Beginning with the Dartmouth Conference in 1956, the logo artificial intelligence technology was officially born. Artificial intelligence has gone through three ups and downs, and now it has once again entered the public eye. In the two years of 2016 and 2017, artificial intelligence was very hot, especially AlphaGo defeated Li Shishi and Ke Jie. In the world of Go, AlphaGo can be said to be rampant and no one can compete.
Starting with AlphaGo, artificial intelligence has entered the public's field of vision. The media is competing to report that the artificial intelligence is heated to a very high level. Although artificial intelligence is still very hot, the heat has been taken away by the blockchain. But whether it is the world's capable countries or the powerful international technology giants, they are all in the field of artificial intelligence. For example, Google has acquired more than a dozen companies related to artificial intelligence, including those acquired by Google. Developed AlphaGo's DeepMind.
Undoubtedly, the future of artificial intelligence is bound to be the direction of our development. So what are the pain points that need to be solved in the process of artificial intelligence development? The Innov100 platform analysis considers the following 10 pain points.
1. Scarcity of talent
Artificial intelligence is extremely scarce in terms of talent. According to LinkedIn data, there are less than 250,000 people in the field of labor in the world, of which the most talented people in the United States. The rest are mainly distributed in Europe, India, China, Canada, etc. Less than 30% of the work experience has more than 10 years. So in that country, companies want to make achievements in the field of artificial intelligence, the first is the competition for talent. For example, ZTE was recently sanctioned by the United States, causing questions about “National Coreâ€. The first problem is the scarcity of talents, the excellent level of talents, and the degree of talent concentration, which determines whether a company is going to be glory or decline.
2. Moral value judgment
How to make choices when artificial intelligence encounters an injury event.
For example, driverless driving is also in full swing. Artificial intelligence is likely to be applied first to the driverless field. However, in the unmanned field, sometimes this happens. When a driverless car is walking on the road, a person suddenly rushes out in front of him, and there are people on the left and right sides. No matter how the vehicle is operated, it cannot be avoided. Then how to make artificial intelligence at this time.
In the world of computers, it is a world of probability. If the elderly are in front of you and the children are left and right, will it be more valuable to analyze children than the elderly, and then complete all the optimal solutions. But this is obviously contrary to moral common sense.
3. Moravik Paradox
Artificial intelligence, simple understanding is the intelligence like human beings. Then the logic or method that artificial intelligence follows should be human-like. However, artificial intelligence is completely different from human intelligence.
Moravec's paradox is a phenomenon that is found by artificial intelligence and robotics to be familiar with common sense. Unlike traditional assumptions, computers require very little computing power, such as reasoning, to perform high-order intelligence that is unique to humans. But completing unconscious skills and intuition requires great computing power. This concept was explained in the 1980s by Hans Moravik, Rodney Brooks, Marvin Zucker and others.
4. Calculation limit
At present, the construction of an artificial intelligence computing platform requires a large amount of CPU and GPU. The TPU used by Google's AlphaGo is similar to an algorithm chip with GPU, and the energy efficiency ratio is very high. The power required to train AlphaGo is equivalent to approximately 12,000 consumer-grade 1080TIs on the market, at least 10 million.
For giant companies like Google, Facebook, Tencent, etc., this overhead may not be a big deal. But for some smaller companies, this will be a very big problem. After all, artificial intelligence wants to enter the mature period, and must solve the problem of computing power.
5. Privacy security issue
Privacy and security issues are very important topics in many industries. Why does the privacy of the artificial intelligence industry become a pain point for him?
Because if you want to use artificial intelligence to improve people's living efficiency and quality, you must get as much personal information as possible. Because the AI ​​model needs training, it is very likely that you need to upload your personal information to the cloud. In addition, there is currently no way to rely on local computing power to support artificial intelligence. Privacy and convenience are often contradictory, but if artificial intelligence wants to have good development, it must be both.
Recently, because of the leakage of up to 80 million user information, Zuckerberg was asked by the US government to attend the US Congressional hearing and was questioned for 10 hours. In the hearing, he repeatedly mentioned the use of artificial intelligence to solve some business needs.
6. Need a lot of data tags
Currently existing AI models require a large amount of data tagging, since most of the models are supervised learning models. A large number of data tags will not only require more human resources, but also people's participation will inevitably bring a certain degree of error to the data.
The problem that can be solved very well at present is to use reinforcement learning to conduct unsupervised learning. Google’s AlphaGo is trained using unsupervised learning.
7. Data scarcity
The AI ​​model not only requires manual identification of information, but also requires huge amounts of data to achieve the correct recognition of human beings. Take AlphaGo as an example. In the version of AlphaGo that defeated Li Shishi, 30 million maps were learned. Defeat the version of Ke Jie, carried out more than 4 million times, self-game.
In addition to the huge demand for data volume, the dimensions of the data are also required to be as comprehensive as possible. In short, the best you can give me, the more comprehensive the better. But the reality is that structured and comprehensive data is difficult to obtain in real life, and it is difficult to obtain more accurate data.
8. Black box problem
At the beginning of artificial intelligence design, the corresponding development direction is to perform corresponding tasks according to artificial rules and artificial manufacturing logic. However, there is no way to actually make artificial intelligence have a very satisfactory practical application.
Until now, very popular deep learning, through a certain degree of human intervention, AI model through data training and results intervention, will generate a fitting algorithm to generate human expected results. However, since the AI ​​model is automatically generated, there are unexplained problems. If one day the AI ​​model draws or does something unexpected, we have no ability to explain the reasons behind this time.
9. Poor portability of the model
The availability of the AI ​​model increases as the amount of data trained increases. However, the amount of data required is a very large order, but even a model with a high degree of repetition has no way to gain some experience from the previous model, and can only start training from scratch.
This will bring some problems, such as increasing data acquisition costs, time costs, energy costs, etc., which will cause great problems for developing companies. The poor portability of the model will inevitably affect the speed of development of technology and increase the cost of communication. This is indeed a very important issue for an emerging technology that needs rapid development.
10. Trustworthiness
This is a compound problem. The AI ​​model may perform very well in applications that measure measurable results. For example, in the field of image recognition, we can use a certain probability to evaluate the correct recognition of the AI ​​model, or it can be said that the trustworthiness is clearly visible.
However, if you are in a future application scenario, you need an AI model to provide some business decisions, or give some advice. We do not have a good reference system to evaluate whether the decisions and recommendations of the AI ​​model are accurate and whether they are optimal solutions. This will affect the direction and accuracy of the decision-making, and will also cause unnecessary trouble for many business people.
At present, there are already many intelligent models in life to assist decision-making, but such models can all be explained, and only play a certain reference role for decision-making. But the AI ​​model, we can not explain the reasons for its conclusion, then its credibility will be an inescapable problem.
Artificial intelligence has been around for 60 years since its birth. Nowadays, it has begun to play an important role in some areas. Although there are many pain points to be solved, which new technology has not experienced various problems from the beginning to the maturity. Moreover, the future potential of artificial intelligence is enormous, and I believe that there will be better development and breakthroughs in the future.
DADNCELL alkaline zinc-manganese dry battery has the characteristics of low self-discharge, which is suitable for household batch procurement storage, and has the discharge performance of continuous discharge time with more long-lasting and large instantaneous discharge. Widely used in flashlights, smart door locks, infrared thermometers, cameras, flash lights, razors, electric toys, instruments, high-power remote control, Bluetooth wireless mouse keyboards and other common household electronic equipment instruments. DADNCELL battery advocates providing a more comfortable and smooth power supply experience for household appliances. The alkaline battery capacity and discharge time of the same model of DADNCELL battery have reached four to seven times that of ordinary batteries, respectively.
The performance difference between the two is greater at high and low temperatures, because its unique internal components are structure and battery materials with better performance, and the power capacity and electrical performance can be improved.
At present, all types of batteries developed by DADNCELL Lab do not involve any heavy metals in production, use and waste. They are green and environmentally friendly and can be treated with domestic waste.
1.5V Am-4 Alkaline Battery,Ideal Aaa For Household And Office Devices,Batteries Long Lasting,Lr03 Non-Rechargeable Batteries
Shandong Huachuang Times Optoelectronics Technology Co., Ltd. , https://www.dadncell.com