One objection to working on AI governance in the European Union (EU) is that the EU is not a superpower in AI. The EU is not currently leading AI development and will probably not lead AI development in the foreseeable future either. According to the objection, if AGI gets developed elsewhere, then EU AI governance will have little impact on ensuring that it gets developed safely and is aligned with human values. As a caveat to this post, it is not entirely clear that the EU is not a superpower in AI even when it is not leading the rankings or that it is not relevant to AGI development even when it is not the leader in AGI investments. Finally, there could be other pathways to impact besides direct AGI development.
Top AI companies in the world are US and Chinese-based. Top companies explicitly trying to develop AGI are all US-owned. According to Disfold, among 30 of the best funded AI startups in the world, 16 are in the US, 9 in China, 3 in the UK, 1 in Canada, and 1 in Singapore.[1] “Of the 2,451 AI new companies that Statista reports as of 2018, 675 have a place with European nations (the UK had 245).”[2] According to McKinsey, “Europe attracted only 11 percent of global venture capital and corporate funding in 2016, with 50 percent of total funds devoted to US companies and the balance going to Asia (mostly China). That share was about the same in 2018. Only four European companies are in the top 100 global AI startups.”[3] While funding for startups in Europe seems to be rapidly increasing, with total investments surpassing €120bn in 2021, triple the value of 2020, the regional gap in top AI-companies is unlikely to be significantly reduced in the near future.[4]
According to the Center for Data Innovation’s 2021 report, “China’s AI capabilities relative to the EU and the United States have improved in several ways. First, China has surpassed the EU as the world leader in AI publications. Second, the quality of its AI research has generally trended upward year to year. Third, its software and computer services firms have increased their R&D spending. Fourth, China now has nearly twice as many supercomputers ranked in the top 500 for performance as the United States—the United States led in this indicator as recently as 2017. Finally, China likely continues to lead in the amount of data generated. Overall, however, China has not significantly reduced the gap in AI between itself and the United States, but its trend of consistent progress could eventually evaporate the U.S. lead.”[5]
According to the Global AI Index, the United States of America is leading AI development in the world followed by China on the benchmarks of talent, infrastructure, operation environment, research, development, government strategy, and commercial. European countries such as the UK, Germany, the Netherlands and France are in the top 10.[6] This ranking doesn’t consider the EU on the whole, however.
According to the AI Index 2019, Europe has consistently been the largest publisher of AI papers, but recently China has surpassed it and has been growing the most from 2000 to 2018. The US is third in terms of number of AI papers. China has 28% of AI publications, Europe has 27% of AI publications, and the US has about 19%. The citation impact of US publications is about 50% higher than China’s. North America accounts for over 60% of global AI patent citation activity between 2014-2018.[7] These data do not distinguish between the EU and the rest of Europe.
According to the Global AI Talent Tracker, “the United States has a large lead over all other countries in top-tier AI research, with nearly 60% of top-tier researchers working for American universities and companies.” And “China is the largest source of top-tier researchers, with 29% of these researchers having received undergraduate degrees in China. But the majority of those Chinese researchers (56%) go on to study, work, and live in the United States.”[8] Most AI researchers want to move either to the US, UK, or Canada.[9]
Who will lead AI in the foreseeable future? According to an article in Nature, China has increased its share of authorship of the top 10% most-cited papers while the USA’s share is declining.[10] The article includes a prediction that “it could take 5–10 years for China to reach the level of innovation in fundamental theories and algorithms occurring in the United States and the United Kingdom — but that it will get there.”
As far as we are aware, the EU is not usually considered in these discussions about future AI leadership. The European Investment Bank reports that there is an "annual shortfall of up to €10 billion in investments to keep the European Union in the global artificial intelligence and blockchain race."[11] That said, according to the AI Watch, EU AI investment in 2019 compared to 2018 increased by 39%, is on course to reach €22.4 billion by 2025 and surpass the €20 billion target by over 10%.[12]
As far as we are aware, the EU currently has very few companies that are explicitly trying to develop AGI.[13] According to the Global Catastrophic Risk Institute's 2020 project, these are Aleph Alpha, Animats, Alice In Wonderland, Curious AI, EleutherAI, Fairy Tale AGI solutions, FLOWERS, GoodAI, Mauhn, SingularityNET, Xephor Solutions. We do not have a full picture of total funding in the EU, but Aleph Alpha seems to have about €30m in funding[14], while SingularityNET has raised €41m so far.[15] The Curious AI Company seems to have been discontinued, with the team moving to do AI research for a top tech company. This is minor compared with the US where OpenAI and DeepMind on their own surpass everything else on AGI development and generally the EU AGI initiatives do not appear to have achieved much noteworthy. We are not aware of Chinese companies explicitly trying to develop AGI.
Thanks for all the numbers. I think putting them into plots would make the case even easier to understand, especially when talking to policymakers and other influential people who get a wall of numbers thrown at them every day.
If you currently have little time, just taking the most important stat and putting the respective plot on top of the article gets you quite far already.
According to this CSET report, Europe (especially the Netherlands, UK, Germany) plays a role in the semiconductor supply chain. Is this significant enough to grant Europe a "seat at the AI table" in the future?
Interestingly both EU, China, and the US are taking large steps to become less reliant on each other for semiconductor production. My guess is that, however strong an argument this is currently, it will be less so in the future.
I wonder to what extent governments will be successful in this effort. Are semiconductor companies easy to replicate or should is it a field where we should expect the current leaders such as ASML (Dutch) to keep lead even after governments pour in money?
What do you think of the "Brussels Effect" as a way EU AI law may matter?
I think this article is a spin-off of the one below where the Brussels effect is discussed: https://forum.effectivealtruism.org/posts/fbG6wWZhJ3jt3xHxS/is-the-european-union-relevant-for-agi-governance
Seems a race to the top dynamic only applies when a company wants to operate in the market with stringent policies, not when you're in the R&D phase, which is where the biggest concerns are.
Interesting to see that the EU punches well above its weight on academic publications and citations, while lagging behind on funding, top talent, and other metrics. Does that indicate anything about the types of AI projects being pursued in the EU, and what the high impact opportunities might be?
Yeah, I don't think the EU should be the priority, but it makes sense to at least experiment to see how much influence we can have here.
Agreed. While the cause itself might not be the most impactful, work on EU AI policy could build valuable career capital for individuals that allows them to have a greater impact in the US or China years down the road. It seems like there are more entry level AI policy jobs in the EU, which could help early-career practitioners.