AI: Arms Race 2.0
An analysis of the current state of the competition between US, Europe, and China in AI, examining research, patent publications, global datasphere, devices and IoT, people, and more.
On February 11, President Trump signed an executive order outlining the American AI initiative. Among other things, the order discussed the need for the U.S. to maintain its current leadership in AI. This was followed by another announcement from the Department of Defense on February 12th, which released the summary of its strategy on artificial intelligence.
And yet, one can argue that continued U.S. leadership is far from certain: in particular, as we’ve discussed in “Reflections on the State of AI: 2018”, China has already surpassed the U.S. in terms of investments in AI startups, with close to 50% of AI investment dollars going to Chinese startups (in terms of the number of deals, the U.S. is still in the lead, although the share of AI startups hailing from the U.S. has been steadily declining over the last few years).
Source: Top AI Trends To Watch In 2018
China is now also challenging the U.S. in terms of both the number of patents and publications in the field. True, the quality of some of those publications might still lag behind the U.S., but China has been catching up, and the rate of its advancements in the field over the last few years has been simply staggering.
The desire to dominate AI space is quite understandable — after all, the idea that AI would one day enable a whole new world of possibilities has been around for decades. Until recently, however, it was largely relegated to the realm of science fiction and the works of select few researchers & futurists. It all started changing in the early 2010s when the technology and, perhaps equally as important, the computational resources, finally caught up, and we got AI (or rather, machine learning) capable of solving real-world problems for the first time.
As it usually happens with any kind of game-changing advancements, however, different countries have found themselves facing new opportunities and challenges offered by AI in vastly different circumstances.
For the rich western democracies, the emergence of machine intelligence offers opportunities to explore new frontiers, build a new generation of successful companies, and further improve their societies. However, it also means having to face the dangers that AI could pose to their citizens if applied recklessly. In the last few years, that meant increasingly prioritizing the “no harm” approach when devising the AI policy — the West, with its emphasis on individualism and strong human rights record, simply has more to lose and less to gain when it comes to AI, compared to other places. While the West, and more specifically, the U.S. might still lead the way in AI research, it’s the implementation that is going to be harder and more challenging, considering a different level of expectations it faces around ethics & privacy concerns.
In contrast, China is facing a different set of challenges altogether: given its historical context and the stage of economic development, the opportunities potentially stemming from AI often outweigh the dangers of its abuse, which in turn has led to embracing AI and executing an aggressive investment and deployment strategy.
It’s also worth noting here that with the broad AI deployments, China and the West might be optimizing for different results. In China, it would often be optimized to deliver the best results for society as a whole, even if it means inadvertently harming minorities in the process. On the contrary, the West focuses on human rights and fair treatment of every person, including any outliers, which in turn creates unique challenges for AI adoption.
As for the rest of the world, most countries today fall somewhere in between the extremes represented by the West and China.
Now, let’s dig a bit deeper into the key factors that will determine the leader in the currently unfolding global AI arms race.
Building on what we’ve discussed above, we propose segmenting the world into 3 major groups: the West, China & the rest of the world. Obviously, such segmentation is quite subjective, but we believe it frames the conversation around AI policy in a useful way.
Now, when thinking of any problem that could be tackled using machine learning, there are three building blocks to be considered: data, people, and money.
Source: Evolution One
Note: Quantities of each resource here are subjective, and serve the illustrative purposes only; we will elaborate on how we’ve got to those for each section below.
The last couple of decades has brought us tremendous growth in the amount of data generated, and there is no sign of it slowing down — rather, if anything, it’s been accelerating in the last few years, driven by our ability to generate ever-increasing amounts of information, as well as the explosion in the number of sources for new data, both on the hardware & software side.
According to IDC, today more than 5 billion consumers already interact with data every day, and this number will increase to 6 billion by 2025. Still, while in the early 2010s it was the smartphones that were responsible for the bulk of the growth in the amount of data, going forward the growth will be driven more and more by the IoT devices, which are now expected to generate over 90 zettabytes of data per year by 2025 — over 50% of all data forecasted.
One thing worth underscoring here is that the relationship between the number of devices & the amount of data they generate has never been linear, but nowadays, this is becoming especially true. While in the late 2000s & early 2010s, it was the growing penetration of smartphones, coupled with the declining costs of transferring & storing data, driving the amounts of data produced, there were obvious upper limits on the number of smartphones that can be in use at any given time. However, today, at 3 billion smartphones in the world, the growth is slowing down, yet the amount of data is growing fast as ever.
Source: State of the IoT 2018
There are 2 key factors at play here.
First, while smartphone growth is slowing down globally, IoT represents a different story. As of 2018, there were at least 7 billion IoT devices (with other estimates putting this number significantly higher), posed to grow to 21.5 billion by 2025, surpassing all the other categories combined. Perhaps more important than a specific number of devices is the fact that there is no natural limit to the number of IoT devices that can be put out there: it’s quite possible to imagine the world where there are dozens or even hundreds of devices per every living person, measuring everything from the traffic on the roads to the temperature in our apartments (and this is even before accounting for the IoT devices used by enterprises).
Second, the amount of existing data is to a significant extent defined by our willingness and ability to collect, share and store it (be it temporarily, or permanently). And here, the choices we make around what types of data we are willing to collect and retain are becoming crucial — any data that’s not captured today is by definition lost, and this effect is compounding over time.
Imposing restrictions on data collection out of concern for people’s privacy and to prevent potential abuses might be a reasonable thing to do, but in the narrow context of machine learning, those choices affect the amount of data available to train the models on. This, in turn, means that countries less concerned about privacy (with China being a prime example — for instance, see its experiments with AI-powered security cameras to catch criminals) will likely gain an edge when it comes to data.
That being said, it’s also important to recognize that privacy concerns aren’t applicable to every single problem, and there are some fields (such as driverless cars, or machine translation — see some interesting expert opinions here) where the West would actually have better datasets.
People represent the second crucial building block, as it is they who define the approach used to tackle any problem that could be addressed with machine learning.
Here, the situation is somewhat opposite of what we’ve seen in Data — the West, and the U.S. in particular, has a natural advantage, stemming from the fact that it remains one of the most desirable locations to work and live in, and thus has an easier time attracting people from all over the world. It could also be more tolerant towards unorthodox ideas, which provides for a more creative environment and helps to find and nurture innovative ideas.
In fundamental research, the U.S. has also historically had an advantage, thanks to its established system of research universities, not to mention its ability to attract top talent from all over the world. Still, in recent years, China has established a system of top-tier research universities and continues to aggressively invest in it. Today, China is already conferring more doctoral degrees in natural sciences & engineering and produces more articles in peer-reviewed journals than the U.S., according to the Economist. Moreover, in AI-specific research, the U.S. lead is even less certain, as was mentioned before (see CB Insights report for details).
Finally, when it comes to the practitioners who are focusing on implementation (rather than pure research), both the U.S. and China have some unique strengths; two possible proxies to evaluate those are the number of startups founded in each respective country, and the number of professionals joining the field.
The U.S. has the highest number of startups and also an established ecosystem of big tech companies such as Google, Microsoft, and Facebook investing in the field. Still, China is #2 here (#3, if looking at Europe as a whole); moreover, it receives an unprecedentedly high amount of investments (more on that in the section below), and is also a home to select few companies that could rival the biggest players in the U.S. (namely, Alibaba, Tencent & Baidu).
However, in terms of the workforce, China has a clear lead — today, it produces 3 times more college graduates with STEM degrees compared to the U.S. that faces chronic shortage of qualified personnel. Unlike in research, where it is the select few who often matter the most, with the practitioners, numbers do matter, and producing enough engineering and science major becomes crucial to establish and maintain leadership in the field.
According to CB Insights, investments in Chinese startups contributed 50% of the dollars invested in AI startups in 2017 globally, growing from just 11.6% in 2016. It comes as no surprise then that top 2 most well-funded companies in 2018 — SenseTime and Face++ — were both from China — we have already briefly discussed AI investment landscape of 2018 in our recent article and concluded that China already leads the race today when it comes to early-stage investments.
Still, now that President Trump has announced his American AI Initiative, we feel it might be a good time to go back and consider how this announcement affects the balance of power.
Before we do that, however, let’s pause for a second and think through the funnel that could help analyze the efficiency of the investment strategy and determine its ultimate success or failure.
Source: Evolution One
The following three steps could help to frame the discussion:
- First, consider the overall size of the proposed investment, and whether it would be enough to achieve a meaningful difference given the stated goals
- Second, consider how efficient and developed the ecosystem supposed to absorb the funds already is
- Finally, determine how focused the proposed strategy is and whether it targets the right areas that have the potential to yield the best possible returns (the areas themselves would differ based on the overall goal — e.g. supporting an already established and well-developed ecosystem might require a different strategy than when building the basic institutions from scratch).
Now, applying this framework to evaluate President Trump’s strategy for the AI, one could safely conclude that it doesn’t really change anything, given how vague and generic it is. That is not to say that the U.S. is falling behind China when it comes to investments — rather, it becomes clear that both countries are equally well positioned in terms of the amount of funding available, the robustness of ecosystems and availability of multiple areas to focus on that pose significant opportunities for advancement.
While today many view AI as a new arms race, where countries are posed to fiercely compete against each other (and the tone of President Trump’s announcement doesn’t help the matter), we believe that collaboration in AI leads to consistently better outcomes for all.
Interestingly enough, the West is particularly likely to benefit from promoting global collaboration (more than its counterpart that is better positioned to thrive in a siloed world), as it was the freedom to think and create that historically made places like U.S. attractive for talent from around the world.
The route to sustainable leadership in AI for the West would likely rely on:
- Focusing on fostering global collaboration, including researchers and companies from places like China
- Investing in the development of frameworks for ethical usage of AI, while also paying attention to not putting undue restrictions on the initiative of private businesses
The role of the Western governments should thus be focused on helping to frame and guide the discussion, rather than trying to impose unnecessary restrictions stifling innovation.
Bio: Alex D. Stern is a Product strategist, with stints in venture capital and startups, currently working for Microsoft's Cloud & Enterprise.
Eugene Sidorin is an Entrepreneur, physicist, and tech geek, currently working at Microsoft, focusing on Azure finance & strategy.
Original. Reposted with permission.
- On-line and web-based: Analytics, Data Mining, Data Science, Machine Learning education
- Software for Analytics, Data Science, Data Mining, and Machine Learning
- Reflections on the State of AI: 2018
- Two Major Difficulties in AI and One Applied Solution
- State of the art in AI and Machine Learning – highlights of papers with code