AI Race between the US, Russia, and China… and where is Germany?
Disclaimer: the expressed opinions are my personal ones and are not necessarily those of Georgia Tech or ML@GT.
It pretty much does not matter what position one has on Artificial Intelligence (AI) and Machine Learning (ML), but the strong interest in AI and ML can hardly be denied. In fact, doing a quick informal poll among colleagues and friends, it seems that many Thanksgiving table and holiday conversations were dominated by either AI or cryptocurrencies (Bitcoin and the likes). Both of these technologies have the potential for significant disruptions and whether or not they are overhyped (they probably are) they provide huge fundamental value.
While I will write about cryptocurrencies elsewhere, here I want to talk about AI and the global race for leadership in that space. AI is dominating many discussions last but not least due to its sci-fi appeal and the thought experiments of various thinkers and leaders. This interest in the technology has reached a point where AI is now a topic of politics and important verticals of interest include ethical questions (e.g., killer robots and the terminator scenario), AI security and safety (provable guarantees for the operating envelope and AI hacking), as well as the impact of AI on armed conflicts. In fact the question of leadership and dominance in AI research, development, and deployment has become central over the last few months with various world leaders and politicians acknowledging the potential role of AI. For example, Vladimir Putin recently said:
“Artificial intelligence is the future, not only for Russia but for all humankind.”, “Whoever becomes the leader in this sphere will become the ruler of the world.” — Vladimir Putin [Source: Wired]
The current discussion is dominated by the idea of an arms race between the US, Russia, and China [see e.g., here, here, and here] due to its potential and risk of a significant power redistribution around the world and as such is considered being of highest political relevance and national interests. A significant part of this discussion is dominated by fear and winner-takes-all scenarios particularly fueled by Silicon Valley household names and scientists. So where does all the fear come from? If the entrepreneurs and investors from the valley have understood one thing then it is the power law (sometimes also Pareto effect or 80–20 rule) and the related Matthew effect. The Matthew Effect basically asserts that “more gains more” and the power law stipulates that a small fraction, say 20%, of a group is responsible for the majority, say 80%, of an effect. These two effects have been widely observed in society (e.g., wealth distribution in countries etc). With the advent of the internet this effect became significantly more pronounced and we saw an accelerated version of the power law: even smaller fractions, say 1%, accounted for even larger majorities of the effect, say 99%. This law has been guiding many investment decisions in the valley and Amazon, Google, Facebook, and partially Uber are great examples of this force being at play. The belief is now that AI will result in a significantly further accelerated version of the power law due to its potential impact: 0.0000001% of the participants will generate 99.9999999999% of the effect, or put differently, the winner-takes-all. Marc Cuban aptly summarized this in a recent interview with CNBC:
The world’s first trillionaire will be an artificial intelligence entrepreneur — Marc Cuban [Source: CNBC]
With such a pronounced difference between the first and the second, the saying “the second is the first loser” reaches a completely new level of severity.
This sounds all very serious. So where is Germany in the picture? Looking at GDP, Germany is the 4th largest economy (after the US,China, and Japan) and ranks 3rd in exports after China and the US. From howmuch.net:
So one would naturally expect that Germany has a significant interest (and the necessary resources) to strive for leadership in AI, at least in specific domains — a domain that immediately comes to mind is manufacturing. However a recent article on economist.com draws a different picture, showing that Germany is significantly falling behind not just the US and China, but also several other countries.
A systemic issue.
It is not hard to understand why Germany’s presence in AI is not as strong as expected. Germany never had a startup culture or a strong tech culture and its large companies are too big to be innovative. Moreover, it seems that the mind set is not aligned with innovation: Failure is not an option and in fact constitutes life-long stigma whereas in the US a culture of fast failure promotes innovation. On top of that, due to labor laws and company structure several large companies suffer from legacy leadership that missed the point of adaptation. I will use the German automotive industry, one of the main, dominating industries, to exemplify the challenges. The German automotive industry focuses on engineering. However, the next evolution in the automotive space, autonomous cars are (mostly) a software problem not an engineering problem — we know how to build a great car (engineering problem: solved) but we do not know how to teach it to drive by itself (software or AI problem: unsolved) and in fact highly innovative companies in that space are not traditional car companies (e.g., Apple, nuTonomy, Intel, waymo, Uber). As a function of this basic mindset fallacy and strong resistance to change, German automotive companies have fallen behind now scrambling in a desperate attempt to play catch-up. Competing for top talent (not paying enough: robotics grad pay packages run at $200k/year in Pittsburgh) and technology (not researching enough: the US has a large number of AV startups). Surely, the German giants are still building great cars but their role might be reduced to become “Zulieferer” (Suppliers) for AI companies that build the autonomous technology and/or vehicle. In fact several companies have shifted to providing retro-fit kits for cars and for example in Tempe, AZ (or Pittsburgh or San Francisco for that matter) you will regularly see retro-fit, autonomous Volvo’s on the street functioning as autonomous Uber vehicles collecting important driving data for refinement and training (not always without challenges). Germany banned Uber and tests with autonomous vehicles have been only recently (05/2017) allowed in Germany… The next challenge after the car is the fleet problem: how to effectively manage a fleet of autonomous vehicles and what are models of use? Same story all over.
Mittelstand is not enough and Legacy Leadership.
A typical argument that is often brought forward is that Germany’s strong Mittelstand (think of, often family-owned, SMEs with workforce supplied by Germany’s apprentice system) will fix the issue by delivering the necessary innovation. In the past this has often been the case and in fact large parts of automotive suppliers come from that space. This was easy: process fine-tuning and kaizen-style hardening of manufacturing. However, when it comes to AI a much more concerted effort with pooling of top talent is needed. Critical mass is required and smart people bumping into each other, not an R&D function with a “small r” and a “huge D” where innovation is an artifact that has been replaced by most precise, process-engineered, rehashing of silicon valley ideas. Some of these heavy weights literally have “being a fast follower” as strategy-mantra: “you go first” as innovation strategy…
The above is exacerbated by German labor laws that force legacy leadership on companies. Employees grow through the ranks, rather than hiring leadership from the outside and at the end of this growth process managers are out-of-touch with current technological advances and innovation. In the US, companies strongly fight this by aggressive firing and (re-)hiring to ensure a dynamic mix. Strong labor laws in Germany though cement legacy leadership: the longer an employee is with a company the harder it is to get rid of the employee and this is almost orthogonal to the actual performance of said employee. Now, these managers with a legacy mindset “hire what they know” and what does not threaten them, their position, or their worldview, effectively leading to companies that try to solve today’s challenges with tools from the 60s…
University-driven public AI research.
The cynical reader might now point towards Japan and their 会社 (Kaishas), these huge conglomerates such as Sony, Toyota, etc. that meticulously groom their employees through years of training within the company. While this is indeed a very similar approach to in-company-growing, the major difference is that the Japanese government and industry interest groups strongly invest in innovation and AI through large-scale initiatives, such as RIKEN-AIP; Japan also has a Google Lunar XPRIZE team…
But everything is going well and the world is jealously looking at Germany because of its strong exports. So how can it be so bad?
These exports and productivity stem from products and services that where created with the aforementioned legacy model at the end of the current S-curve: cars and other relatively pure engineering products. However little has been done at the beginning of the next S-curve with some notable exceptions such as e.g., KUKA... oh wait, actually they recently got bought by China’s Midea (see here and here)… my bad…
What is required is some concerted effort of the German government to strengthen Germany’s position in Machine Learning, Artificial Intelligence, and more broadly data-driven methods. Germany did an outstanding job in establishing itself as one of the leaders in optimization and operations research with extremely strong groups spread throughout Germany and reasonable industry uptake of these methods. This made and makes a lot of sense as it is aligned with the need for improving manufacturing processes and more generally engineering excellence. The next evolution of these methods however requires a tight integration with AI methods and data and Germany is lacking severly behind (with few exceptions). While in several other countries talent is formed, honed, and groomed through data competitions and a huge number of startups in that space that bite their teeth out on challenging data and prediction problems, defining and redefining best practices at a mind-boggling pace, German companies and data innovation are crippled by German/European privacy laws. To be clear, I am not against reasonable privacy laws and these are only one piece to the puzzle, but surely the current ones in place are not accelerating innovation.
Don’t hate the player — hate the game.
In the end, there is a very simple proxy when it comes to innovation in the digital realm (including AI and more broadly the data space): the US has Google, Facebook, Amazon, Apple, Uber, … Russia has Yandex, VK.com, … China has Tencent, Alibaba, Baidu, DJI, … and Germany? Cheating car manufacturers is the first thing that I regularly hear. I guess after all, it really is about “being first, being smart, or cheat” and since “first” and “smart” were already taken what else is left? The fallout from innovating-through-cheating has not only severely damaged the “Made in Germany” brand but also sends an even stronger message: the German innovation concept has completed its life cycle and is end-of-life and Germany needs a new mindset to be competitive in the AI, ML, and data space.