers that Europe is not even competing for a bronze medal in the AI race. Uszkoreit argues Lee's framework focuses only on commercialization and product research, and that a fuller picture including basic research tells a different story: Europe still leads in publication volume and produces many of the pioneering researchers behind deep learning, even though the US dominates top-conference papers and best-paper awards and China leads in scale of funding and market size. He attributes US strength to decades of strategic DARPA funding that created a virtuous circle of research, commercialization and venture capital, and attributes China's fast growth to government dedication, market size, public awareness of AI figures and pragmatism about shipping minimum viable products. He closes by arguing Europe has real chances in enterprise AI, robotics, health and smart energy, and urges closer research cooperation between Europe and China rather than reliance on the US.
Key takeaways
- Uszkoreit argues Kai-Fu Lee's ranking of the US and China ignores basic research, where Europe still leads in publication volume.
- He credits decades of strategic, long-term DARPA funding for creating a virtuous circle behind US success stories like Google Translate and GPS.
- He says China's growth stems from government dedication, market size, public name-recognition of AI researchers and pragmatism about shipping minimum viable products.
- He states real US-European AI research cooperation was historically rare, based on his own 30 years working in the field.
- He sees Europe's best chances in enterprise AI, robotics, health and smart energy, combined with closer cooperation with China.
Detailed notes
Questioning Kai-Fu Lee's framework
Uszkoreit starts from Kai-Fu Lee's book AI Superpowers, which splits AI into internet AI, business AI, perception AI and autonomous AI, and scores the US and China against each other while leaving Europe out entirely, with Lee reportedly telling a journalist that Europe is not even in the running for bronze. Uszkoreit says he largely agrees with Lee's scoring in those four categories, including that China is already more advanced than commonly reported in autonomous AI, but argues the framework only measures product research and commercialization, not basic research, where Europe remains strong.
What the publication and talent data show
Uszkoreit presents figures showing Europe produces the most AI papers overall, though China is catching up quickly, while the US leads at top conferences such as NeurIPS 2017 and dominates organizations with two or more best-paper awards, ahead of European and then Chinese universities. Looking at a list of pioneering deep learning researchers, he shows many originated in Europe before moving to the US, illustrating a brain-drain pattern: Americans rarely emigrate, many Chinese researchers move abroad and some return, and many Europeans move to the US with some return, but there is little direct movement between Europe and China.
Why the US and China grew fast
Uszkoreit credits long-term, strategic DARPA funding for US success stories including autonomous driving, Google Translate (started by a German researcher), the internet, GPS and statistical machine translation, arguing this created a virtuous circle: funding produces breakthroughs, breakthroughs attract venture capital, and venture capital fuels commercialization. He says China has understood and copied this mechanism better than Europe, combining strategic government funding, a Thousand Talents program, an enormous domestic market and, in his view, more pragmatism about shipping a minimum viable product and improving it after launch, despite a persistent language barrier limiting international expansion.
Europe's chances
Uszkoreit argues Europe's pressure-free academic climate and freedom to work on unpopular topics sustain excellent basic research, and identifies enterprise AI, robotics, health and smart energy as areas where Europe, lacking its own Google or Facebook but having companies like SAP, still has real opportunities, particularly in combining German robotics expertise with Chinese manufacturing scale. He states that after 30 years working in AI he found genuine US-European research cooperation was rare, largely limited to a handful of jointly funded projects, and argues Europe should instead pursue cooperation with China, which needs European basic research and enterprise-AI expertise even as it takes the lead in commercialization.
Quotes
In the race of AI, Europe is not even in the running for bronze medal.
Kai-Fu Lee (quoted by Hans Uszkoreit)
China will take the lead in commercialization and huge markets, I think there's no way around.
Hans Uszkoreit
We had the illusion that we were working with the US, but that only happens on a corporate level.
Hans Uszkoreit
Questions answered in this talk
Does Europe have any chance in the AI race between the US and China?
According to Hans Uszkoreit at Rise of AI 2019, Europe still leads in basic research and publication volume and has real chances in enterprise AI, robotics, health and smart energy, even though it lags badly in commercialization.
Why did the US become dominant in AI commercialization?
Uszkoreit attributes US strength to decades of long-term, strategic DARPA funding for projects like the internet, GPS and self-driving research, which created a virtuous circle of breakthroughs, venture capital and successful commercialization.
Why is China's AI sector growing so fast?
Uszkoreit points to strong, dedicated government funding at all levels, a huge domestic market, a Thousand Talents program to attract researchers, public awareness of leading Chinese AI figures, and a pragmatic willingness to ship minimum viable products and improve them later.