The 4 waves of AI: Does the winner take it all?
Last June, I read Kai-Fu Lee’s book “AI Superpowers: China, Silicon Valley and the New World Order“. A book that was kind of preparing me for topics that I have been gradually getting to know from a whole new perspective since the beginning of July 2019 when I started working at Parashift.
Kai-Fu Lee is Chairman and CEO of Sinovation Ventures and President of the Artificial Intelligence Institute of Sinovation Ventures. He founded Sinovation Ventures back in 2009 and is now at fund number 7. The team manages a portfolio of over $2 billion and supports more than 350 technology companies across China. Prior to that, Kai-Fu Lee was President of Google China and supported companies such as Microsoft, SGI and Apple in various executive positions.
In his book, Kai-Fu Lee discusses what the driving factors in the AI ecosystem are, how China is aligning its infrastructure to support AI, the direction he anticipates the AI revolutions to take, and how we need to adapt to the new environment.
He also makes a comparative assessment of the level of progress based on his perspective, which provides him with extensive insights into AI research and business applications in the two leading nations – China and the US. He differentiates between 4 AI Waves:
By now, Internet AI has probably more or less reached us all. This particular domain is basically about companies like Google, Facebook and Amazon who have access to huge amounts of data and are using it to improve the user experience. To do that, they mainly use recommendation algorithms which detect behavior patterns based on user behavior on the services. Based on these insights, the services curate personalized content to optimize for X. Most of the time, it’s user engagement.
One of China’s leaders in this category is Jinri Toutiao or ByteDance in English. Founded in 2012, the company is often referred to as “the BuzzFeed of China” because they too have built a hub for timely viral content. However, compared to BuzzFeed there are significant differences. Toutiao does not have people as editors like BuzzFeed does. No. Toutiao’s editors are algorithms. They browse the internet for content and use technologies such as Natural Language Processing (NLP) and Computer Vision to process and analyze articles and videos from partner networks and commissioned contributions. In other words, to understand what the content is all about. The individual user footprint, which includes clicks, articles read, videos watched, views, comments and much more, then serves as the basis for personalizing a news feed that is talior-made for the user’s interests. In this process of personalization the algorithms sometimes edit the title of articles to increase the click rates. And the more users click through the pages and actively “label” data points, the better the recommendation algorithm becomes. The result is increasingly specific content.
In a national comparison, China and the USA are neck and neck. However, a rather trivial fact is favorable for China’s future pole position: China alone has more internet users than the USA and Europe combined. In addition, China already offers a mobile-first infrastructure that enables seamless payments on the internet and thus stimulates the development of creative and economically viable internet applications.
“There’s no data like more data” – Robert Mercer
The second wave, the one of Business AI, makes use of already labelled company-specific data. In some cases, this data goes back several decades and is therefore very valuable for the development of more accurate prediction models. For example, banks and insurance companies, but also a number of medical institutions, are usually sitting on very large data sets where things like credit histories, claims and fraud cases or archived diagnoses and health status developments have been recorded and stored for years.
Prediction models developed on such data sets are particularly valuable as machine predictions are more subtly structured than ours. While we humans base our predictions on obvious connections (so-called hard features), AI-based predictions additionally contain subtle connections (weak features) that appear irrelevant to us in the overall picture. The more data the machine can process, the higher the probability to find more relevant correlations and thus increase the quality of the predictions significantly. Pretty straightforward.
So, as long as there is a large enough data pool with structured respectively categorized data with relevant outcomes, these technologies are able to outperform even top-notch experts in their analytical tasks. A good example of the added value of these technologies can be found in their use for the diagnosis of diseases (interesting results were recently obtained in the case of breast cancer). Here, AI is brought in to support the expertise of field experts and thus increases the accuracy of the diagnosis.
Already in 2004, companies such as Palantir and IBM Watson became active in the Big Data sector and offered consulting services to companies and governments based on their expertise. For a long time, these players were market leaders in what they did. But when Deep Learning, a special technology associated with Machine Learning, spread rapidly in both exploration and application in 2013, new players such as Element AI and 4th Paradigm also came into play.
While US companies today have a clear advantage in the immediate and profitable implementation of Business AI, Kai-Fu Lee expects China to become a leader in the use of AI in public services and also in some individual industries where legacy systems are still in use these days. For example, China’s relatively immature financial and healthcare systems provide strong incentives to question how services such as consumer credit or medical care should be designed and delivered. And this is precisely where Business AI can come in and turn weaknesses into strengths through a radical bottom-up transformation of structures and processes.
As the name suggests, Perception AI is about giving machines senses and thus opening up the spectrum for context. The result is a fusion of the digital and physical world.
Algorithms learn to group pixels of photos and videos into relevant clusterings and recognize objects in a snapshot, pretty much identical to how we do it. Similar with audio data. Here too, algorithms gain a better understanding of individual words over time and learn bit by bit about the meaning of sentences and words in different contexts.
Essential for progress in this specific domain is the digitization of our environment through sensors and other smart devices – keyword Internet of Things (IoT). So when you talk to Amazon’s Alexa or drive a Tesla, for example, you are making a major contribution to the further development of such technologies.
Kai-Fu Lee says, “Perception AI will bring the convenience and abundance of the online world to our offline reality”. Various sensor-based hardware devices act as connectors. A technologically quite remarkable example, which can already be used conventionally in some locations today, is Amazon Go.
The fact that China is the leader in Perception AI and can still make drastic gains from here should come as no surprise. The Chinese culture and its somewhat relaxed approach to privacy as well as Shenzhen’s strength in hardware production form the basis for a relevant edge in the global competition.
To me, the most interesting wave is the wave of autonomous systems. But at the same time it is also the one that is the most difficult to assess in its development. The wave is based on all previous AI milestones and aims at creating systems that can act completely autonomously (i.e. without any human interaction). To achieve this, these systems must not only have a representation of an environment, but must also be able to react to changes in it and to cope with potential deviations and irregularities. For the majority of you, autonomous driving will probably be the application that comes closest to your mind. But also apart from that particular use case, autonomous systems will gradually change many other areas of our daily lives.
According to Kai-Fu Lee, China must be making incredible investments in this area of AI. Moreover, they are also using their political structures to bring about goal-oriented actions faster and gain a competitive advantage. As an example: In the Zhejiang province, regulators and government officials have begun planning China’s first intelligent highway. It is equipped with all kinds of sensors, has solar panels embedded in the ground and enables wireless communication between car, road and driver. The aim is to increase traffic efficiency by 30 percent and significantly reduce accidents. Fun Fact: In the future, these roads should continuously charge autonomous vehicles while driving.
Another interesting project is being developed near Beijing. In Xiong’an. There, over the next 20 years, $580 billion will be invested in infrastructure. The goal of the project is to build the first city designed for autonomous vehicles of any kind. In this context, Baidu, one of the leading companies in AI, is working closely with the government to move the project forward as quickly as possible.
Nevertheless, in 2018, the USA was the leader in autonomous systems. Silicon Valley companies in particular have a substantial lead in research and development. Google began testing self-driving cars in 2009 already and many of the engineers who were involved at the time later started their own startups. In China, this movement did not start until 2016, where companies such as Baidu, Momenta, JingChi and Pony.ai are particularly strong and are rapidly catching up.
Kai-Fu Lee makes the question about the long-term leader in this sector dependent on the following question: Will the predominant bottleneck in the context of the widespread use of such technologies be technological or regulatory in nature? If the decisive factor is policy, China will most likely have a relevant advantage. Otherwise, the USA. The only chance for China to still have a say at the technological level would be if new and unexpected advances in computer vision were to spread rapidly around the world, thereby closing technological gaps. Then again, China’s regulatory advantages would be a trump card.
Does the Winner take it all?
So now, an interesting question is in what way and to what extent the leading forces will be able to make the predicted economic added value their own. Because there is an incredible amount at stake. According to a PwC report, AI products and services alone can be expected to generate an additional economic output of $15.7 trillion – yes, you read correctly: trillion, not billion – by 2030.
I find this question particularly exciting because AI has a characteristic and natural tendency to create monopolies. This means that in many industries of the 4 Waves there is an economic dynamic, where in fact only a substantial winner is the result. We already know this from various internet companies. Google dominates search, Facebook social networks, where opportunities for new players may slowly emerge, and Amazon, which is quite consistently strengthening its dominance in e-commerce. The situation will be no different for AI companies. Rather more radical. And so, various American and Chinese companies, which are leaders in their specific field, will be able to build an incredible value creation concentration. What impact this will have on all sorts of systems in our world is a question where, ideally, we can achieve greater clarity very soon.