2019 China Artificial Intelligence Conference (Chinese Congress on Artificial Intelligence 2019, referred to as "CCAI 2019") It will be held in Qingdao from September 21st to 22nd. Dr. Xiao Jing, the chief scientist of Ping An Group, will give a keynote speech entitled “Intelligence + Financial Strategy Practice” for this conference.
Xiao Jing has long been engaged in the research of artificial intelligence and big data analysis and mining, and has been elected as an important international academic conference. The committee and members of the China-America Fund Review Expert Committee have held senior R&D management positions at Epson America Research Institute and Microsoft Corporation of the United States. At present, Xiaojing is responsible for innovative technology and product development and application in Ping An Group, including the development and application of technologies such as intelligent big data analysis in finance, medical and smart cities.
Deep Learning advances in face recognition, machine translation, and human-machine games, Caused a new round of "artificial intelligence fever." Before the opening of the conference, we may wish to first understand how Xiaojing, a scientist in the industry, sees the development of the current artificial intelligence industry. What thoughts has he made about the advancement and landing of intelligence in the future?
Future artificial intelligence should be like "crow"
Xiao Jing mentioned in public that artificial intelligence is not new. As early as the 1970s, artificial intelligence assistants appeared in TV commercials to help people check emails and realize voice interaction, but at that time computing power. Very weak; by the 1990s, due to the emergence of the Internet, data, storage capabilities, algorithms have significantly improved, artificial intelligence technology has begun to flourish; then deep learning emerged as unstructured data The processing has brought about a very big breakthrough, resulting in a lot of application scenarios, thus forming the artificial intelligence boom we see today.
For the current In the deep learning of the fire, Xiao Jing said that it has many shortcomings. For example, deep learning only considers relevance and ignores causality. AlphaGo knows where the chess will win, but I don't know why, it is not explanatory. In addition, deep learning is only Classification does not energize, it tells you to win, but you don't know how much you can win. Finally, deep learning is too dependent on big data. In AlphaGo Zero, it doesn't need large numbers. Training, because the rules of Go clear, complete information,The machine can realize self-training learning by combining deep learning and reinforcement learning. However, most scenes do not satisfy the conditions of complete information, so the accuracy of the model depends on a large amount of training data, which is often difficult to obtain. Therefore, Xiao Jing pointed out that there is a need to improve the method of deep learning, so that the machine can achieve interpretable, quantifiable, small data learning, readable and writable, adaptive and other capabilities.
On a more macro level, Xiao Jing divided the development of artificial intelligence into three stages: weak artificial intelligence, strong artificial intelligence and super artificial intelligence. He believes that the current artificial intelligence is still in the first stage, and it will “calculate” but not “calculate”. Its computational intelligence is far superior to human beings, but it still does not reach human level in terms of perception intelligence and cognitive intelligence. . Xiao Jing used the crow as an example to propose the evolutionary direction of artificial intelligence in the future. Crows who love nuts will take the initiative to stop at the traffic lights, throw nuts at the green light, crush the passing cars, and eat them when the red light is on. Xiao Jing said that in the future, artificial intelligence must at least be like a crow, and will "calculate" and think.
Universal AI is reserved for research institutions, companies should focus on solving actual business pain points< /p>
The breakthroughs in current artificial intelligence are mainly concentrated in the dedicated field. Many industry voices believe that general artificial intelligence is the future trend. In this regard, Xiao Jing in the corporate world has his own views.
In Xiao Jing's view, general artificial intelligence will not appear until the strong artificial intelligence stage.He believes that general artificial intelligence is a very interesting research direction, but in most practical businesses, general artificial intelligence may not bring additional application value like dedicated intelligence.
So he suggested that research on general artificial intelligence should be left to research institutions to study; for companies, they should focus more on developing smart technologies. Solve the actual business pain points. Even if you want to target general artificial intelligence, you should plan as a long-term goal, otherwise there will be unrealistic flaws.
For the application of artificial intelligence at this stage, he pointed out that it is very important to define the target scenarios and problems. The problem is too big, too general, often If you don't have good results, you should consider how to standardize and expand your application after the target is clear and targeted.
How to achieve Intelligent transformation?
With the development of artificial intelligence, more and more enterprises have proposed intelligent transformation goals, hoping to use artificial intelligence "Dongfeng" to achieve developmental transition.
Xiao Jing believes that today's intelligence is more complicated than the Internet revolution of the past 20 years. In his view The Internet's innovation model has been relatively simple for the past 20 years. It has only transferred traditional business to the Internet and created new channels, but it has not made much changes to the business itself.
Intelligence is a technological innovation. Under the premise of being familiar with traditional business processes, information transformation should be carried out first, information flow will be realized, and data will be completed.
Xiao Jing summed up the five factors that need to be established in the traditional enterprise intelligent transformation:
First, there must be technology, algorithms, computing platforms, Computational ability;
Second, there must be data to achieve data;
third There must be a scene, and iteratively in the actual scene, in order to make the intelligent program continue to improve, and finally really play a role;
fourth, there must be industry experts The guidance, such intelligent transformation can effectively solve the actual pain points, and not only the illusion;
fifth,There must be a top-down mechanism to coordinate the promotion.
Xiao Jing also proposed that the implementation of intelligentization should be gradually implemented. It is not the best way to use the most sophisticated and advanced deep learning as soon as you come up.
In Xiao Jing's view, the value of data mining is mainly reflected in three stages.
The first phase is business rules and business experience, and a comprehensive deterministic expert system and rules engine is required. Many aspects of a business process rely heavily on rules, experience, and knowledge that are often not available from historical data.
The second stage is the business intelligence stage, the purpose is to find the correlation between the data, the users are reasonably grouped according to the data characteristics and classified, mainly It is applied to the service "head-end users", that is, the customer groups with particularly obvious characteristics. However, it is often difficult for business intelligence to dig deep into multi-objective and multi-factor complex relationships.Especially the mining and utilization of weak correlation factors. In the era of big data, there is a significant long tail phenomenon, and it is necessary to dig as many weak correlation factors as possible to make full use of the value of big data. Therefore, it is necessary to enter the third stage.
The third stage is the artificial intelligence stage, using artificial intelligence techniques such as machine learning and deep learning to deepen and make full use of large The value of the data enables a more accurate analysis. Generally speaking, because long-tail customers do not have obvious characteristics and lack correlation with each other, they need to be touched by deep intelligence methods such as machine learning. At this time, it is especially important to model according to different projects, different groups of people, and different scenarios.