The Four Waves of AI as Described by Kai-Fu Lee

September 3, 2018 – If you are not familiar with the name Kai-Fu Lee, you are not alone. He is the founder of Sinovation Ventures, a venture capital firm, and President of the Sinovation Venture Artificial Intelligence Institute. Prior to starting Sinovation, Lee served as a Vice President at Google, and later as President of Google’s operations in China. He is a graduate of Columbia University and received his Ph. D. from Carnegie Mellon. In 2013 he was named by Time Magazine as one of the 100 most influential people in the world. Lee holds ten U.S. patents, has published seven books and has authored more than 100 journal and conference papers. He is followed by more than 50 million on his social media sites.

The posting that follows is very much a description of what the Sinovation Venture Artificial Intelligence Institute is working on including: automated machine learning, content recommendation engines, computer vision including commodity, human behaviour, and scene recognition, natural language understating including speech intention and emotion recognition, conversational systems, domain knowledge systems, and machine learning for analyzing public opinion.

There doesn’t appear to be an individual more tuned into the evolution of artificial intelligence (AI) than Lee. So when he shared his ideas on this subject recently with Peter Diamandis I felt compelled to pass it along to my many readers. Some of Lee’s prognostications may appear disturbing in just how invasive AI may become. As always I welcome your comments.

The First Wave: Internet AI

In this first stage of AI deployment, we’re dealing primarily with recommendation engines — algorithmic systems that learn from masses of user data to curate online content personalized to each one of us. Good examples include Amazon’s spot-on product recommendations, or the “Up Next” YouTube video you feel compelled to watch or the Facebook ad that seems to know what you’ll buy before you do.

Powered by the data flowing through our networks, Internet AI leverages the fact that users automatically label data as we browse. Clicking versus not clicking; lingering on a webpage longer than we do on another; hovering over a Facebook video to see what happens at the end. These cascades of labeled data build a detailed picture of our personalities, habits, demands, and desires, a perfect recipe for more tailored content to keep using a given platform.

Lee estimates that China and the U.S. stand head-to-head in deploying Internet AI, but that China’s data advantage will give its technology giants a slight edge over American within the next five years. While you’ve most definitely heard of Alibaba and Baidu, you’ve probably never stumbled upon Toutiao. Starting out as a “copycat” of America’s wildly popular Buzzfeed, Toutiao reached a valuation of $20 billion by 2017, dwarfing Buzzfeed’s valuation by a factor of 10. With almost 80 million daily active users, Toutiao doesn’t just stop at creating viral content. Equipped with natural-language processing and computer vision, its AI engines survey a vast network of different sites and contributors, rewriting headlines to optimize user engagement, and processing user’s online behavior to curate individualized news feeds for millions of consumers. As users grow more engaged with Toutiao’s content, the company’s algorithms keep getting better and better at recommending content, optimizing headlines, and delivering a truly personalized feed.

The Second Wave: Business AI

While Internet AI takes advantage of netizens constantly labeling data via clicks and other engagement metrics, business AI jumps on the data traditional companies have already labeled in the past; for example, banks issuing loans and recording repayment rates; hospitals archiving diagnoses, imaging data and health outcomes; and courts noting conviction history, recidivism and flight.

While humans make predictions based on root causes referred to as strong features, AI algorithms can process thousands of weakly correlated variables or weak features that may have an unsuspected influence on outcomes. These hidden weak correlations which escape our human linear cause-and-effect logic, business AI can leverage to train algorithms to outperform us. Applying these data-trained AI engines to banking, insurance, and legal sentencing, and the results you get include minimized default rates, optimized premiums, and plummeting recidivism.

In this AI field, the U.S. is clearly in the lead but Chinese startups may soon have a significant advantage. For example, in China, credit and debit cards that proliferated in the West in the 1970s, never happened. Because of this China has moved from a cash-based transactional economy directly to mobile payments using smartphones and other digital devices. We have seen a similar leapfrogging of technology with wireless telecommunications and mobile phones in the Developing World where African and Asian countries lacking landline infrastructure have gone wireless.

In China in 2017, mobile payment spending was 50:1 greater than in the U.S. With no competition from credit cards, 70% China’s 753 million smartphone users left behind the notion of credit and jumped directly into mobile payments. An AI-powered application for micro-finance, Smart Finance depends almost exclusively on algorithms to make millions of micro-loans. For each potential borrower, the application simply requests access to a portion of the user’s phone data. On the basis of variables as subtle as typing speed and battery percentage, Smart Finance can predict with astounding accuracy the likelihood of loan repayment.

The Third Wave: Perception AI

The third wave upgrades AI to give it eyes, ears, and other senses, merging the digital world with the physical. As sensors and smart devices proliferate through homes and cities, we are moving towards a trillion-sensor economy. Companies like China’s Xiaomi are putting out millions of IoT-connected devices, and teams of researchers have already begun prototyping smart dust, solar cell and sensor-geared particulates that store and communicate troves of data anywhere, anytime.

Perception AI “will bring the convenience and abundance of the online world into our offline reality,” states Lee. Sensor-enabled hardware devices will turn everything from hospitals to cars to schools into online-merge-offline (OMO) environments. For example, imagine when you enter a grocery store, it scans your face and pulls up a record of your most common purchases. With shopping cart in hand accompanied by a virtual assistant (VA), it plans your route through the store and the items you choose. It can even suggest a wine for an upcoming spouse’s birthday or anniversary because it knows your personal data.

While we haven’t yet leveraged the full potential of perception AI technology, China and the U.S. already are making incredible strides with China appearing to have the edge because of the proliferation of hardware to support the evolution and the country’s official attitude regarding data privacy.

Companies like Xiaomi are turning bathrooms, kitchens, and living rooms into smart OMO environments. Having invested in 220 companies and incubated 29 startups that produce its products, Xiaomi surpassed 85 million intelligent home devices in 2017, making it the world’s largest network for connected products. A KFC restaurant in China has teamed up with Alipay (Alibaba’s mobile payments platform) to pioneer a “pay-with-your-face.” Forget cash, cards and cell phones, and let OMO do the work.

The Fourth Wave: Autonomous AI

Autonomous AI integrates all three previous waves giving machines the ability to sense and respond to the world around them. While today’s machines can outperform us on repetitive tasks in structured and even unstructured environments (think Boston Dynamics’ humanoid Atlas or oncoming autonomous vehicles), machines with the power to see, hear, touch and optimize data will present a whole new ballgame.

Think swarms of drones that can selectively spray and harvest entire farms with computer vision and remarkable dexterity or heat-resistant drones that can put out forest fires a hundred times more efficiently than crews of human firefighters or Level 5 autonomous vehicles that navigate smart roads and traffic systems all on their own.

While autonomous AI will first involve robots that create direct economic value, these intelligent machines will, in time, revamp entire industries from the ground-up. Today the U.S. leads in the development of autonomous AI for self-driving vehicles. But China in Zhejiang province is building the first intelligent superhighway, outfitted with sensors, road-embedded solar panels and wireless communication between cars, roads, and drivers. Aimed at increasing transit efficiency by up to 30% while minimizing accidents and fatalities, the project may even allow autonomous electric vehicles (EVs) to continuously charge as they drive. Another Chinese initiative is happening in the Xiong’an New Area, a city near Beijing, where an infrastructure is to be built around the use of autonomous vehicles. China’s Baidu is working with Xiong’an to include sensor-geared cement, computer vision-enabled traffic lights, intersections with facial recognition, and creating parking lots that turn into parks on the fly in low traffic periods. Toronto’s Quayside, a project being developed by Sidewalk Labs, a Google subsidiary, has similar plans but on a tiny scale. Where Xiong’an is investing $580 billion U.S., Sidewalk Labs’ investment is $50 million.

China is expected to be the leading developer of autonomous drones. Premier drone maker DJI, located in Shenzhen, already owns an estimated 50% of the North American drone market and will only see its economic advantage grow.

In Lee’s Artificial Intelligence Institute, all the tools to make these four waves happen are well along in development. These questions arise.

Will China’s more authoritarian environment where managing its human population through technology perpetually give it the edge in future AI development?

Or will the evolution of this technology be dictated in nations where individual privacy will put constraints on just how much allowance to give AI?