Looking at the Global AI Development in 2019 Through 5 Hot Topics
-
Looking back at 2019, the development of AI technology is changing the direction of global technology.
In 2019, AI, as one of the core technologies of the "Fourth Industrial Revolution," remained a global tech hotspot and seemed to gain even more momentum. For instance, at the recently concluded NeurIPS 2019, a record-breaking 13,000 attendees and 1,428 published papers showcased the thriving ecosystem and abundance of talent in the AI field.
This year, AI development has penetrated deeper into various industries and become more closely integrated with people's daily lives. As you can see, many trending topics on social media are closely tied to AI.
Today, let’s explore the global AI development highlights of the year through five trending topics!
On October 23, as Nature celebrated its 150th anniversary, Google announced the achievement of "quantum supremacy."
Google used its 54-qubit quantum processor, Sycamore, to complete a computational task—random sampling of quantum circuits—in just 200 seconds, a task that would take the world's most powerful supercomputer, Summit, 10,000 years to finish.
The announcement instantly became the biggest tech news globally.
AI industrial development relies on three key elements: computing power, algorithms, and data. "Quantum computing" will unlock a new frontier in human computational capabilities, opening up entirely new possibilities for AI.
This is why Google began investing in quantum computing 13 years ago and later established the Google AI Quantum team to explore its potential.
Although, as Pichai put it, the achievement of quantum supremacy is still at a "Hello World" level of computational tasks, it is enough to make 2019 a significant year in the history of AI computing power.
If "quantum computing" still seems far from practical use, 2019 also saw substantial progress in more reliable computing power improvements. The industry's advancements mainly focused on two areas: chip development and new cloud computing centers.
In terms of chips, both academia and industry are exploring more possibilities.
For example, more general-purpose chips.
On August 1, Nature introduced the "Tianji" AI chip architecture developed by a Chinese team. Led by Tsinghua University and involving multiple universities from China and abroad, this architecture combines brain-inspired computing with AI algorithms and successfully powered autonomous vehicles. This achievement demonstrates a potential path toward achieving the ultimate "general-purpose" AI.
And, larger chips.
Also in August, Cerebras Systems unveiled the Wafer-Scale Engine (WSE) in Silicon Valley—the largest AI chip in history, 56 times larger than conventional chips, with 1.2 trillion transistors. Shortly after, they launched the CS-1, a new deep learning computer equipped with the WSE. Such supercomputers provide unprecedented computing power for large-scale AI problems, significantly reducing training times.
And, more specialized chips.
In November, following NVIDIA and Qualcomm's earlier releases of AI-specific chips (TPUs), Intel finally introduced the Nervana NNP processor for complex deep learning computations. Meanwhile, Baidu developed China's first full-featured cloud AI chip, "Kunlun," and launched the Kunlun Cloud Server in December, supporting mainstream AI frameworks/platforms, including PaddlePaddle.
For edge devices, the focus is even more specialized. In July, Baidu adopted a "software-defined chip" design approach and released the Honghu chip, tailored for far-field voice interaction, primarily for scenarios like in-car voice assistants and smart homes. It’s safe to say that specialized AI chips have become the standard, whether in edge devices or servers.
In cloud computing centers, the theme is using AI to optimize AI-specific cloud computing infrastructure.
With the growing demand for AI, the need for computing power continues to rise, and existing data centers can no longer meet these requirements. According to IDC, by the end of 2020, 55% of the world's enterprise data centers will be forced to upgrade or optimize due to these demands.
Cooling and energy efficiency are critical aspects. Gartner predicts that by the end of 2020, as AI computing density increases, 30% of data centers will no longer be economically viable.
Earlier this year, China's Ministry of Industry and Information Technology, along with other departments, issued the "Guidelines on Strengthening the Construction of Green Data Centers," explicitly stating that by 2022, China's data center energy efficiency must reach world-leading levels, with large and hyperscale data centers achieving a Power Usage Effectiveness (PUE) of no higher than 1.4.
Following DeepMind's success in 2018 in using AI to cool data centers with remarkable results, China has also made encouraging progress. Huawei has continued to release innovative technologies like iPower, iCooling, and iManager, using AI to optimize power distribution systems and end-to-end energy efficiency management, improving resource utilization by 20%.
Baidu's under-construction cloud computing center features a super-large neural network, a massive high-performance computing cluster, and adopts Baidu's fourth-generation data center infrastructure module architecture. With an AI control system developed on PaddlePaddle, it achieves an annual average PUE of 1.15, becoming the first fully distributed, prefabricated data center in terms of power supply and cooling.
In 2019, global AI computing power continued to advance, with chips becoming more powerful and versatile, while data centers became more energy-efficient and cost-effective.
In the future, computing power will remain a critical factor for breakthroughs in the industry, and quantum computing will experience a new wave of growth. Edge AI chips will become more cost-effective, specialized, and integrated into solutions, while NPUs will become a standard module in the next generation of general-purpose edge CPUs.
On September 25, Beijing Daxing International Airport officially opened, becoming the world's largest and most "high-tech" airport.
Among its features, the deployment of automated parking robots caught public attention and became a trending topic. Developed by JD Digits and Shougang S-PARK, these robots allow drivers to simply park their cars in a designated area and walk away, leaving the robots to handle the rest, including finding a parking spot.
These are essentially Automated Guided Vehicles (AGVs), which have long been used in logistics and warehousing, representing a major category of industrial robots.
Traditional industrial robots rely solely on electromechanical systems and conventional control theories, making them suitable for repetitive tasks on production lines. However, with the global push for Industry 4.0, there is an increasing demand for highly intelligent and flexible applications.
Powered by industrial big data and sensors, AI-enabled robots can now "see" and "understand," becoming more intelligent. They can handle more complex scenarios and even collaborate with humans.
For example, at the Baidu AI Developer Conference in July, Baidu showcased a robotic arm called "Tea Master," based on a service robot solution. Using Baidu's 3D vision technology, the robot can detect and track the position of teacups. Through motion planning and control, the arm can perform collision detection, avoid obstacles, and generate tea-pouring trajectories autonomously. Combined with voice and semantic interaction, the robot can "understand," "speak," and "act" like a human.
Meanwhile, Boston Dynamics, known for its viral robot stunt videos, began industrial applications this year. In April, after acquiring Kinema Systems, the company commercially launched the logistics robot PICK, which can handle mixed-SKU loading and unloading using recognition technology, almost like a human worker. In July, it announced the commercial release of the quadruped security robot SPOT.
It’s safe to say that in 2019, AI achieved more substantial industrial applications, integrating more deeply with production practices across various sectors.
Beyond manufacturing, where industrial robots enhance efficiency, AI has also made strides in agriculture. For instance, China Agricultural University and JD Digits introduced the "Shennong Brain" system for pest control and smart farming. Baidu and BOE’s plant factory developed an intelligent soilless cultivation solution that saves energy and water, reduces costs, and increases yields.
These AI applications digitize and productize the expertise of agricultural specialists, enabling scalable production.
In the future, AI will penetrate even deeper, with industries widely adopting deep learning technologies to innovate, accelerate transformation, and achieve intelligent upgrades.
On July 1, as Shanghai implemented its "strictest-ever" waste sorting regulations, garbage classification became a regular topic in China. "What kind of garbage is this?" even became a catchphrase.
A month later, the news that "Shanghai will deploy 2,000 AI trash cans" went viral. These bins can automatically identify and sort waste into the correct category each time a person disposes of an item.
Relevant applications include Baidu App's 'AI Waste Sorting' mini-program, which leverages Baidu's AI visual and voice capabilities to help users identify waste types through voice or image searches.
In reality, the technologies behind these AI applications are quite mature. Developers only need to call basic AI algorithms and map them accordingly. This can often be achieved directly using various AI frameworks and their preset models, sometimes without the need for retraining data.
A small AI trash bin application reflects the broad societal demand for basic AI capabilities. The rapid deployment of hundreds of similar applications in 2019 highlights the increasing platformization and industrialization of AI frameworks, as well as their improved usability.
For example, the progress of PyTorch and TensorFlow in 2019 shows both frameworks addressing their weaknesses to expand their influence in the AI arena. PyTorch, known for its simplicity and Python-like coding, is more popular in academia. However, TensorFlow, with its superior deployment capabilities, has been more widely used in industrial applications due to its support for edge device deployment.
At this year's f8 conference, PyTorch introduced an experimental PyTorch Mobile. Version 1.3 supports a complete end-to-end workflow from Python development to deployment on iOS and Android. Meanwhile, TensorFlow significantly enhanced the developer experience at its October World conference, integrating more preset models, including those from DeepMind, and emphasizing 'Deploy ML anywhere' through TensorFlow Extended, Lite, and .JS.
Domestically, Baidu's PaddlePaddle, the only open-source, fully-featured deep learning platform in China, now boasts over 1.5 million developers. This year's upgrades focus on standardization, automation, and modularization to meet the growing and increasingly complex demands of developers and industries. Mid-year, PaddlePaddle partnered with Huawei's Kirin chips to integrate deep learning frameworks with AI computing power, strengthening China's AI competitiveness.
In 2019, major AI frameworks have evolved to become more comprehensive and user-friendly. For simple requirements, technology may no longer be a barrier—just plug and play. AutoML may soon become a reality, where frameworks can automatically select data, optimize models, and perform training and deployment based on input meta-knowledge.
As AI frameworks become more powerful and easier to use, the world may see the rise of multiple 'AI factories.' These entities will aggregate and abstract demands, then propose scalable solutions. For example, voice customer service could be replicated on a large scale for one-on-one education and financial services.
In the future, identifying needs and knowing how to implement them correctly may become the real challenge and the key talent demand in the market.
On November 1, China's three major telecom operators officially launched 5G plans, marking the formal commercialization of fifth-generation mobile communication technology in China, which naturally becomes the world's largest 5G market.
5G's new capabilities, as infrastructure, will synergize with AI to unlock more opportunities. The 'high concurrency' application scenarios of 5G will enable more devices to connect, truly realizing the Internet of Things (IoT).
According to IDC, global IoT spending is projected to reach $726 billion this year and surpass $1 trillion next year. This year, from wearable devices to smart home products, more smart hardware devices have emerged, with sales growth hitting new highs.
Xiaomi, with its range of smart hardware from sockets to watches, reported over 210 million connected devices on its IoT platform in Q3, a 62% year-on-year increase. In smart speakers, China's top-selling Xiaodu speaker achieved a staggering 3700% growth in Q2.
On the developer side, platform capabilities continue to improve, and manufacturers are accelerating their growth. In February, the Alexa platform released its annual report, announcing compatibility with over 100,000 devices, reaching 100 million devices, and achieving a 100% increase in user interactions. It now supports 15 languages. DuerOS launched version 5.0, introducing industry-leading full-duplex wake-free capabilities. According to Tianyancha data, the number of registered smart hardware companies in China increased from 1,459 in 2018 to 2,027, a nearly 40% rise.
5G's 'low latency' application scenarios are enabling the deployment of high-reliability demands like autonomous vehicles and manufacturing.
As major automakers in Japan, the U.S., and Europe announced massive layoffs this year, the commercial progress of autonomous vehicles in the U.S. has slowed. High costs and financial constraints have led GM to delay its self-driving taxi rollout. Additionally, developing fully autonomous vehicles for all road conditions is extremely challenging. Waymo CEO John Krafcik stated in January that autonomous driving might never achieve full road condition adaptability.
In August, The Information revealed that Waymo's self-driving taxis in Silicon Valley and Phoenix had a 70% user satisfaction rate. By strict taxi safety standards, this implies that users perceived danger in 30% of cases, such as following too closely, inaccurate parking, or stopping abruptly on the road.
China has made encouraging progress. Baidu's Apollo leads in road testing, with L4 autonomous driving city road tests covering 3 million kilometers. With the launch of Apollo Enterprise earlier this year, Apollo has entered its commercialization phase. Xiaodu Vehicle OS has partnered with over 300 leading ecosystem partners, and the jointly developed 'Hongqi EV' Robotaxi fleet began trial operations in Changsha.
With 5G, smart city infrastructure will provide more possibilities for autonomous vehicles in China. Narrowly defined, smart cities involve intelligent upgrades like street sensors, smart lampposts, waste bins, and even smart shading facilities. Leveraging 5G's low latency, these sensors can transmit real-time road data to vehicles, significantly improving the road environment for autonomous vehicles and easing their deployment.
On October 31, Google's Sidewalk Labs smart city project in Toronto received 'conditional approval' after facing public and governmental challenges. A final vote is scheduled for March next year. Once completed, it will be a flagship smart city project in the Western world.
China is progressing even faster. 'Building a strong cyber nation and smart society' has become a national strategy, with smart cities playing a key role. In November, CCID, JD Cloud, and JD City jointly released the '2019 China Smart City Development Strategy and Tactics Research,' proposing five strategic pathways: industrial development, good governance, public welfare, symbiosis, and foundational construction.
Domestic companies have also accelerated their smart city initiatives in 2019. Early in the year, JD's iCity conference in Beijing announced JD City as a top-tier strategic focus. By June, the Smart Suqian App was launched as the first milestone. At November's JDD, the Smart City System 2.0 was officially deployed in Xiong'an. Baidu partnered with Neusoft to launch the 'Cloud Intelligence Future City' smart city solution. Alibaba Cloud collaborated with Sena in May to improve Malaysia's traffic system and with AutoNavi in August to release a joint smart traffic management solution.
In the future, as more high-reliability devices connect and 5G and edge computing advance, computing power will become more distributed. Smart cities will develop faster, with intelligent transportation accelerating in scenarios like campuses, logistics, and public transit.
In 2019, what you see may not be real. China's ZAO App, which allows users to star in famous movie scenes using deepfake technology, gained massive popularity upon release. While celebrating technological progress and novel applications, public concern about deepfakes peaked in 2019.
This technology can also be misused—for example, fabricating speeches by politicians like Obama, which has already impacted Malaysian politics and led to fraud cases in the UK. Discussions on AI regulations and ethics have become a priority for governments worldwide.
On November 29, China's Cyberspace Administration issued new regulations prohibiting the unchecked dissemination of AI-generated fake videos starting next year, sparking widespread discussion. The rules mandate platforms to deploy "non-authentic audiovisual identification technology" promptly.
A month earlier, California Governor Gavin Newsom signed AB-730 into law, criminalizing the use of Deepfake for political manipulation. Virginia had similarly banned Deepfake abuse in July. Tech companies joined the effort too - Twitter released its first anti-Deepfake policy draft in November, while Microsoft and Google accelerated authentication technology research.
Beyond synthetic videos, AI text generation made strides. OpenAI's GPT-2 model, initially withheld due to concerns over its hyper-realistic article generation capability, was fully released in November after proving manageable. Other breakthroughs included Baidu's ERNIE framework topping the GLUE benchmark and Carnegie Mellon's NLP advancements.
Simulated training data gained traction where real-world data proved costly. OpenAI Five defeated Dota 2 champions using simulated training, while Amazon's Alexa and autonomous vehicles like Aurora leveraged synthetic environments. Baidu pioneered an Augmented Autonomous Driving Simulation (AADS) system featured in Science Robotics.
AI demonstrated profound social impact: Amsterdam researcher Huang Zhisheng's Weibo suicide prevention team intervened in 507 cases, Google allocated $25M for AI-powered nonprofits, while Baidu's AI-assisted missing person searches reunited 9,000 families and launched the world's first sign-language translation app for hearing-impaired children.
As AI capabilities grow exponentially, 2019 marked pivotal developments in computational power (quantum computing, 5G), platformization (TensorFlow, Baidu Brain 5.0), and policy frameworks. China's smart city initiatives and industrial upgrades showcased its AI leadership, paralleling global efforts to balance innovation with ethics. The stage is set for AI-driven industrial revolution, defined by quantum supremacy and responsible deployment.