WAIC in the eyes of 11 scientists: When scientists start to pay attention to the market

AI Technology Review reports

"There are fewer things in the white snow in the spring, and the fireworks have become heavier."

This is the most intuitive feeling for this year's attendees to participate in the World Artificial Intelligence Conference (WAIC).

From the perspective of scientists, this collision of "Yangchun white snow" and "fireworks" has a deeper meaning.

In the industrial chain of AI from theoretical research, applied technology to industrialization promotion, scientists occupy the most upstream position. But in recent years, more and more scientists have shifted from academia to industry. How to think from the perspective of the industry is the biggest change they need to make.

WaIC, as an industry conference, is the touchstone for testing this change.

At WAIC, AI Tech Review met with 11 scientists. Their research fields vary, but one thing in common is that they are all closely linked to the industry, either directly joining a certain company, or although they are still in academia, they have close cooperation with the company.

At the scene, the 11 scientists shared what they saw and thought at the WAIC site. In their view,

Research progress in the field of artificial intelligence in the fields of big data, large computing power, big models and other fields is still impressive, but in WAIC, this "technology" is transforming into "technology", and technology companies are also leaning down to explain the subtle changes of their technology to daily life in a more down-to-earth way.

The elephant is invisible, and the sound of the sound is loud. When the AI represented by big data, large computing power, and large models has developed to a certain extent, it makes it difficult for people to detect the changes in it; and the sinking of technology companies in the construction of AI infrastructure and the further combination of AI and industry are the reasons why the audience feels that there are fewer technologies and applications in this WAIC.

This is also a new stage in the development of AI: if there are several WAIC sessions, AI is more of a "top-down" driving paradigm, through cool research results, pulling ordinary users' attention to AI; and this year's WAIC is more from the application, "bottom-up" way of promotion.

This also raises new questions for scientists: they are not only concerned with the change in research methods of artificial intelligence from perception to decision-making, but also the impact of this shift in driving paradigms on their research value judgments. When scientists start paying attention to the market, they may be able to see more as a result – this is another opportunity for artificial intelligence to take another leap forward.

(The following rankings are in no particular order)

Caio Yang Qiang, WeBank: Scientists should be able to clearly explain the position of the work they do in the entire industrial chain

Compared with academic conferences, the World Artificial Intelligence Conference has many new features, for example, there are many cases of landing applications, and more emphasis on these cases, I think this is a benefit, for scientists, should be more concerned about the intersection of artificial intelligence and practical applications, at the same time, scientists should be able to clearly explain the work they do, the position in the entire industrial chain, rather than just the pursuit of accuracy, error rate and other more abstract indicators.

I think this year's World Artificial Intelligence Conference has several highlights that are particularly meaningful, the first is more international, we saw a lot of IJCAI board members from all over the world at the opening ceremony; another highlight is the social responsibility of AI, such as privacy computing, this time has been mentioned many times and there is a special track in the talk, and then, there is also the social responsibility of AI is also in the form of different panels, Both scientists and industry are increasingly aware of some deficiencies in artificial intelligence itself, and how to make up for these deficiencies has become a hot topic.

Another feature is that more young people are participating, especially this time, IJCAI and the Shanghai Artificial Intelligence Association including WAIC held an IJCAI YES youth academic conference, which was attended by hundreds of young scientists and industry leaders. At the same time, IJCAI and WAIC have cooperated in depth, not only debuting at the opening ceremony of WAIC, but also setting up a permanent office. In addition, on July 9th, an IJCAI WAIC morning run was held, with WAIC participants and leaders participating, which is a sign of WAIC's internationalization and socialization.

<h2 toutiao-origin="h3"></h2>

He Xiaodong, executive vice president of JD Artificial Intelligence Research Institute: AI can bring practical productivity progress

Looking beyond the improvement of single point technology, a new highlight of WAIC this year is that there are many industry + AI exhibitions, such as the booth of the Bank of Communications, showing that AI can bring real productivity progress. Since the impact of AlphaGo on the public in 2016, in the past 5 years, we have seen a boom in the AI academia, with many new algorithms emerging. The next 5 years will be a window period for large-scale technology landing industry, and I believe that more highlights will emerge in cross-field technology integration and industrial landing.

In fact, the general public may be more likely to pay attention to single-point AI technologies, such as speech recognition, language understanding, image recognition and generation, etc., but they do not have enough understanding of highly complex problems. Taking human-computer dialogue and interaction as an example, the average person only sees speech recognition, or further into language understanding, but in fact, the essence of dialogue is the continuous game and decision-making of both sides of the dialogue, and only taking speech recognition as the core technology of dialogue is as one-sided as image recognition as the core technology of Go AI, and it cannot really solve the problem. From this point of view, the new opportunity lies in looking at the development of AI from a higher level, focusing on more essential technical problems, and focusing on more complex large systems.

At this year's WAIC, Jingdong exhibited many AI products and services, the most worthy of introduction is the "digital bank teller" created by Jingdong intelligent customer service Yanxi, which integrates cutting-edge intelligent voice, intelligent dialogue, and virtual digital human technology, combined with financial field knowledge and industry know-how, in various customer service, marketing, investment and other scenarios to provide customers with efficient and high-experience intelligent services, can achieve independent response, active service, quality inspection and compliance, intelligent workbench, staff training, Full-process services such as customer insights.

Wang Shijin, President of iFLYTEK A.I. Education Research Institute, HKUST: Scientists should pay more attention to the introduction and understanding of artificial intelligence applications

From the theme of the conference, it can be found that from the beginning of the "new era of artificial intelligence empowerment" to the current focus on urban development, this represents the industrialization of artificial intelligence application more and more deeply, and the players involved in it tend to be rational and mature.

At the same time, the challenge of security is also growing. At the opening ceremony of WAIC, Yao Zhizhi, an academician of the Chinese Academy of Sciences, stressed at the opening ceremony that "combining artificial intelligence with 'multi-party computing' technology is expected to achieve privacy protection of data." Many of the guests at this year's conference mentioned the technology of privacy computing. It is foreseeable that private computing will usher in explosive research and application, which is of great significance for privacy protection and even national security.

In this year's WAIC, smart driving is one of the concepts that is highlighted. As an emerging product of the cross-border integration of artificial intelligence, next-generation communication technology and traditional automotive technology, the mass production and commercialization of intelligent driving technology are increasingly worth looking forward to.

At present, the state promotes the establishment of a demand-oriented mechanism for the transformation of scientific and technological achievements, and scientific and technological innovation has risen to a national strategy, in this context, scientists need to pay more attention to the strategic, basic and leading industries that support economic and social development, and strive to solve the "card neck" problem. The application of artificial intelligence has changed from the past paper talk or showmanship to practicality and serve the national strategic needs, which I think can be said to be a manifestation of the development and progress of academia.

In this diverse industry conference, scientists should focus more on the introduction and understanding of artificial intelligence applications, especially including exploring and elaborating on new applications in new scenarios, and how new problems in mature applications are solved, which can include exploration attempts and planned data verification. For example, in the "AI-enabled Education Digital Transformation Forum" of this conference, I focused on the smart classroom and personalized learning application products for the teaching scenarios according to the talents, and analyzed the transformation of artificial intelligence technology on the teaching mode and the iterative development of products.

Liu Jie, dean of the Institute of Artificial Intelligence of Harbin Institute of Technology: Artificial intelligence should be landed, and the evaluation criterion is efficiency

Every time artificial intelligence rises, people will dream that it has the ability to replace or surpass people, but "prosperity is over", people will find that artificial intelligence is far from real intelligence. This gap between ideals and reality is actually the reason for the previous "artificial intelligence winters".

This year's World Artificial Intelligence Conference is relatively pragmatic, manifested in two aspects, one is that the bubble is slowly being squeezed out, and there are fewer and fewer things to look forward to, tell stories, and draw big cakes; on the other hand, some very significant progress in artificial intelligence, including large models, super-large-scale computing power, etc., there are many news reports, so that everyone feels a little aesthetic fatigue.

Artificial intelligence will eventually land, and the evaluation criteria are the substantive benefits. Some people say that data is the new oil, and some people say that artificial intelligence is the new power, no matter what, these "electricity" must be able to land. Analogy with electric lights, telephones, elevators, only by creating the Internet of Things or physical intelligence, can artificial intelligence truly enter thousands of households and empower various industries to run towards a better future.

This year is the first time that the conference has a sub-venue in China, and we host the Harbin sub-venue, which highlights the characteristics of Harbin and the interaction with Shanghai, and hopes that everyone will pay more attention to the progress of Harbin artificial intelligence academic and industrial development.

Li Chunxiao, assistant professor of Antai College of Economics and Management, Shanghai Jiao Tong University: Academic direction is clear, and the industry recognizes the boundary

I hear a lot of voices shouting "it's time for artificial intelligence to reflect", after all, this wave of AI has lasted for so long, we really need to sort out which theories have not kept up and which theories are at the forefront. In 2021, both for academia and industry, it is a good opportunity for reflection and criticism.

The current artificial intelligence, represented by deep learning technology is only in the initial stage, in some specific areas is very "smart", but does not represent "all-you-can-eat" all fields. Therefore, we should clarify this boundary and not create the misunderstanding that "AI can fight anywhere". Isn't every previous wave of AI retreating stemming from this misunderstanding?

The current artificial intelligence is still based on perception, but in the actual scene, it needs to combine perception and cognition, so the research trend of academia should be: while expanding the research of the perception layer, actively explore the research of cognitive theory.

In fact, there are some things in the industry that are not clear, for example: 1. It is not clear which scenarios are suitable for AI to deal with; 2. There is no clear which scenarios AI cannot handle. In addition, there are some basic facts that I would like to remind you: the knowledge graph can only solve some cognitive problems; deep learning is not something that can be done.

From the government's support for artificial intelligence, it can be felt that China is a country that actively embraces technology. But this World Artificial Intelligence Conference can also see some "preferences": more emphasis on application, too little emphasis on science. Only by advancing science can we use technology to solve the pain points of the industry and help us achieve a better life.

Tang Dayan, head of the deep learning laboratory of Mingluo Technology: The AI industry can repeat the glory of the computer industry in the future

There are two new trends worth paying attention to in this WAIC, one is AI empowerment, in addition to perception, it is more necessary to add "cognition", and the combination of the two can assist or even replace human decision-making; the second is the localization of AI hardware, autonomy has a tendency to blossom everywhere, but as an asset- and intellectual-intensive industry, many small companies may only be martyrs.

It can be seen from this year's World Artificial Intelligence Conference that although intelligence is everywhere, there is still a long way to extend human intelligence-assisted decision-making; AI computing hardware manufacturers seek cooperation in many ways, but it is still difficult to break through the international giant (NVIDIA).

The AI industry is developing and advancing, and the public and even capital hope that there are landing cases, scenarios, and products that can be effectively perceived in order to continue the confidence in the AI industry. AI companies must also come up with grounded results to prove that they continue to move forward and occupy a place in this field, which is the so-called "spring and snow". Since the concept and trend of AI have been popularized to the public, it is necessary to deal with everyone's expectation that AI can change work and life, this process has been repeated once in the past ten years of computer popularization, and I hope that the AI industry can also repeat the glory of the computer industry.

Zhang Huichu, chief scientist of Boyu Education: AI development, talent first

AI development should not only focus on academic frontiers and technology landing, but also pay attention to talents. At this year's conference, the topic of talent training was mentioned in an unprecedented important position. 5 theme forum activities, covering the cultivation of young people, industrial talents, innovative talents and young scientists.

AI empowers all walks of life, what is needed is the popularization of artificial intelligence thinking and literacy improvement of the whole people, and everyone needs to recognize the value of AI technology in their own occupations and professions as soon as possible. This year, many children can be seen in the venue, and even primary school students have formed a group to visit, and "artificial intelligence from the doll" has begun to show results.

The breakthrough of AI core technology is inseparable from top talents, and the cultivation of innovative talents and scientists has already provided model innovation and policy support. In the case of a constant proportion of "geniuses", we must expand the talent base in order to allow China to maintain its leading position in the artificial intelligence competition and take the lead in entering the era of strong artificial intelligence and even super artificial intelligence.

Shan Haijun, vice president of CLP Jinxin Research Institute: The development of artificial intelligence technology has moved from technology-driven to market-driven

WAIC2021 has several distinct look and feel, summed up in a few key words: market, security, and transformation.

First, the development of artificial intelligence technology has moved from technology-driven to market-driven, and companies in the AI field have developed from focusing on cutting-edge technologies and discussing the possibility of scenario landing to launching mature solutions around specific scenarios. This exhibition has a large number of domestic AI startups, covering chips, AI platforms to AI algorithms to intelligent business, showing the fiery and pragmatic market;

Second, data security and privacy protection have become new industry hotspots. From the perspective of thematic sub-forums, the private computing academic exchange meeting and the trusted AI forum have become hot spots; major manufacturers such as Ant Group and JD.com have also launched solutions. It is believed that under the impetus of policies and technologies, it will become an important business in the field of artificial intelligence in the next few years.

Third, traditional industry manufacturers have begun to contact a large number of "AI", Schneider Electric, State Grid, SAIC, etc. have used a lot of AI technology in products and businesses, indicating that a new round of popularization of AI technology and the transformation of the industry are deepening.

CLP Jinxin also participated in the academic conference IJCAI YES co-organized by IJCAI and WAIC, and we are the diamond sponsor of this academic conference. The combination of AI academic research and industry is getting closer and closer, I think whether it is academia or industry, the future is definitely young people, so the young people's conference is to give these young people more stage to show, give them more opportunities for cooperation, can better promote the development of our AI academia, the significance of the conference lies in this.

Hao Jianye, director of Huawei's Noah's Ark Decision and Reasoning Lab: Perception AI is moving towards the ground, but the investment in decision-making AI is still not enough

In the past two or three years, the development of AI technology has been relatively flat, and from a historical point of view, all technical directions will go through a relatively gentle stage. At this stage, it is more necessary to pay attention to the improvement of the technology front and back end and the platform system, so as to make the technology better precipitate and land.

From this point of view, the overall content of this conference feels more pragmatic, focusing on the software and hardware platform behind the landing of AI technology, as well as security, ethics and other issues. If there is no relevant supporting software and hardware, including the support of the general environment, then a cutting-edge technology may just be a castle in the air.

From the perspective of AI technology evolution, large pre-training models are still relatively new things, and at present, only a few companies or scientific research institutions have enough computing power to support, and how to land large models for NLP and CV on a large scale in the future is also a problem that requires continuous exploration and thinking in the industry. Facing the future, in addition to perception and cognition, it is more important to make decisions. The mainstream technology in this regard is reinforcement learning, and there is still a certain distance from large-scale landing. For the public, the understanding of this technology is relatively small. It is necessary to further expand the understanding of non-AI people in decision intelligence through such conferences, so as to increase investment in decision intelligence.

In addition, WAIC is different from ordinary academic conferences, and the participants have a very diverse background, including experts, entrepreneurs, and government personnel in various industries. I suggest that scientists or researchers share their views on research directions, such as what impact does AI technology have on the development of society as a whole? What specific angles should the general direction of AI evolve? Include AI security, ethics, and explainability, and more, rather than showing the complex technical details of cutting-edge technology.

Professor Huang Xuanjing of the School of Computer Science and Technology of Fudan University: Yangchun white snow should not be the purpose of the AI conference

The AI sense of the conference is getting stronger and stronger, but the highlights are unattainable. Taking the best papers of the AI summit as an example, the pre-trained model has won the best papers of NAACL 2018, NAACL 2019, and NeurIPS 2020. But this is a special case, and the best papers for more summits are still incremental breakthroughs rather than subversive results.

Perhaps we can focus on some topics that have not yet produced sufficient advanced results, and perhaps this contains bright spots in the future, such as responsible AI, environmentally friendly AI, brain-computer interfaces, etc.

WAIC's mice cover the upper layers and apply them to the underlying architecture. Popular virtual idols, self-driving guided minibuses, low-altitude delivery drone networks and other applications are certainly lively, and the underlying AI platform is even more exciting. Several companies have released hyperscale pre-training models, pushing Chinese natural language processing to new heights. It is expected that with the help of domestic AI platforms, domestic artificial intelligence research, application and industrial landing can be more brilliant.

I think the AI Conference must shoulder a mission, that is, to introduce the latest academic progress and the most potential industrial applications to the media, supervision, and the public, so that artificial intelligence can enter people's hearts, into life and work, and be more grounded. Yangchun white snow should not be the purpose of the AI conference.

Although WAIC emphasizes the landing and application of AI technology, it also attracts the attention of the academic community. In addition, in addition to the Internet and artificial intelligence industry, there are many audiences from the manufacturing and financial industries. Scientists who wish to better present themselves at WAIC recommend learning techniques such as "how to make a good TED talk", not trying to introduce the entire field of research to the general public, but designing a compelling opening statement, detailing the content, and telling the story well. Another idea is to adopt flexible and diverse forms of expression, for example, in the roundtable discussion of the "Graph Neural Network and Cognitive Intelligence Frontier Technology Forum" organized by AI TIME, controversial questions were designed, allowing guests to express their opinions, colliding with the spark of ideas, and encouraging the audience to ask questions, which received good results.

Yang Hongxia, artificial intelligence scientist at Alibaba Damo Academy: It is a big surprise that pre-training technology has gone to commercial landing

Artificial intelligence has reached or surpassed human standards in the field of perceptual intelligence such as "listening, speaking, and seeing", but it is still in its infancy in the field of cognitive intelligence that requires external knowledge, logical reasoning, or domain migration. Cognitive intelligence is considered a key breakthrough towards the next generation of artificial intelligence, and hyperscale pre-training models are considered to be the infrastructure of cognitive intelligence.

Since BERT, the pre-training field of NLP has grown from small to large, and the model has grown from text to multimodal. These studies are innovative in a variety of ways, and many exciting results have emerged. Among the many studies, OpenAI's GPT3 and DIL-E are particularly striking, showing stronger language capabilities, logic capabilities, and cross-modal comprehension capabilities.

Internationally, Google, Microsoft (OpenAI) and NVIDIA have successively broken through the trillion parameter scale, the current Beijing Zhiyuan Research Institute and Alibaba DAMO Academy have developed a trillion-level parameter pre-training model, to achieve a powerful, universal multi-modal encoder and generator, and even have realized the commercial landing of some scenarios, at the same time we found that with the increase of parameter scale, the pre-training model shows a certain degree of reasoning cognition and creativity.

It can be predicted that the emergence of pre-training technology has brought great opportunities for the combination of soft and hard, new commercialization models, etc. This is what surprised me very much about WAIC and look forward to the landing of more basic areas of technology in the future.

Wang Jinqiao, a researcher at the Institute of Automation of the Chinese Academy of Sciences: Scientists should condense common needs from fragmented scenarios and design more general models

Topic innovation and cool demonstrations do not solve the problem of AI monetization, but AI is more pragmatic and no longer advertised by To VC. Models and algorithms are penetrating into the deep details of the scene, becoming more and more in line with the actual business, forming a closed loop of data-scene and SaaS services from the dot line surface, that is, know How to better understand the scene. I think it's also an improvement — getting AI into reality.

AI companies from the output of the algorithm to provide the overall solution, the scene is more and more subdivided, while the scene is also more and more involuted, the traditional enterprises began to lay out AI, thereby intensifying the competition in the subdivision industry.

The cool aura of AI is fading, and papers emphasizing new methods and cool demos often have a certain gap with the needs of actual application scenarios. When people discuss AI, they are becoming more and more grounded, and they no longer say "how many PhDs, how many Nature papers".

In addition, AI as an infrastructure, especially the emergence of pre-trained large models, the emergence of intelligent computing centers, will also change the current AI a model, an algorithm research and development model, "multi-modal + large model + multi-task" has become an important direction. With the development of large models and the explosive growth of data, artificial intelligence is becoming a new productivity tool, computing power has become productivity, data has become a new productivity data, and city-level intelligent computing centers will become an important base to support smart cities, promoting urban digitalization and intelligent upgrading of industries.

As a scientist, you should communicate with the experts of the industry scene through this conference, and introduce your scientific research direction from two aspects: on the one hand, the innovation of the combination of scene and AI, go deep into the subdivision scenario to solve the application problem of AI, systematically analyze from multiple aspects such as models, hardware and algorithms, and solve real problems through front-line practice; on the other hand, condense some common needs from fragmented scenes, combine current technological developments, and better design general models. Enhance the generalization capabilities of the model.

Due to the trial of out-of-order pushing on the WeChat public account, you may no longer be able to receive pushes of AI technology comments on time. In order to receive the first report of AI technology reviews, please set "AI technology reviews" to star account watching" .