Top 10 AI News of 2016

2016 was a year full of ups and downs. We witnessed Britain’s surprising exit from EU, the drama of the US Presidential Election, the persistent chaos of the Middle East, and China’s further economic rise as a global power. The world changed every second. These changes may not have matched our expectations, but we can see that behind them, technological innovation was among the most powerful shaping forces of 2016.
Last year, deep learning stepped into the scene and expedited the growth of artificial intelligence (abbrv. AI). AI became a gaming wizard, a Go player, a speech transcription expert, a driver, a translator, an artist, and a weapon. In the future, it is expected to have continuous influence on numerous fields, such as imaging, translation, advertising, communication, and transportation. 
Some farsighted scholars think that AI will not only change commerce, but also the way we understand the human race, intelligence, and the essence of the universe.
In the upcoming list, Synced will sort out for you the top 10 AI milestones of 2016.

NO.1 Human-Computer Chess Match: Lee Sedol vs. AlphaGo

Milestone Rate: ★★★★★ 
Attention Rate: ★★★★★

In March 2016, machines conquered the last besieged fortress of board games: Go. Google DeepMind developed a computer program named AlphaGo, which beat Lee Sedol, the international Go champion in Seoul, Korea with a score of 4–1. This attracted public attention on a great scale and raised discussions on the progress of AI. After the match, the Korea Baduk Association awarded AlphaGo with the highest Go master rank: “honorary 9 dan”. AlphaGo unseated the former human champion to become the world’s number 1 ranked Go professional on July 27th.
AlphaGo’s development team is a British AI company named DeepMind. The company was acquired by Google in 2014, and soon became the most important company in the field. Under the leadership of genius researcher Demis Hassabis, DeepMind started to expand and attract talents from all over the world. DeepMind has made a number of important research breakthroughs, including winning Atari games, speech synthesis, lip reading, subway route planning, and building neural networks. In addition to research, DeepMind has also been making forays into product design. In July, Google announced that DeepMind’s Deep Learning capabilities could potentially save its servers millions of dollars worth of electricity costs in the future. DeepMind’s ongoing collaboration with the National Health Service (NHS) is seen as an important trailblazer in the medical application of AI.
After AlphaGo’s triumph, DeepMind turned its focus towards more complicated games requiring real-time strategies. In November, DeepMind collaborated with Blizzard in opening up the game StarCraft II to AI players and machine learning researchers around the world. This time, researchers will try to challenge the game from a human player perspective and adapt to real-time gaming speed.

NO.2 Autonomous Cars are Now on the Road

Milestone Rate: ★★★★★ 
Attention Rate: ★★★★★

Many companies are trying to out-compete Google in self-driving cars. The list includes big tech companies such as Baidu, Uber, and IBM, along with traditional car manufacturers like General Motors, Honda, and Tesla Motors. Independent research projects from prominent academic universities like MIT and Harvard are looking into the social impact, ethics and law of autonomous vehicles. 
By the beginning of 2016, the first self-driving bus hit the roads of Wageningen, Netherlands. The vehicle, named WePod, was designed by French car manufacturer EasyMile in conjunction with Citymobil2, a pilot platform for automated road transport systems. WePod takes up to 6 passengers, yet can only travel 200m far at 8 miles/hr. 
nuTonomy, the world’s first driverless taxi, had a road test in Singapore in August 2016. People interested in taking a ride could use smartphones to book seats. The nuTonomy project was initiated by researchers from MIT. While companies like Google and Volvo had privately tested their autonomous vehicles, nuTonomy became the first company to conduct a road test in public. 
Google made significant progress by the end of 2017, and separated its autonomous driving project into an independent company named Waymo. 
The rapid development of autonomous vehicles also caused issues. For example, Tesla’s self-driving feature “autopilot” had caused several incidents in the past year, and one accident was lethal. Consequently, Tesla warned drivers not to rely completely on the autopilot function. 
Nevertheless, self-driving car technology as a whole improved greatly in 2016. Based on currently planning in China and the US, and provided that we manage to resolve the current safety hazards, we may be able to see these autonomous vehicles commonly on the road in less than ten years.

NO.3 Is Artificial Intelligence Dangerous? Theories of Threat

Milestone Rate: ★★★★☆
Attention Rate: ★★★★★

Tay’s inappropriate comments

In 2016, the idea that the continued development of AI is dangerous to human beings originally voiced by individuals like Stephen Hawking and Elon Musk continued to escalate. After the triumph of AlphaGo in March, these warnings started to sound more persuasive than usual to the general public. 
Later on in March, Microsoft’s AI Chatbot “Tay” also stirred up quite a scene on Twitter. Tay was originally designed to imitate the speaking habit of 19-year-old girls and learn through interacting with internet users. Within one day of its public launch, however, the internet taught Tay to say things that made the Chatbot a racist, Nazi supporter. 
Google also got into trouble this year: the company’s image recognition program identified a black man as a gorilla, and its search ads suggested job advertisements which differed between men and women. Other news, such as the Dallas police using robots to bomb snipers and the effects of social networks on the U.S. presidential election, all contributed to public anxiety regarding AI. 
Will AI be a threat? Some researchers think AI can cause further social inequality and instability, and this includes using algorithms with biased loopholes or filtered data. These social reactions could lead to major public misunderstanding of AI.
Further education is definitely needed. Many organizations and researchers are working on various projects to change the phenomena. For instance, in September 2016, Google’s DeepMind collaborated with Microsoft, Amazon, Facebook, and IBM to form an organization called “Partnership on AI”, which tries to explore the development of safe AI and promote public education. This is a good thing because there had been many publications by academics, the tech industry, and media that exaggerated public issues. If you are interested in learning more, check out articles by Microsoft on “The Partnership of the Future”, the Allen Institute for Artificial Intelligence on “Designing AI systems that Obey Our Laws and Values”, and the Nature International Weekly Journal of Science on “More Accountability for Big-Data Algorithms”.
The future relationship between AI and human beings will be refined through careful research and practice rather than verbal debates. Governments and relevant organizations around the world are stepping in to make laws on AI relating practices. In the words of U.S. Council of Economic Advisors chairman Jason Furman, “we need AI for the future.”

NO.4 Government Starts to Regulate and Encourage AI Development

Milestone Rate: ★★★★☆ 
Attention Rate: ★★★★★

The White House released two reports in October 2016, “Preparing for the Future of Artificial Intelligence” and “National Artificial Intelligence Research and Development Strategic Plan”, detailing the future challenge, development, and strategic planning of AI. These reports described the future opportunities and challenges posed by AI facing the American government. Venturebeat summarized the above reports in 7 takeaways from the White House report on AI:

AI should be used for public good
Government should embrace AI
Automated cars and unmanned aircraft need regulation
No child left behind (on learning about technology)
Use AI to supplement, not supplant, human workers
Eliminate bias from data, or don’t use it at all
Think safe, think global
U.S. President Barack Obama talks about the future of artificial intelligence, self-driving cars, and humanity, with reporters from Wired.

In December, the White House released another report named “Artificial Intelligence, Automation, and the Economy”. This report forecasted the economic influence of AI and automation technology, while offering strategies to solve issues like technological unemployment. 
Meanwhile, the European, Chinese, Japanese, and Singaporean governments also signalled over the last year that they viewed AI as an important driver of national development moving forward. Li Keqiang, the premier of China, repeatedly emphasized in public statements that the Chinese government will be putting in a concerted effort to support tech developments and the robotics industry. 
Many governments and research labs paid extra attention to AI regulations in 2016. Researchers are working to address and preempt potential negative repercussions of automation such as social inequality, technological unemployment, and capital redistribution. They are also working on safeguards against potential abuse of AI technology. These are all important issues for which we have not found complete solutions to yet. 
One report submitted by the Council of Economic Advisers to President Obama in July mentioned that an estimated 83% of jobs with wages under $20/hour will eventually be automated.The council suggested two basic strategies to deal with the situation: 1) allow flexible social experimentation; and 2) encourage alternative methods of employment. 
Governments are learning to deal with the radical new changes that will be ushered in by AI. Their careful decisions and adaptive policies will directly determine the future of our societies.

NO.5: Microsoft’s Speech Recognition Application is Now a Pro

Milestone Rate: ★★★★☆ 
Attention Rate: ★★★★☆

In October, researchers from Microsoft published an article named “Achieving Human Parity in Conversational Speech Recognition”. Apparently, speech recognition systems now have an word-error rate (WER) of 5.9% , outperforming human professionals. The 6.9% WER historical record was broken in December. The development speed is definitely threatening to current human voice recording employees. 
Other industry players are also entering the field. Yonghao Luo, the CEO of Smartisan, demonstrated the company’s new iFlytek voice input method at a product launch also in December. 
The continued development of speech input methods will popularize speech assistants. At AWS re:invent 2016, Amazon launched various kinds of tools to help developers use the company’s speech assistant Alexa more easily. Rohit Prasad, the vice president of Alexa’s development team, disclosed that the AI application has more than 5,000 skills up to now. The Amazon Echo, which had Alexa installed, could easily help users with daily tasks like scheduling and music searching via speech commands. Also, big companies like Google, Microsoft, Samsung, and Apple are catching up. In a few years, machines will be around to wait for your speech commands.

NO.6 The Breakthroughs of Machine Translation with Neural Networks

Milestone Rate: ★ ★ ★ ★ ☆ 
Attention Rate: ★ ★ ★ ★ ☆

The Model Architecture of GNMT

Since the fall of the “Tower of Babel”, language unification has been an universal dream. Now, AI based on neural networks is providing us with the solution. 
In September 2016, Google published a paper on arXiv introducing the Neural Machine Translation System (GNMT), an advanced training technique which maximizes the quality of machine translation. The system is currently being applied to Google’s Chinese-English machine translation. Two months later, Google announced further breakthroughs in the realization of multi-lingual translation by neural machines and zero-shot translation. 
Sougou’s new real-time machine translation application was unveiled at the Wuzhen World Internet Conference in November 2016. Machine translation is becoming more and more of an established trend. It is now plausible that the goal of language unification will be achieved in our lifetimes.
In addition to the translation of text-to-text sequences between language pairs, 2016 also saw a noteworthy breakthrough in the “translation” of lip to text (video sequences to text sequences). In November, at Oxford University, Google’s DeepMind and the Canadian Institute of Advanced Studies (CIFAR) jointly published an important paper describing the use of LipNet (a sentence-level lip reading AI) to actualize automatic lip-reading. The application currently has 93.4% accuracy, easily beating experienced human lip readers.

NO.7 The Start of AI Hardware Warfare: Giants vs. Start-Ups

Milestone Rate: ★ ★ ★ ☆☆ 
Attention Rate: ★ ★ ★ ★ ☆

As deep learning algorithms become more complex, the data sets used grow as well, making demands on upscaled hardware. 
In 2016, building platforms for artificial intelligence became a major new direction in the development of computational hardware. In addition to chip giants like Intel and NVIDIA making high-profile moves in AI, start-ups with core technologies are also trying to make pivotal changes (although high potential start-ups are acquired at faster rates). Even companies like Google are trying to make their own moves in this area. 
The traditional chip manufacturer NVIDIA combined use of GPUs with deep learning algorithms to further its own development, helping its stock prices to soar and emerge notably as this year’s biggest winner in AI computing. 
Intel, with a bigger market share, naturally did not wait for new markets to be claimed by new comers and sought to catch up through acquiring start-ups. In 2016, Intel acquired several AI start-ups such as Movidius (computer vision) and Nirvana (in-depth learning chip). By November, Intel announced its roadmap for AI with Nirvana and FPGA vendor Altera Plus, which was acquired in 2015, and introduced its corporate strategy and planned production ecosystem in AI chips. 
AMD emerged as a noteworthy competitor in this field in 2016, as the company announced its first VEGA GPU architecture-based machine learning chip. At the same time, DSP vendor CEVA, FPGA vendor Xilinx and processor technology vendors like Imagination also made their presences felt in the machine learning world.

Gregory Wong, CEO of NVIDIA, giving a speech at GTC Europe 2016

At the same time, Internet giants also found new opportunities in the same field. In May 2016, Google released a new custom-designed chip tensor processing unit (TPU/Tensor Processing unit). This chip is specifically tailor-made for Google’s open source machine learning framework TensorFlow. Microsoft has also indicated support for FPGA through Project Catapult. In addition, IBM’s progress in neurological morphology has attracted a great deal of attention, and may even herald a new direction for the development of AI moving forward. 
Many start-up companies, such as Wave Computing and Kneron, performed very well in 2016. The Chinese AI sector also saw tremendous growth over the past year. Chinese start-ups like Cambrian and Shenjian Technology successfully developed deep learning chip platforms. The Institute of Computer Science of the Chinese Academy of Sciences notably launched Cambrian 1 A Processor, which is allegedly the world’s first commercial processor dedicated to deep learning.

NO.8 Stanford Releases One Hundred Year Study on Artificial Intelligence

Milstone Rate: ★ ★ ★ ★☆ 
Attention Rate: ★ ★ ★ ☆ ☆

In Fall 2014, Stanford University launched a research project on AI spanning the past century. 
Two years later, in early September 2016, the research efforts of giants like Google and Microsoft, well-known universities like Harvard University and MIT, as well as the famous Allen institute of Artificial Intelligence organized by Stanford University jointly published the research results of many experts and scholars in a report called “Artificial Intelligence and Life in 2030: One Hundred Year Study on Artificial Intelligence”.

This report encompassed the present development of Artificial Intelligence and its projected future impact on areas such as employment, environment, transportation, public safety, healthcare, community involvement, and politics. The report attracted a great deal of attention, and to a certain extent has guided the subsequent development of AI around the world.
In 2016, many other research institutions, science and technology media also released their own forecast reports. Synced published its own series of monthly reports titled “AI100: the world’s most noteworthy 100 AI companies”, showcasing one hundred NLP, computer vision, chips and hardware companies operating currently in the field of AI. This report will continue to be updated on a monthly basis. Click here to find out more.

NO.9 The Artificial Intelligence/Robot in Our Lives

Milestone Rate: ★ ★ ★ ☆☆ 
Attention Rate: ★ ★ ★ ☆ ☆

Although today’s AI appear far less advanced compared to the sensible, self-aware robots depicted in science fiction, they have already infiltrated our lives in many ways.
The past year saw tremendous advancements in voice assistance technology, such as Amazon’s Alexa, the new Google Home, and Microsoft’s Xiaona and Xiaobing (“twin sisters” directed at helping people with real life scenarios in the Chinese market).
We also made tangible progress on the development of intelligent hardware (namely, robots). Autonomous cars began carrying passengers, unmanned aerial vehicles carried freights, and humanoid robots in Japan took up receptionist roles inside enterprises while nursing robots communicated with seniors and autistic children in Europe. In the United States, robots even became an integral element of law enforcement for the first time. A robot carrying a detonating bomb was used to end a standoff between police and a sniper in Dallas. 
AI was also used as a key research aid in many scientific fields. In July 2016, NASA sent a mission update to the Curiosity Rover on Mars, granting it the autonomy to select research targets. AI also helped scientists to look for exoplanets and signs of extraterrestrial life. AI was widely employed to analyze data in fields such as economics, atmospheric systems, and genomics. In early December 2016, Science magazine published its first-ever issue on the use of robots in science, serving as direct validation of the growing importance of AI in academia.
AI also became an artist in 2016. Applications like Prisma became omnipresent on the internet. Intelligent systems are now able to compose classical music, edit porn, and even write movie scripts. The following video is a sci-fi short film based a script written by the AI program Jetson.

NO.10 Industry and Academia Crossing Over

Milestone Rate: ★ ★ ☆ ☆☆
Attention Rate: ★ ★ ★ ☆ ☆

In 2016, the value of deep learning began to manifest after many years of research. Industry players focused on AI like never before. 
In the past year, we witnessed researchers and scientists entering the commercial space, for instance:

  1. In August — Shiguang Shan, a computer vision expert from the Chinese Academy of Sciences founded the facial recognition company Seeta.
  2. In October — Ruslan Salakhutdinov, a machine learning expert from Carnegie Mellon joined Apple, while Canadian professor Joshua Bengio from University of Montréal joined Element AI, a deep learning incubator.
  3. In November — Stanford professor Feifei Li joined Google; and Carnegie Mellon professor Xing Bo founded the machine learning platform Petuum.

While academics flocked into industry, industry players also began producing high-quality research papers. Companies like Google, Microsoft, Facebook, Tencent, and Baidu built their own research institutions dedicated to AI. These institutions helped companies to enhance their products and also published tons of research results. In December, the aforementioned Ruslan Salakhutdinov announced at NIPS 2016 that Apple will continue to publish AI results to make the field more open. The company subsequently published its first paper on AI, “Learning from Simulated and Unsupervised Images through Adversarial Training”. 
The message is clear: in order to lead in the age of AI, we need to participate.

Many, many more things happened in the world of AI in 2016, such as realist speech synthesis (WaveNet), the steady development of image recognition systems, video prediction applications, confrontation networks, and even quantum computing, optical computing, bio-computing, BCI, robots, and new cryptography applications. 
In the future, can people live without AI? We are excited for what’s to come!

Original Article from Synced China |Author: Pan Wu, Zenan Li | Localized by Synced Global Team: Meghan Han, Jiaxin Su, Rita Chen