The first session of the Global Economic Development and Security Forum of Boao Forum for Asia, with the theme of "Economic Security and Sustainable Development under Great Changes", will be held in Changsha, Hunan Province on June 65438+ 10 18 -20, aiming at discussing how to deal with global economic risks and structural problems, and will discuss topics such as intelligent manufacturing and scientific and technological revolution.
Beijing business today today: The big model is called the core of the new generation of artificial intelligence ecology. Can you briefly introduce what a big model is? What problems will be solved?
Huang Tiejun: The big model is the basic general model that endows intelligence with various applications. Generally speaking, it is a basic platform for learning the knowledge and laws contained in data from massive big data, and condensing them into a large model on neural network to provide services for various general intelligent tasks.
For example, on the mobile Internet, cloud service providers can have many service capabilities, but without a carrier like App, it is difficult for users to get all kinds of cloud services. From this perspective, App itself is an industrial ecology. In fact, the big model also needs to solve similar problems at present.
The big model is a public service with strong radiation and technology. In the future, all walks of life will have some specific needs, and some enterprises need to develop large-scale model transformation and customized interfaces.
Beijing business today today: How will the big model connect the artificial intelligence technology ecology with the industrial ecology? What will be the application direction of artificial intelligence in the next information field?
Huang Tiejun: The understanding and application of artificial intelligence in many industries are still in the exploratory stage, and there is a certain distance between them. How to connect this interface, in fact, needs a group of enterprises, which can transform the ability of the big model into the content needed by all walks of life.
It is difficult to predict the application in the next information field. I think in reality, copywriting and information processing will be replaced by artificial intelligence, or most of them will be solved by the big model of artificial intelligence, which will bring great application possibilities.
All kinds of applications of search engines are, in the final analysis, the application of information organization mining. For example, individuals can collect data and do some information processing through search engines, but now the big model has solved the problem of collecting massive data. Its data is not made by any one person or a group of people, but collects all the data, which in turn serves various information processing applications of copywriting. Maybe the last exit is still people, but most of the tasks behind it are done by artificial intelligence. The possibility of this application direction is enormous.
Beijing business today today: How did artificial intelligence technology develop into the era of big models? What are the differences compared with the past?
Huang Tiejun: The development of artificial intelligence to large models is determined by the basic laws of the development of artificial intelligence technology. There are two schools of thought about artificial intelligence. One school thinks that the scientific mechanism, theory, mathematics and algorithm behind artificial intelligence are very important; Another school thinks that artificial intelligence is a general technology, and it is necessary to build an intelligent system and then understand its mechanism. The latter is the mainstream view of artificial intelligence.
In the process of building artificial intelligence, it was the work of a few researchers at first, and then some enterprises gradually participated. In the future, Industry-University-Research and the whole society will build a model. Why are you doing this? In fact, the reason is very simple. If the data learned by an artificial intelligence system or model is not complete and timely, it is hard to believe that its intelligent model ability is strong. The so-called big model is to integrate all kinds of data resources, the strongest algorithms and computing power in society into a public basic platform that everyone can use. This is the inevitable direction for us to build an artificial intelligence system.
In this process, the ability of big models is enhanced, which in turn will play a role in all aspects of society. If it works, more people will build it, and it is a benign iterative effect.
In fact, the potential of artificial intelligence depends on the data it can obtain, just as people "read thousands of books and walk Wan Li Road", so does artificial intelligence. The physical world and even the universe are so vast that if it can be transformed into data and information, artificial intelligence can learn, and this space will be very large.
I dare not say whether the big model will surpass everyone, but at least no one has obtained all this information, let alone discovered the laws behind it. The body and life cycle of each of us determine that the data we can obtain is still relatively limited.
Beijing business today today: What challenges need to be overcome in the development of the grand model? What is the future development path?
Huang Tiejun: Now there is a set of techniques and algorithms for training large models, but whether there is a better algorithm is still being searched and explored by academia and industry. At present, training an intelligent model requires a lot of carbon emissions, and training a model in the future may require less carbon emissions. I think that one day, the cost of artificial intelligence training may be less than that of people, which is another milestone.
Therefore, with more and more data contacted by artificial intelligence, and with the increasing efficiency of learning and training, its result is a large model, and it may be a super-large model and a huge model in the future, and it will continue to iterate. This road is already very clear.
But where is its upper limit, there is no basis at present. At present, the bigger the better is right. It is possible that it will not be a simple linear growth after expansion to a certain extent, and it is also possible that the growth will start to slow down after expansion to a certain extent, but these are still speculations.
Beijing business today today: How to consider some safety and ethical issues in the development of large models? How will it be avoided?
Huang Tiejun: The safety and ethical issues of artificial intelligence cannot be achieved overnight. For example, information security issues are constantly emerging with the development of information, and we have to solve them after they appear.
In the process of developing a large model, there are also some inherent risks, such as what knowledge the model has learned does not conform to ethics and principles, and these risks can be controlled in advance; But there are also some risks, which are brought about by the continuous progress of technology, and its solution needs to be solved by technical means. "You must untie the bell and the person who tied it." If this technology is not developed because of some potential problems, it is not in line with the law of scientific and technological development.