Will artificial intelligence surpass human intelligence in the future? Today, the AI industry has basically reached a consensus on this issue. Beijing Zhiyuan Research Institute gave the definition of artificial general intelligence (AGI): it must be an intelligence stronger than human beings to be truly called AGI. Therefore, autonomous intelligence, superhuman intelligence, and strong artificial intelligence actually talk about a kind of intelligence that surpasses human beings in an all-round way.
So when the generative artificial intelligence (GAI) model ChatGPT was born, and the long-dormant AI industry boiled again, collective speculation on artificial intelligence security occurred.
"There has never been a precedent in history where more intelligent things are controlled by less intelligent things. If frogs invent humans, who do you think will gain control? Frogs or humans?" - Godfather of AI, winner of the Turing Award , Geoffrey Hinton, one of the "Three Giants of Deep Learning";
"Imagine a future AGI system that might have 100,000 lines of binary code, and human supervisors are unlikely to find out if such a model is doing something nefarious." - OpenAI CEO Sam Altman;
"We should suspend the development of systems more powerful than GPT-4, not that we should suspend artificial intelligence, the purpose of the suspension is only to make artificial intelligence more like biotechnology. Making machines explainable is the key." - "Pause AI research Initiator of the initiative", Max Tegmark, current tenured professor of physics at MIT;
These views are excerpted from the "Zhiyuan Conference", the world's highest-level AI industry conference. In June this year, guests attending the 5th Zhiyuan Conference included 4 Turing Award winners: Geoffrey Hinton, Yann LeCun, Joseph Sifakis and Yao Qizhi, OpenAI CEO Sam Altman, Zhang Yaqin and other well-known academicians and scholars, as well as well-known artificial intelligence experts such as DeepMind and Midjourney. Smart team.
The organizer Zhiyuan Research Institute is an artificial intelligence non-profit research institution that is extremely low-key in the field of public opinion. In 2018, the year after the release of OpenAI's first-generation GPT large model, Zhiyuan Research Institute was established. At the beginning of its establishment, the goal of Zhiyuan Research Institute was clear and huge: to dig deep into artificial intelligence and produce a number of original innovations that are still meaningful in retrospect 20 years later.
Zhang Hongjiang is the soul figure with high hopes in this "big gamble". He has done R&D, management, and investment. He is well-known as the CEO of Kingsoft Group and Kingsoft Cloud, and one of the founders of Microsoft Research Asia. Zhang Hongjiang is known by the industry as the "founder" in the field of computer video retrieval research. He is one of the scientists with the highest impact factor in the world's computer field and is currently a foreign academician of the National Academy of Engineering.
"Hongjiang is the first wave of scientists who put scientific research into the industry. In the past 20 years, they have cultivated many artificial intelligence talents." Huang Jiangji, co-founder of Xiaomi Technology, told "Deep Web"; former vice president of byte, now intelligent research at Tsinghua University Ma Weiying, the chief scientist of the Academy, told "Deep Web", "Hongjiang is one of the few scientists in the world who can achieve a certain level of both academics and industries."
After becoming the chairman of Zhiyuan Research Institute, Zhang Hongjiang built a team from scratch. In 2020, Zhiyuan established a large-scale model team. In 2021, it launched the "Enlightenment 2.0" large-scale model with a parameter volume of 1.75 trillion, which is nearly 10 times that of GPT-3.5. It was China's first and the world's largest trillion-level model at that time Model. At the Zhiyuan Conference in June this year, Zhiyuan released a complete Enlightenment 3.0 large-scale model series (including language large-scale model series, large-scale model language evaluation system, visual large-scale model series, multi-modal models, etc.), and entered the comprehensive open source stage.
"I am a so-called debut (retired) person. I don't need to do something for my own position. I will not do some repetitive things." In an exclusive interview with "Deep Web", Zhang Hongjiang said that the research should be Coming out is the direction of Zhiyuan Research Institute, "it cannot be for publishing a few articles."
Zhang Hongjiang directly pointed to the problem of artificial intelligence research in the interview of "Deep Web", and answered the root cause of why Microsoft seized the opportunity of this wave of artificial intelligence. Zhang Hongjiang believes that basic research must either not be done, or it must be done as charity, and don't think about short-term returns.
"We have to ask, is there an environment that can tolerate researchers doing something at a low ebb. Conversely, we have to ask ourselves, how many of our scientific research practitioners are out of love for what he does?"
Zhang Hongjiang has always been full of confidence in young people, "Generation after generation is stronger, their burdens are less than ours, they are confident, their knowledge is wider than ours, and their global vision is much stronger than ours at his age."
Zhang Hongjiang agrees with the book "Why Greatness Cannot Be Planned", that the process of exploring any complex issue will be filled with countless deceptive "stepping stones", and the target may be a wrong compass. "Every effort is made to increase the odds. Funding agencies often ask me a question, what can you achieve in five years? I can't guarantee what I can bring, but my probability of success will increase."
In the past three years, "Deep Web" has had many conversations with Dr. Hongjiang Zhang, and the topics involved basic scientific research, large models, talent education, AI technology explosion, etc. The following is the actual record of the dialogue between "Shenwang" and Zhang Hongjiang, which has been deleted without changing the original meaning:
"Do Basic Research as Charity"
"Deep Web": Why is Microsoft seizing the opportunity of artificial intelligence?
Zhang Hongjiang: The core is that we can’t just ask why I didn’t do it every time we see some successful people after a certain trend comes. Why didn't China do it? Why didn't Tsinghua University or Peking University do it?
We have to ask, is there an environment that tolerates these people doing something when they are low. Conversely, we have to ask ourselves, how many of our scientific research practitioners are out of love for what they do, and how many are just as a profession?
Neural network is actually the core algorithm in the last wave of artificial intelligence boom. When technology ebbed in the late 1980s and early 1990s, people began to reject or abandon research on neural networks and turned to statistical learning. This is not a bad thing in itself. At that time, the neural network really couldn't go down, and the research must have taken many detours.
No one was investing in neural networks anymore, but Geoffrey Hinton (Vice President and Engineering Fellow at Google, Chief Scientific Advisor to the Vector Institute, Professor Emeritus at the University of Toronto, and Gatsby at UCL) The founder of the Center for Computational Neuroscience. One of the "Big Three" in deep learning) has not given up and is still doing it. And Yang Likun (Yann LeCun) (French computer scientist, chief artificial intelligence scientist at Facebook and professor at New York University, winner of the 2018 Turing Award, known as the father of convolutional networks. One of the "Big Three" in deep learning) is a Ph.D. The time was exactly when the last round of AI bubble burst, and he was not affected, and continued to study neural networks.
If we look at people like Geoffrey Hinton or Yann LeCun, what we should take away from them is that their love for the research they do doesn't change from high to low .
When the professional cost is too high, it is difficult to have breakthrough things. But this is an ecological construction, and we hope that Zhiyuan can continue to contribute to a virtuous circle of ecology.