Directly hit the opening of the WAIC丨World Artificial Intelligence Conference: generative AI ushered in the outbreak, robots will be more "at their disposal"

21st Century Business Herald reporter Tao Li, Zhang Zitong, intern Yan Shuo reported from Shanghai

In Shanghai in July, the sun is on fire. Even hotter than the temperature is the wave of innovation in emerging technologies such as large models and robots at the 2023 World Artificial Intelligence Conference.

On July 6, the 2023 World Conference on Artificial Intelligence (WAIC) opened in Shanghai. The theme of this year's conference is "Intelligent Connected World, Generating the Future", focusing on the frontier of science and industrial development, focusing on ten cutting-edge trends such as large models, smart chips, robots, and metaverse. The 21st Century Business Herald reporter noticed at the opening ceremony that both the exhibitors and the number of visitors significantly exceeded previous years.

Among the many cutting-edge technologies, large models and generative AI have received unprecedented attention. Both the government and academia have expected the artificial intelligence industry and put forward rational suggestions.

Tesla CEO Elon Musk said in his speech that the emergence of generative AI is having a very profound effect and impact on human civilization. With the explosive growth of digital computing power, the proportion of machine computing power and biological computing power is expanding, and the proportion of machine intelligence will exceed biological intelligence in the future. Over time, the proportion of AI in all intelligence will become lower and lower, which will be a fundamental and profound change compared to machine intelligence.

Yao Zhizhi, academician of the Chinese Academy of Sciences and dean of the Institute of Interdisciplinary Information Science of Tsinghua University, said that after ChatGPT, the next important goal of artificial intelligence research is to allow robots with multiple perception capabilities to learn new skills independently in different environments.

Talking about the development of large models in the vertical field, Yao Zhizhi believes that the easiest scenario to think of is paperwork, based on the language of large models, more work can be done by machines.

He also pointed out that the current reinforcement learning methods in the industry are too slow, often taking months to learn new technologies, and if there is a reinforcement learning method that allows robots to complete self-learning in a few hours, it will be a major breakthrough.

The "big model" is integrated into the industry

The scale of participation in this year's World Artificial Intelligence Conference has increased significantly compared with previous years. The exhibition area reaches 50,000 square meters, and the total number of enterprises participating in the exhibition has reached a record high, and it is expected to exceed 400.

If it comes to the most "eye-catching" existence of the World Artificial Intelligence Conference, the big model is not allowed. At the WAIC booth, Ant Group systematically demonstrated the technical architecture of Ant Trusted AI for the first time, as well as the newly upgraded trusted AI industrial detection platform - Ant Jian 2.0.

The platform can generate massive test sets through intelligent confrontation technology to induce AIGC generative models interactively. It can be understood that it is like a 24-hour sleepless "security hacker" looking for a large model, so as to find the weaknesses and security problems of the large model.

SenseTime brought the "SenseAvatar" AI digital human video generation platform this time. The relevant person in charge of the company told the 21st Century Business Herald reporter that the Ruying digital human generation solution only needs one live-action material input to generate hyper-realistic digital twins and achieve exponential improvement in video production efficiency.

At present, the main scenario of this platform is to help enterprises and individuals easily carry out content creation such as short videos and live broadcasts for digital humans, and also provide video solutions for education and training, corporate publicity, entertainment and culture.

In fact, AI has become an important productivity tool to promote digital development, and how to make AI applications comply with safety specifications has become a key problem in the industry in recent years. With the emergence of the "strongest brain" represented by large models, the evolution route of artificial intelligence is becoming clearer.

"Of the more than 170 large models in the world, more than 100 are in China." Sheng Xuefeng, president of the Shanghai Smart City Development Research Institute, told the 21st Century Business Herald reporter frankly that if AGI or ChatGPT is a huge transformative event in the industry, it is likely to open a new era. "This era can be compared to the mobile Internet."

ChatGPT-led conversational AI makes it easy for ordinary people to perceive and understand AI, and the combination and interaction of people and data is smoother than ever. However, the development and landing of large models also require time to explore. Any technology, in the final analysis, needs to serve the industry and scenarios.

For example, the landing of large models of technology companies such as iFLYTEK and NetEase Youdao took the lead in targeting the scene in the field of education. The latest application results of NetEase Youdao's self-developed large model "Ziyue" also made its debut. At the conference site, the 21st Century Business Herald reporter noticed that the virtual population language coach can interact with the user, which can not only provide emotional feedback for the practitioner's answer, but also actively throw questions.

However, the large-scale application of artificial intelligence in the industry must also solve the problems of explosive data growth, trusted data flow, user privacy and data security, complex correlation of massive data, and reduction of computing energy consumption.

Robots "at their disposal"

Robots are known as the "jewel at the top of the manufacturing crown", and its research and development, manufacturing and application are important symbols to measure a country's scientific and technological innovation and high-end manufacturing level. At this World Artificial Intelligence Conference, robot exhibits are dazzling.

It is currently widely believed in the industry that the next wave of artificial intelligence is Embodied AI. As a carrier for artificial intelligence to enter the physical world interaction, embodied intelligence combines artificial intelligence technology with robotic entities, so that the "brain" has a "body" that can be controlled, perceived, interactive and actionable, and as an important partner and assistant of human beings, it brings great convenience and benefits.

Tesla CEO Elon Musk pointed out at the opening ceremony of the World Artificial Intelligence Conference that based on the existing development trend, the number of robots will increase in the future. In terms of specific proportions, Musk believes that the ratio of robots to humans will be more than one to one, that is, the number of robots on the planet will exceed the number of humans.

In terms of the role that humanoid robots can play, Musk said that Optimus humanoid robots are aimed at doing things that humans don't want to do, such as self-driving, factory handling, etc. "The Optimus humanoid robot can do some work for people, it is not very intelligent, but it has enough intelligence to do some boring, repetitive and dangerous work that humans don't want to do, and that's what we're aiming for."

Musk calculated that under normal circumstances, a family car is about 10 to 20 hours a week of use, most of the time in the parking lot lying on its stomach, but for fully autonomous cars, it may be used 50 to 60 hours a week, a total of 168 hours a week, so the use rate of fully autonomous vehicles is nearly 5 times higher than that of non-fully autonomous vehicles.

Not only at the opening ceremony, but also in the booth located in the Shanghai World Expo Exhibition and Convention Center, the 21st Century Business Herald reporter also found many "humanoid robot" products.

"Just like the development trend from desktop computers to personal computers to mobile phones, general-purpose humanoid robots will gradually become the mainstream of the industry in the future." The person in charge of Fourier Intelligence told the 21st Century Business Herald reporter at the booth that the current hardware foundation of the general humanoid robot is basically laid, and the wave of applications promoted by large models is coming. "We feel that the commercialization of general-purpose humanoid robots can be expected in the future, so they also seize the opportunity and gradually cut the development strategy into the field of trillion-level general-purpose robots."

IT IS UNDERSTOOD THAT THE HUMANOID ROBOT GR-1 BROUGHT BY FOURIER INTELLIGENCE IS 1.65 METERS TALL, WEIGHS 55 KG, HAS 40 DEGREES OF FREEDOM THROUGHOUT THE BODY, THE MAXIMUM PEAK TORQUE OF THE JOINT MODULE CAN REACH 300NM, THE WALKING SPEED CAN REACH 5KM/H, AND THE LOAD IS 50 KILOGRAMS.

This robot has a highly bionic trunk configuration, anthropomorphic motion control, fast walking, agile obstacle avoidance, steady uphill and downhill, anti-impact interference and other motor functions, combined with cognitive intelligence, can complete actions collaboratively with people, and has great potential in industry, rehabilitation, home, scientific research and other application scenarios.

Another brand, Cloudminds humanoid robot Ginger Ginger 2.0, the next generation of Cloud Ginger, also appeared at WAIC as a "welcome sister" at the booth. Cloud Ginger 2.0 is an "encyclopedia" when it comes to interacting with audiences, allowing for multiple rounds of dialogue and multimodal interaction.

It is understood that this latest humanoid robot uses flexible joint-SCA2.0 and a variety of sensors to achieve high-precision intelligent grasping, and has visual and laser positioning and navigation capabilities. Based on precise positioning and autonomous walking, it can provide welcome reception, business guide, brewing coffee, item delivery, program performance, accompanying passport viewing, education and scientific research and other services.

In the eyes of the industry, with the rapid evolution of AI, robots have become more "at their disposal", and all aspects of future landing life scenes also have more possibilities.

AI ethics and mechanisms

On the other hand, the popularity of generative AI has also worried the industry. The in-depth application of artificial intelligence not only puts forward higher requirements for data, algorithms, and computing power, but also poses more challenges to security, privacy, and ethics. Under the premise of ensuring data security and privacy protection, and improving the ethics and security of artificial intelligence, AI technology can truly release the application value.

It took a long time for humans to establish the ethics and governance mechanisms of the present. If AI is allowed to develop consciousness and intelligence like humans, how can humans meet the requirements of ethics and governance? What is the mechanism of human intervention in artificial intelligence?

At the opening ceremony of the World Artificial Intelligence Conference, Yann LeCun, winner of the 2018 Turing Award and chief AI scientist of the Meta AI Basic Artificial Intelligence Research Institute, said that he completely disagreed that "because anyone can do anything with AI, the situation can be very dangerous, so AI needs to be strictly regulated." ".

In his judgment, the only way to make AI platforms safe, good, and practical in the long run is open source.

Yang Likun talked about the ways in which AI can become intelligent and controllable. "If you think that the way to reach human-level AI is to do larger autoregressive LLM [large language models] and then test them with multi-model data, then you might think that these AI systems are not safe." But in fact I don't think this system can be very intelligent, I think the way they become intelligent is also the way to make them controllable, which is the idea of goal-driven artificial intelligence. Essentially, this is the goal they must meet. ”

Yang Likun believes that artificial intelligence systems will not deceive or dominate humans through bit by bit infiltration. Humans can set goals that force AI to be honest, such as forcing AI to submit to humans, being careful of targets that want them to be curious, and being careful about giving them access to resources they shouldn't have. "So I think these systems are going to be completely controllable and manipulable." Systems can be designed to be secure, but it's not easy. Designing these goals to make the system secure will be a formidable engineering challenge. ”

But Yang says humans don't have to design it well from the start, but can start by building a system as smart as a mouse, with the goal of making it a good mouse. "There are some people who want AI to be strictly regulated because they are afraid of the consequences of AI."

On the contrary, Yang Likun believes that in the future, everyone will interact with the digital world through artificial intelligence assistants, and all human information will pass through this artificial intelligence assistant, which is not a good thing if the technology is controlled by a few companies.

For more information, please download 21 Finance APP