The era of intelligence has entered a new chapter, and new technologies such as AI (artificial intelligence) are accelerating the integration with various industries in the real world. The question also arises: Where will the surging tide of AI take us? How should we face technological change?
From September 5th to 7th, the 2024 Bund Conference was held in Shanghai. As a globally influential financial technology and cutting-edge technology conference, the Bund Conference has built a bridge between technology and humanities and has become an important window for observing the implementation and development of China's AI industry.
The Huangshan Scenic Spot intelligent agent was officially released, iFlytek Spark Medical large-scale model, Guodun Quantum product display, the School of Cyberspace Security of the University of Science and Technology of China was one of the organizers of the exhibition competition of the Global Deepfake Offensive and Defense Challenge... many "Anhui scenes" "Anhui Elements" also appeared simultaneously, demonstrating the humanistic depth and warmth of science and technology together with the Bund Conference, making "the more technology moves forward, the more it benefits everyone" continues to become a reality.
More professional AI
Just like the steam engine changed the world because it began to integrate with ships, trains, and textile machines, the field of AI is also undergoing a key evolution from "Battle of Hundreds of Models" to "Industrial Applications."
Digital avatars assist in completing work, brain computers assist in deep sleep, and artificial muscle fibers give you a warm hug... The Bund Conference was a concrete demonstration of the current global craze in the AI industry. Experts predict that industrial implementation within five years is the key to the continued evolution of large model technology.
As the saying goes, "Leave professional things to professionals." The needs of each person and each industry are unique. For artificial intelligence that empowers thousands of industries, more professional problems may require more professional AI to solve.
Therefore, promoting large models from general capabilities to personalized applications is one of the core propositions of the AI industry.
At this Bund Conference, Huangshan Scenic Spot and Alipay launched the "Huangshan Space Intelligence" with a "space" service concept for the first time, using the Ant Bailing large model as the technical base. As a result, Huangshan Scenic Spot became the first "full AI escort" in China. Scenic spots.
As a product form of generative artificial intelligence, intelligent agents are the core direction of current large model applications, allowing large models to grow "hands and feet". The spatial intelligent agent uses "AOI" (Area of Interest) as its main service organization and presentation method. When the intelligent agent senses that the user has entered the Huangshan Scenic Area, it will use AI to analyze and push the services that the user is most likely to need. .
For example, before a user purchases a ticket, the "Huangshan Space Intelligence" will inform the user of the temperature of the scenic spot, clothing recommendations, and intelligently recommend route guides; when the user arrives at the entrance of the scenic spot, the ticket QR code will be automatically retrieved; when the user recognizes When the user starts climbing the mountain, each scenic spot along the way will be explained to the user; when the user reaches the exit of the scenic spot, interesting data about the trip can be generated with one click. In addition, shared bicycles, online ride-hailing and other services commonly used during travel can be quickly invoked and opened.
"Under the wave of AI development, the cultural tourism industry is also embarking on and experiencing an unprecedented revolution in technology and scenarios." Zhang Dehui, chairman of Huangshan Tourism Development Co., Ltd., said that compared with the traditional "one-code tour" and "one-machine tour" model, "Huangshan Scenic Area Intelligent" integrates large model technology technology, cultural tourism knowledge base, big data algorithm, voice interaction and the latest application products that can realize independent learning and self-evolution. "Although it is an initial product for exploration and trial, it needs continuous iterative upgrades, but this marks our progress in cultural and tourism development." The digital transformation of the travel industry has taken another revolutionary step.”
Enlightenment from “Questions of Science, Technology and Humanities”
Every technological innovation in human history is accompanied by "excitement" and "anxiety".
How to deal with AI competing for jobs with humans? Will human thinking ability degrade due to AI? How far is the era of "human-machine symbiosis"? Is it possible for AI to transcend tool properties, gain consciousness, and even establish an AI civilization? … It can be seen from the Ten Questions on Science, Technology and Humanities released at the Bund Conference that the public’s feelings towards AI are more complex, with a lot of technological anxiety.
The future is upward and forward, and science and technology are good and beautiful. No matter how breakthrough technology changes, people-oriented is always the first principle.
At the Bund Conference, Zhou Jian, a 24-year-old man who lost his right arm at the age of 12, used artificial intelligence and brain-computer interface technology to play a piano song "Big Fish" with his bionic hand, letting the audience express their appreciation for the development of technology to serve mankind. Well-being is intuitively felt.
It is worth noting that while new technological changes bring new scenarios, new businesses, and new markets, they also bring unprecedented risks and challenges in terms of technological ethics. How to promote the rapid development of AI technology while ensuring its safety, reliability and controllability, so that AI can benefit human society more responsibly?
Take AI large models as an example. Currently, large models are being implemented in various industries at an accelerating pace. In particular, the reshaping of rigorous industries such as medical care and finance is unprecedented. However, these rigorous industries also have the highest ethical standards and the most complex ethical relationships. field.
Zhao Jinghe, vice president of iFlytek Medical Research Institute, shared the current application progress and safety ethics exploration of iFlytek's Spark medical model at the Bund Conference. Regarding the problems faced by the application of large models in medical scenarios, he believes that "the main issues are data privacy, intellectual property rights and moral responsibility. Only by allowing large models to truly solve these problems can we truly assist doctors, patients, scientific research and other scenarios." .
“In terms of interpretation of the law, it is not only necessary to regulate and legislate at the government level to ensure the legality and compliance of large-scale models; companies must also abide by the nature of medical care, take measures to strengthen data protection, and obtain necessary copyright authorization. Ensure the fairness and compliance of the model while improving transparency and strengthening public education," said Zhao Jinghe.
As one of Anhui's important scientific and technological innovation companies, Guodun Quantum also displayed a variety of products related to office information security. "For example, our Bluetooth headset incorporates the voice communication privacy protection technology independently developed by Guodun Quantum, which can effectively protect users' private information during calls, voice messages, etc., making users feel more at ease during communication." Guo said. Zhang Rutong, technical director of Shield Quantum Solutions, introduced.
Some experts are more optimistic about how to ensure that technology always benefits human welfare. "Many of the problems we worry about today may not be the hardest to solve. The real challenges may be unknown problems we haven't imagined yet. One of the best ways to achieve the AI-driven future we want is Get actively involved," noted futurist Kevin Kelly said at the Bund conference.
Use AI to control “AI fraud”
With the rapid development of AI technology, deepfake technology (Deepfake), which uses deep learning algorithms to simulate and forge audio and video content, has become increasingly mature. Vicious incidents such as "AI face-changing fraud" that have occurred in recent years have triggered people's interest in the application of technology. Thinking about risk.
It may be an effective way to use AI technology to accurately identify and prevent the risk of "AI face-changing" and "use AI to defeat AI". However, can the fakes created by AI be identified by AI?
A competition focusing on the fraud risks of "AI face-changing" technology provides the answer. In the exhibition competition of the AI Innovation Competition of the Bund Conference and the Global Deepfake Offensive and Defense Challenge, the question was jointly presented by the School of Cyberspace Security of the University of Science and Technology of China and Ant Digital Dimensity Laboratory. The contestants used the AI models they studied to modify the AI. Detect and identify images and videos that have been processed.
In the image track, there are 9 pictures that are difficult to distinguish between true and false with the naked eye, but the contestants need to use their own participating models to give the probability of forgery for each picture, and compare the two photos with the highest probability values with the correct answers. In the video track, players also have to use AI models to identify the authenticity of videos.
It is worth noting that in terms of on-site contestant scores, on some moderately difficult questions, the results of both sides were close, and the forgery recognition rate was almost above 80%. However, on some highly difficult questions, the model results were very different.
Someone at the scene asked: "Why can't AI fully answer a picture that looks fake to the naked eye? And a picture that is difficult to distinguish with the naked eye, AI can recognize in one second?"
In this regard, Zhou Tianyi, chief scientist of the Frontier Artificial Intelligence Research Center of the Singapore Agency for Science, Technology and Research, believes that humans rely on intuition and logic to judge whether an image is fake. For example, if another person's face is replaced with a movie actor's face, humans can tell at a glance. The AI doesn't understand the story behind it. But AI is not as susceptible to emotional interference as humans, and it will complete instructions meticulously. From this perspective, AI is “superior” to humans in terms of work efficiency and stability.
However, developing detection algorithms is only one way to curb the abuse of deepfake technology. Yao Weibin, one of the game question makers, admitted that this method is relatively lagging behind. To prevent fraud from the root cause, we should require the establishment of a universal identifier for AIGC content from the perspective of standards and legislation. For example, every video or picture generated by AIGC should have a signature or watermark in the file information, so that platform management has more basis.
"For the general public, deep forgery is now very likely and has gradually entered our lives, and even leaked people's privacy." Professor Zhou Wenbo of the School of Cyberspace Security, University of Science and Technology of China, suggested not to do so easily in public places. Show your own face data and avoid the leakage of face data as much as possible. "As the synthesis technology gets better and better, we will be deceived more and more deeply, but as long as we can stay vigilant, we can still have a big impact." Probability is not compromised by fraud.”