21st Century Business Herald reporter Luo Yiqi reported from Guangzhou
The possible negative impact of the rapid outbreak of artificial intelligence is being highly valued.
Recently, the field of AI has once again triggered a round of open letters led by various industry leaders. Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear (To the effect that reducing the risk of species extinction from artificial intelligence should be a global priority, along with other society-scale risks such as pandemic diseases and nuclear war.) ）
(Screenshot of part of the open letter)
On the one hand, this drives the need for the rapid implementation of relevant regulations, and this action may need to be promoted more widely; On the other hand, it also shows that this round of AI industry development has come to a particularly rapid turning point.
At the recently held 2023 Xiaomanwaist Technology Conference and AIGC Artificial Intelligence Summit, Meng Meiling, professor and head of the Department of Systems Engineering and Engineering Management of Hong Kong Chinese University, also told the 21st Century Business Herald reporter that while it is clear that generative AI will bring great impetus to the digital economy, it is also necessary to consider that there are many risks. In terms of AI ethics, fake news, infringement and other aspects should be clear from the level of laws and regulations how to protect personal rights and interests, and individuals should also be vigilant in the process of using AI large models.
Looking back at history, Zhu Jiaming, an economist and chairman of the Academic and Technical Committee of the Digital Chain Digital Finance Research Institute of the Guangdong-Macao Intensive Cooperation Zone in Hengqin, said that the AI industry has developed for at least 70 years today, which can be divided into three stages: first, with Turing's "Machine and Intelligence" published in 1950 as a starting point; The second was an artificial intelligence conference held in 1956, until 2012, when deep learning entered a new stage, which lasted for more than 50 years; The third is the full arrival of the era of large models from 2022.
Zhu Jiaming pointed out in an interview that the basic feature of the era of large models is to integrate artificial intelligence and human life, economy, learning and education models more closely. This coercive change on humanity will be stronger than ever. Therefore, it is normal for human beings to have worries, "Now it is more important to 'literacy', so that the public should not be afraid and do not refuse, but first learn to understand and use." ”
The wave of AI development
During the summit, Zhu Jiaming said in his speech, "There are several basic characteristics in this era of large models: a large model is based on artificial neural networks; Legoization, all models will be combined in different ways to form a large model cluster; Pre-training promotes parameter scale, and large models cause the data storage scale to transform to EP, ZP and even YP stages. More importantly, it has the ability and mode to understand natural language, has formed a thinking chain, and then moved towards the thinking tree (ToT); Huge corpus is particularly needed; Artificial feedback and reinforcement learning mechanisms implanted in cybernetics; The implementation provides a great platform for hybrid quantum-classical computing. ”
The core values of AI are also more diversified: one is to accelerate the trend of artificial intelligence Internetization or Internet artificial intelligence; The second is to trigger a revolution in knowledge, learning and education, because it changes the knowledge map; The third is to change the paradigm of scientific research, that is, human beings have entered the historical stage where basic scientific research relies on artificial intelligence; The fourth is to accelerate the formation of cross-group hybrid intelligence, human beings are no longer the only component of intelligence, it will enter the era of thinking subjects with the mixing of human beings and artificial intelligence and machine intelligence; Fifth, it will trigger profound changes in the economic structure and economic system; The sixth is to reconstruct the model of human society, physical space and information space.
"Personally, I believe that the era of big models is promoting artificial intelligence towards generalization, that is, general artificial intelligence (AGI). This pace is accelerating. Zhu Jiaming believes that in the next 3 years, it will be the fourth wave of artificial intelligence development, mainly manifested in the combination of artificial intelligence and industrial applications, especially for everyone's life pattern will change, and will quickly penetrate into all stages of education from preschool to high school.
He told the 21st Century Business Herald reporter that when everyone discussed digitalization, their understanding was still relatively superficial. Today, the profound foundation of digitalization has become artificial intelligence, in this case, the policy layer needs to play an important role to give the original technology a new connotation, that is, to promote the full entry into the intelligent era.
As AIGC continues to evolve, everyone will have their own digital twin in the future, and the AI Internet will also come to the fore. "This amounts to a mass migration of human thinking patterns and behaviors, requiring a transition to the intelligent age." This will affect 8 billion people in the world, and it will take a long time. Zhu Jiaming said.
Specifically, "in the future, digital human agents will constitute the Internet, similar to WeChat groups that transcend the current people." But the most important thing is to make a big difference in the way people learn. From this point of view, he added, the profession that faces the greatest challenge should be the teacher, because the speed and effect of students' learning may be much higher than the speed of teacher teaching.
How big models affect society
At the specific landing level, how will the AI big model penetrate into the vertical field and even the professional role of individuals layer by layer?
Chen Shi, investment partner of Fengrui Capital, pointed out that from the perspective of generative AI entering the vertical industry, it is necessary to predict it from the perspective of the next ten years.
"Personally, I think that ten years from now will be the era of neural intelligence models. At the top level is the full-stack large-scale language model, which is a GPT4-like model, which aggregates all human knowledge, has or even exceeds human intelligence, and can empower all walks of life. But such opportunities will be few. He added that for the next level of industry, it may be necessary to establish an industry vertical model, which uses existing knowledge, rules, and unstructured text as training data to instill it into the language model to produce an intelligent model. This is similar to a large tool library that empowers industry process reengineering and empowers every step.
Further down, "enterprises must also have their own models in the intelligent era, and the enterprise model must have depth, otherwise it may be broken down." The so-called depth is whether the general model of the previous level has irreplaceable uniqueness, otherwise it will be easily replaced by the previous model. Chen Shi pointed out that the employee personal model will be divided into two parts, one is based on the ability and quality required by the position model, and it is constructed with thinking data and cognition; The other category is models that can be tooled, such as co-pilots, intelligent assistants, etc.
"I think in ten years' time, it's going to be the era of extensive model building, and the software layer will be very thin, and most of it will be models." He added that before entering the vertical industry, there is still basic work to be completed, that is, data and online.
Dataization solves the problem of data source, without data composed of knowledge, laws, etc., it is impossible to establish a model and train; Going online can make the scene truly embedded with intelligent capabilities. "The barriers of future industries or enterprises may be reflected in the models generated by the accumulation of these data."
Meng Meiling analyzed to the 21st Century Business Herald reporter that in the field of education, which has attracted much attention, AIGC can help teachers quickly solve some teaching work, such as giving preliminary suggestions when asking questions, and then teachers make targeted adjustments according to the suggestions. This process will lead to a significant increase in the overall efficiency of teaching and learning. At the same time, when revising students' essays, AI can also suggest some corrections, further helping teachers save time and focus on inspiring students.
"I think it's not just those of us who study science and technology who need to understand AIGC, ordinary people need to deepen their understanding of this technology." She also believes that humans should actively apply large model tools to continuously improve their productivity and efficiency.
"Although GPT4 is strong, the more applications you can still see the shortcomings. This means that the current GPT4 is still far from human intelligence. However, it has a large-scale knowledge base and powerful computing power behind it, and the limited memory of the human brain is actually not fair if it is compared. As for how far away from general artificial intelligence, it is still difficult to predict, but I estimate that the integration of people and AI will be stronger in the future, because where AI is weak, it is where human beings are strong. Meng Meiling concluded.
For more information, please download 21 Finance APP