Summary and Reflection on the hottest AI field fiv

  • Detail

Summary and thinking in the field of AI: five challenges in the future

these two days, a friend sighed: "2016 is a year of extraordinary significance for artificial intelligence. Perhaps the perception in the field of technology is not obvious, but the" success "at the commercial level is unprecedented." Yes, from the go war between alphago and Li Shishi at the beginning of the year to the press conferences related to artificial intelligence. Anyway, AI has finally jumped out of the confinement of the laboratory and become the core force active in the field of science and technology

first, the era of artificial intelligence is coming

whenever a thing rises, it is followed by a large number of views and conjectures, of which the most popular are often the boldest; Then every additional argument will make us more convinced of this view. For example, after alpha go defeated Li Shishi, AI recovered strongly in public opinion, and then Robin Lee's remarks at the world Internet Conference also strengthened people's attention to it again

not only Baidu, but also Ma Huateng said in his speech in June 2015: AI is what I want to do most. Ma Yun also wrote in his internal letter in May 2015: cloud computing, big data, artificial intelligence and other technologies will make countless dreams come true in the next three decades

at present, international Internet giants have entered one after another. Alexa of Amazon, Siri of apple and Cortana of Microsoft, as the first stepping stone to artificial intelligence, have been widely used; Search, translation, maps, unmanned cars, the shadow of deep learning everywhere, artificial intelligence is reconstructing human life

at the same time, with the rapid development of Internet and the continuous progress of underlying technology, the "energy" required by artificial intelligence is constantly improving

data volume: since 2000, the rapid development of Internet and mobile Internet has enabled the accumulation of data. According to IDC prediction, the total amount of big data in the world will be 40zb in 2020, of which 70% will be stored in the form of pictures and videos, which provides rich soil for the development of artificial intelligence

deep learning algorithm: students of Geoffrey Hinton, a professor at the University of Toronto (who is committed to the study of neural networks and deep learning), used the algorithm of deep learning in Imagenet, a well-known image recognition competition in the industry, to reduce the recognition error rate by 10% in one fell swoop, even surpassing Google. Deep learning became famous. In 2015, the visual computing group of Microsoft Research Asia won the competition, reducing the system error rate to 3.57%, which has exceeded the human eye

High Performance Computing: GPU has fast response speed and low demand for energy. It can process a large amount of trivial information in parallel and analyze massive data at high speed, effectively meeting the needs of the development of artificial intelligence

infrastructure cost: the popularity of cloud computing and the widespread use of GPU have greatly improved computing efficiency and reduced operating costs to a certain extent. IDC reports that the cost of data infrastructure is declining rapidly, from $9 per unit in 2010 to $0.2 in 2015

at the same time, giants and start-ups have also invested resources and costs to explore commercialization, but the technology itself still has enough room for growth, which is still in its early stage

second, opportunities brought by AI

we can see that at present, enterprises in the field of AI mainly focus on the following three levels:

foundation layer: pay attention to the foundation of AI supporting hardware or data platform

technical level: including algorithm and technical design related to machine recognition and deep learning

application layer: including general applications and industry vertical applications

according to Jiang Tao, founding partner of geek group venture capital, large companies take all the winners in these three levels, while small companies can only rely on a single point of breakthrough and break through on the advantages of traditional industries

the main battlefield of large companies (with a market value of more than 10billion) is to compete for the commanding heights of future artificial intelligence, which is divided into two directions. The first direction is to compete for the entrance of future artificial intelligence, including the entrance of home and car. These future entrances play an important role in interaction, such as voice interaction of Google and Baidu brain of Baidu

the second direction is the competition of the ecosystem. The entry is easy to switch, so we need to improve the switching cost through ecology, through open source technology, through recommended algorithms, of course, we also need to rely on the extension and development of IOT. For large companies such as and Dangdang, their biggest competitiveness lies in logistics and massive data, so they can buy technology, but they are not so anxious

the application of the main battlefield of small companies in the vertical field, through the wave of artificial intelligence to improve the industry that has not completed mobility. For example, the market scale and space of the financial industry in the era of artificial intelligence should be larger than that in the mobile era; For example, enterprise level services are now in a very backward state in China. Jiang Tao said, "the relatively easy things have been done, and the rest are tough, but I believe there will be big companies running out, of course, companies with data will be easier to run out."

in fact, at present, the application and landing methods of artificial intelligence are extremely limited. Almost all the latest advances in artificial intelligence are accomplished through one type: input data (a) quickly generate simple responses (b), for example:

such a simple input a and output B will change many industries, and the construction of a → B technology is called supervisory learning. The a → B system is developing rapidly, and deep learning is largely inspired by the working principle of the brain. However, the a → B system is far from the robot with emotion in science fiction films, and human intelligence is far more advanced than the a → B system

what can a → B system do? As for its disruptive impact, here is a rule: if human beings carry out a task that takes less than one second to think, then perhaps we can use artificial intelligence to automate this task in the near future

Wu Enda, chief scientist of Baidu, is one of the most authoritative scholars in the field of artificial intelligence and machine learning in the world

Wu Enda, Baidu's chief scientist, said that people have done a lot of valuable research in the application of artificial intelligence: detecting suspicious behavior in surveillance video, braking automatically when the car is about to hit a pedestrian, and automatically deleting the Yellow storm content on the, all these tasks can be completed in one second. Of course, these technologies are more suitable for combining with large industrial businesses

interconnection realizes that infrastructure can run and data can be connected. In fact, AI improves our overall application efficiency in another dimension. It tries to solve the problems of means of production and labor. Artificial intelligence is a powerful tool for the upgrading of industrial intelligence, which is changing all fields including communication, medical treatment, education, etc

1. Communication field

communication networks generally have two major tasks, one is network control, and the other is network management and maintenance. Network control is how to effectively schedule resources in a communication network, so as to improve the efficiency of network use and better serve users. Network management and maintenance is to accurately understand network requirements and optimize network design and deployment; It can perceive the network condition in real time and eliminate the fault in time. Artificial intelligence will make the future communication network less and less need people. The control of the whole network is basically fully automatic, and the whole communication network can be completed with little expert participation

2. Medical field

when introducing the application of Baidu artificial intelligence in the medical field, Robin Lee mentioned four levels, namely o2o service, intelligent diagnosis, genetic analysis and precision medicine, and new drug research and development

the first level: Baidu doctor now has 500000 doctors participating in the consultation, and a total of 8million people have obtained relevant medical services through Baidu doctor platform

the second level: in the small test of intelligent inquiry, the diagnosis of Baidu doctor is consistent with that of Peking University International Hospital in 80% of cases, and it may perform better in some rare cases. Of course, in addition to machine learning a large amount of medical knowledge of enterprises because of what is well-known and why they produce experimental machines for so many years, these technologies also need to constantly improve the understanding ability of patients' statements

the third level: using genes to treat diseases, the biggest problem is that most of the diseases caused by known genes are caused by single genes, and most of these diseases are rare diseases, and most common diseases are caused by multiple genes. Through a large number of calculations, artificial intelligence can help doctors figure out which genes work together to cause a disease

the fourth level: the number of small molecular compounds known today that may form drugs is about the 33rd power of 10, which may be more than all atoms in the universe combined. How can such a quantity be combined with the protein that produces disease with its molecular formula to treat disease? How to screen a large number of unknown molecular formulas to find effective new drugs? Computer science and artificial intelligence can help in this regard

3. Education field

the education industry is actually an industry with high trial and error costs, and no one will take children's grades as an experiment. The same is true in the medical industry. It is true that AI can give advice on image recognition and diagnostic analysis, but who will bear the burden in case of medical disputes or the delay of the patient's condition

on the other hand, these two industries have a long decision-making chain. It involves many stakeholders, including schools, teachers, parents and students in the education industry, and hospitals, doctors and patients in the medical industry. At the same time, these two industries are relatively highly regulated by the state

Jia Jing, partner of deloupe capital, said that no matter how difficult the education and medical industries are, capital is still very concerned. Because users who pay for education and health have strong willingness and ability to pay. Although this road is tortuous, its future is particularly bright

in fact, the education industry is more advanced than the medical industry. At present, there are many applications of AI technology in the education industry. For example, AI is deeply involved in teaching, learning, practicing, testing and evaluation, and speeds up the process of personalized teaching. However, this requires the accumulation of a large number of real and effective data. Whoever can accumulate enough data in the whole education link is likely to get ahead

on the other hand, the United States accounts for 22.1% of them. The problems that the education industry has been trying to solve are how to achieve economies of scale on the supply side and how to train and manage teachers. If AI is involved in the education industry, 70% to 80% of the problems previously solved by teachers may be solved by AI. This reform has been carried out in terms of production costs, which fundamentally solves the distribution of means of production and labor, not just the minimization of transaction costs. Therefore, AI brings much greater changes to the industry than mobile Internet

4. To C application

some AI companies that came out a few years ago have relatively mature technology development. For example, iFLYTEK, which just came out to make products in those years, is not so smooth, but now it has done well. therefore

Copyright © 2011 JIN SHI