Japan: Rationally Treating AI Myths

More than ten years ago, the third wave of artificial intelligence had just begun to emerge; and now, the AI ​​myth has become overwhelming and has become the focus of media hype. From the perspective of AI’s development process in the past 60 years, the previous two climaxes have failed because of In the real world, it has been widely used and eventually fell into a trough. This AI boom has taken advantage of deep learning as an opportunity to display the prospects that can be widely used and create many mythical miracles. But will myths be realized? For this, the reporter of the Science and Technology Daily Recently interviewed Dr. Yang Keji, senior researcher at the Japan Institute of Structural Planning.

At least for now, AI is not 'magic'

He is also a member of the Bridge Health Monitoring Committee of the Japan Civil Society. He is a member of the blast-resistant and impact-resistant design committee. He has long served as the president of the China Association of Science and Technology in Japan. When he began to develop a bridge intelligent detection and monitoring system more than ten years ago, AI has not yet affected. To civil construction industry.

Dr. Yang believes that today's AI is dependent on a very large amount of data. The technology implemented through machine learning and deep learning algorithms is only a black boxed thinking tool made out of data. It is not omnipotent. Learning methods Breakthrough achieves high prediction accuracy. As for how to achieve high precision, it is the result of high-density, high-precision interpolation of super-large amounts of data in the high-dimensional feature space. Therefore, it is necessary to use artificial intelligence to obtain accurate results. As a result, the premise is to have a large amount of data that can guarantee high-quality interpolation calculations, and to be able to accurately set the feature amount of the data. To be precise, the result obtained by the current AI is a correlation, not a causal relationship. Therefore, the AI At least for now it's not a magical tool.

Misunderstanding, hype and worship

Dr. Yang said that people’s biggest misunderstanding is to confuse the content that AI has already implemented, the content to be implemented, and the content that is unlikely to be realized. In fact, if you do not understand the practical ability limitations of AI, you may do it. The ignorance of judgment and planning leads to failure. A common failure is to use AI, but it is not possible to spend money as expected; the other is that although powerful AI tools are used, expectations are too high. By satisfying the results, we finally give up and lose opportunities for innovation. Those who have these two types of failure experience may instead strongly oppose the promotion of AI in the future and become a drag on the development of AI.

He believes that in order to make realistic judgments, we must first understand what AI can do. It is very important that we can do what we can and can do and what we can do in the future.

Secondly, on the technical level, the biggest topic of machine learning is the design of feature quantities. Determining the feature quantity manually depends on the level of familiarity of the setter with the object to be implemented. Deep learning solves this problem and achieves automatic extraction of feature quantities. The breakthrough only contributed to the third wave of AI. Now that AI's pattern recognition level is higher than that of humans, such as machines that beat world champions, etc., it has caused people to worship AI. In fact, AI is identifying objects and sequences. Competencies are far from humans.

In addition, it is difficult for AI to understand popularly as a major cause of worship. The more AI's performance improves, the harder it is to explain popularly, and the process becomes black boxed and becomes black magic, causing misunderstandings. One of the most important human actions is through association. Forming inference stories and so on, decision making and intuitive cognition are more linear, and it is easier for high-dimensional non-linear AI to produce worship phenomenon.

Master the data and master the future of AI industry

Avoiding the above-mentioned misunderstandings can more objectively analyze and predict the future of AI development. Dr. Yang believes that the key to the future development of AI industry is the collection and accumulation of data. Whoever masters the data will grasp the future of AI industry. For the third wave of AI, Japanese political and economic circles have not waited.

For the application of AI in the intelligent inspection and monitoring system for the construction industry, Dr. Yang said that in the future, the degree of damage in buildings, the explanation of the causes of damage, as well as the prediction, monitoring, monitoring and detection, and construction management, etc., will all lead to breakthroughs in the introduction of AI. Civil engineering circles in Japan predict that from 2050 to 2060, the civil engineering industry can realize the automatic detection and automatic repair of structural damage, automatically build houses in the required places, and intelligentize the entire neighborhood.

If intelligentization of neighborhoods is achieved, then the construction engineer is not unemployed? Answers to the reporter of Science and Technology Daily, Dr. Yang will not answer. After realizing the intelligence of the neighborhood, the engineers' work can be focused on the interactive interface, reducing the cost of AI calculation. On the other hand, it is responsible for decision-making responsibilities under very few data conditions.

According to data from the Japan Civil Society Society, the number of workers in the construction industry in 2014 was 3.43 million, and it is expected to be reduced to 2.16 million by 2025. The Japanese government put forward a target at the future investment conference in September 2016, and it is necessary to establish the construction industry by 2025. Labor productivity increased by 20%.

2016 GoodChinaBrand | ICP: 12011751 | China Exports