Artificial intelligence, technology is ready, what about us？
In recent years, artificial intelligence technology represented by deep learning and reinforcement learning has expanded from engineering fields such as language translation, image recognition and industrial automation to economic and financial fields such as intelligent production, intelligent agriculture, intelligent logistics, macro-economic monitoring of big data, and quantitative investment and research, which can be said to be widely used.
Artificial intelligence technology has the inherent advantage of dealing with high-dimensional data, and can avoid many limitations of traditional analysis methods by means of representation learning, value function approximation and feature selection, and obtain better prediction and decision-making effects. In order to make artificial intelligence technology achieve satisfactory prediction and decision-making results, researchers often need to invest a lot of data resources. This technical feature makes data resources a key factor of production. With the popularization of big data, intelligence, mobile Internet and cloud computing, artificial intelligence technology, as the underlying technology to provide information products and services, is also the key to the gradual transformation of industrial economy to digital economy.
What is the artificial intelligence algorithm?
Artificial intelligence algorithms can be roughly divided into supervised learning, unsupervised learning and reinforcement learning. Among them, supervised learning learns laws from human experience through continuous training programs (models). In this kind of machine learning, researchers will constantly adjust model parameters by marking data to achieve learning purposes. Similar to parents will show their children different colors, sizes and even kinds of apples, and teach children to know "never seen" apples. This is the purpose of supervised learning: out-of-sample prediction.
Unsupervised learning enables the machine to extract features directly from the existing data and compress the information for other tasks through training programs. Like traditional principal component analysis, high-dimensional features can be approximated by low-dimensional vectors. For example, we can use principal component analysis technology to compress pictures to save storage space. Therefore, this kind of machine learning algorithm does not need previous experience, and is also called unsupervised learning.
Of course, the relationship between unsupervised learning and supervised learning is not contradictory to each other, and we can also use semi-supervised learning algorithm for data with partial labels. For example, the recently popular antagonistic neural network — — We can use this algorithm to learn a series of Oracle Bone Inscriptions, so that it can generate a number of "Oracle Bone Inscriptions" which is enough to confuse the false with the true, but never represents any meaning, which is equivalent to a calculation program that "paints a tiger as a tiger" but doesn’t know it is a tiger.
In addition, reinforcement learning is different from the above (none, half) supervised learning algorithm. Reinforcement learning is an extension of dynamic optimization, while (none, half) supervised learning is closer to statistics. Reinforcement learning can maximize the cumulative benefits of intelligent programs by making them constantly interact with the environment and adjusting the decision parameters (processes) of intelligent programs. Reinforcement learning is the machine learning algorithm closest to human decision-making process, which is similar to letting an agent perceive the world infinitely and quickly, and optimizing its decision-making process through its own failure or success experience. In this process, computer programs do not need teachers so much. Of course, intensive learning cannot be completely separated from supervised learning. For example, AlphaGo is a computing program trained by means of intensive learning, but in the first stage of AlphaGo training, researchers used a large number of human players’ chess manuals for AlphaGo to imitate learning, where human experience is similar to that of teachers; However, in ZeroGo, an upgraded version of AlphaGo, imitation learning has been completely abandoned.
In order to make artificial intelligence algorithms universally applicable, we often need a lot of data, computing power and effective computing algorithms. A large amount of data is equivalent to hiring a knowledgeable teacher to guide the computer program, and high computing power will give the computer program the ability to learn knowledge faster. An important direction in the field of artificial intelligence research is to continuously develop computing algorithms that can make more effective use of existing data and computing power, which is equivalent to providing better learning methods and paths for computer programs. Therefore, data annotation, cloud computing, chip design and algorithm development are the core parts of the artificial intelligence industry.
What impact does artificial intelligence technology have on social economy?
In fact, artificial intelligence technology originated as a discipline in 1950s, for example, McCarthy and others, the "father of artificial intelligence", proposed artificial intelligence in 1950s; The decision tree model originated from 1950s to 1960s, while the neural network model and Q-learning reinforcement learning algorithm, which are widely used at present, originated from 1980s. However, if artificial intelligence technology wants to reach the accuracy of human decision-making, it needs a lot of training (experience) data and high computing power, so it was not until 2000 that artificial intelligence technology was able to achieve leap-forward development.
With the blessing of a large amount of data and high computing power, some artificial intelligence technologies can replace humans to make large-scale accurate decisions, and also replace more and more manual jobs. Judging from the current impact, on the one hand, the application of machine learning has indeed replaced some traditional labors, resulting in the labor crowding-out effect: automated robots tend to make the production process unmanned, natural language processing technology can better complete most of the translation and even information extraction, and machine learning algorithms can even more accurately characterize the properties of small molecular compounds, thus reducing the labor and time consumption required for large-scale repetitive work to some extent.
On the other hand, like previous technological revolutions, the rise of machine learning has not only improved social production efficiency, but also created new jobs for society. Since the birth of the industrial revolution, steam turbines have replaced grooms and coachmen, spinning machines have replaced textile workers, wired telephones and wireless telegrams have replaced postmen, and electronic computers have replaced hand-operated computers, saving a lot of manual calculations. However, it should be noted that every scientific and technological progress has not caused a large number of social unemployment, but will change the original social production organization structure and produce new formats by improving the production efficiency and technological innovation of traditional industries. Scientific and technological progress not only changes the production technology of industrial enterprises, but also changes the work content of traditional industries, resulting in new job demands.
Like any other technological innovation, machine learning technology has different degrees of influence on different industries and different positions. For those who are engaged in production processes, the impact of machine learning is undoubtedly subversive. However, for those positions that need overall planning, innovation and interaction, machine learning can not make a significant impact at the current stage.
In addition, we also need to realize that artificial intelligence algorithms still cannot reach the level of "intelligence". Any technology is accompanied by security risks, and so are artificial intelligence algorithms. For example, most supervised learning algorithms do not have a clear logic generation process, which not only prevents researchers from effectively interfering with the algorithms, but also makes artificial intelligence algorithms less robust in the training and prediction stages. To give a simple example, in a classification algorithm, if we change a pixel on a three-inch cat photo, it may make the computer algorithm identify the cat as other items. This kind of practice is called reverse attack and involves artificial intelligence technology risks.
Just like other emerging industries in the early stage of development, due to the lack of early supervision, some enterprises will make improper use of their early advantages in data, computing power and algorithms, leading to the abuse of artificial intelligence technology, monopoly operation of some head enterprises, disclosure of private data and even enterprise operation risks caused by over-reliance on algorithm decision-making, which are the application risks and derivative risks of artificial intelligence technology.
Therefore, how to develop and lead this strategic industry has become the top priority at present — — We need to think about how to give full play to the social bottom function of the government during the period of intelligent economic transformation, and how to standardize the operation of the private sector when its computing power and scientific and technological level exceed those of the regulatory agencies.
What are the advantages of the revolution with intelligent technology?
Strengthen investment in research and development, coordinate industry development, achieve core industry leadership, and grasp the dominance of artificial intelligence technology.Artificial intelligence has become a basic core field related to national security and overall development. At present, although the development of artificial intelligence in China is generally upward, there are still many problems in basic research, technical system, application ecology, innovation and development. Therefore, taking interdisciplinary and application transformation as the starting point to drive basic research in the field of artificial intelligence, increasing financial investment in related research, optimizing expenditure structure, and implementing tax incentives for enterprises investing in basic research will all help to coordinate the development of the industry. Focusing on strengthening the originality and leading research in the core areas of artificial intelligence (such as algorithms and chips) can better grasp the dominant position of artificial intelligence technology.
Pre-oriented, professional and flexible industry and technology supervision can better regulate the development of the industry and create a good digital environment.On the one hand, the artificial intelligence industry will have negative effects on monopoly, diversification, privacy and ethics. Therefore, the implementation of the underlying algorithm supervision can effectively prevent artificial intelligence-related and derivative risks such as opaque automated decision-making and privacy violations. On the other hand, the current development of artificial intelligence industry is in the explosive period of technological innovation and industrial growth. While bringing development dividends to the social economy, the flexibility of its application forms and associated formats also means that the regulatory framework and ideas should be adjusted accordingly, so as to further play the dividends brought by technological progress. In addition, we need to equip a more professional industry supervision team, empower artificial intelligence supervision with artificial intelligence technology, standardize the artificial intelligence industry in a pre-position, specialization and flexibility, and flexibly formulate the supervision framework and implementation norms according to the actual development conditions of different artificial intelligence industries, so as to reduce unnecessary obstacles faced by the development and application of artificial intelligence technology, create a good digital environment, and further build the core competitiveness of China’s artificial intelligence industry.
Deeply integrate the real economy, develop the digital economy and explore new formats.As the core technology in the development of digital economy, artificial intelligence technology can effectively transform data production factors into actual productivity. The improvement of production efficiency and the change of production paradigm brought about by the deep integration of intelligent technology and real economy are important driving forces for China’s macroeconomic transformation and upgrading. Therefore, deep integration of the real economy should be a major goal of the development of the artificial intelligence industry. Exploring new formats and developing new models based on artificial intelligence technology, promoting the transformation and upgrading of traditional industries, accelerating the cross-regional flow of production factors, integrating market players, and smoothing the economic cycle at home and abroad are also the inevitable needs of fully basing on and giving full play to the advantages of China’s entire industrial chain and laying out industries with digital economy advantages.
Give full play to the market initiative and realize the simultaneous production, learning and research of artificial intelligence industry.The long-term healthy development of artificial intelligence technology is inseparable from a good market environment and industrial support. Micro-subjects can effectively smell business opportunities, and market economy has comparative advantages in exploring new formats and exploring new models. However, as a typical knowledge-intensive industry, the artificial intelligence industry needs a lot of research and development work and trains a large number of professional and technical personnel in the early stage. Universities and research institutes have comparative advantages in personnel training and technological innovation, and are important core forces in the artificial intelligence industry chain. Therefore, taking market demand as the leading factor, organically combining enterprises, universities and scientific research institutions, we will form complementary synergy in functions and resource advantages, and provide basic support for the development of the intelligent industry. Taking economic benefits as the starting point, we should mobilize the enthusiasm of scientific and technological personnel in colleges and universities, promote the transformation of scientific and technological achievements into real productive forces, and create a healthy and sustainable development ecology of artificial intelligence industry.
Improve the social security system and promote the system of individual development and skills training and re-employment.With the application of large-scale machine learning technology, the subjective initiative, individual innovation and overall thinking ability of the labor force are extremely important to social and economic development and personal development. However, there is still a technical gap between the traditional labor supply and the emerging labor demand — — The traditional labor force is not qualified for the job demand of emerging industries. Under this background, how to effectively promote the re-employment system of individual development and skills training, effectively bridge the technical gap, and how to adjust the social security system to make it more applicable to cross-departmental retraining and re-employment, to ensure people’s livelihood and to effectively improve social welfare, etc., are worthy of our further thinking and exploration.
(Author: Wang Xi, a researcher at Peking University Institute of Economics)