At one time, artificial intelligence was seen as a future technology. But nowadays, people want to see the future beyond artificial intelligence. In the context of the gradual development and rise of the Internet of Things, robotics, nanotechnology and machine learning, this paper attempts to interpret people's views on the development of artificial intelligence in the next five years.
I. Summary
Obviously, artificial intelligence has had a huge impact on many areas in the past few years. However, what people are thinking about now is what areas of artificial intelligence will develop in the next five years. The author believes that it is necessary (writing an article) to describe some of the trends we are seeing today, and to make some predictions about the future development of machine learning. The list presented below does not necessarily exhaust all possibilities, and the reader does not need to take it as a standard. But they stem from some of the ideas that the author thinks are useful when considering the impact of artificial intelligence on our world.
Second, the 13-point prediction about artificial intelligence
1. The amount of data required for artificial intelligence work will become less. Companies such as Vicarious or Geometric Intelligence are working to reduce the size of the data sets needed to train neural networks. The amount of data used to train artificial intelligence is now seen as a major obstacle to its development and a major competitive advantage. At the same time, using the probabilistic induction model (probabilisTIc inducTIon, Lake et al., 2015) can solve this major problem in the development of artificial intelligence. An algorithm that doesn't require a lot of data will eventually learn, absorb, and use the concept in a rich way, whether it's action, imagination, or exploration.
2. The new learning model is a key element. A technology called Transfer Learning allows a standard Reinforcement Learning system to be built based on previously acquired knowledge – something that humans can easily do. It is part of the Incremental Learning technology. MetaMind is studying the problem of MulTItask Learning. Among them, the same neural network is used to solve different types of problems, and when the neural network can perform better on a type of problem, then it can perform better on other problems. The next step in MetaMind's development is the introduction of the concept of a Dynamic Memory Network, which answers specific questions and infers the logical connections between a series of discourses.
3. Artificial intelligence eliminates human (cognitive) bias and allows us to become more "artificial". Human nature will change because of artificial intelligence. Simon (1995) stated that humans do not make completely rational choices because the optimal selection is costly and because of the limited computing power of the human brain (Lo, 2004). What people often do is to seek a satisfactory solution, that is, to pick out at least the choice that makes them happy. Introducing artificial intelligence into life may end this situation. When (equipped with artificial intelligence) humans are no longer constrained by computational power, this will eventually be answered once and for all, whether cognitive bias exists and is human instinct, or whether these behaviors are only carried out in a limited information environment or under restrictive conditions. A shortcut to decision making. Lo (2004) believes that the satisfaction of human beings (when making decisions) is formed in a series of evolutionary attempts and natural selection processes. In it, individuals make predictions based on past data and experience and make choices. They learn based on positive/negative feedback they receive and are able to heuristically resolve related issues quickly. However, once the environment changes, the adaptation process is somewhat delayed and slow, and some old habits are not able to adapt to new changes – which creates behavioral bias. Artificial intelligence reduces these delays to zero, and virtualization eliminates any behavioral bias.
In addition, based on experience learning over time, artificial intelligence becomes a new tool for change: we usually do not evaluate all alternative decisions because we cannot think of all decisions (limited knowledge space).
4. Artificial intelligence will be fooled. Today's artificial intelligence is far from perfect, and many people are focusing on how to deceive artificial intelligence devices. A recent algorithm called Adeversarial Examples; Papernot et al., 2016; Kurakin et al., 2016 was developed and is the first method to mislead computer vision. Intelligent image recognition software is fooled by subtle images that misclassify these images. But what's interesting is that this method does not deceive humans.
5. The development of artificial intelligence is accompanied by risks. The mainstream voice believes that artificial intelligence is becoming a potential disaster for human beings. When a super artificial intelligence system (ASI, ArTIficial Super Intelligence) is created, perhaps its wisdom far exceeds that of human beings, and even it can think of and do what we can't predict today. Despite this, we believe that in addition to these terrible threats related to human survival, there are still many risks associated with artificial intelligence. What we do and how to do super artificial intelligence, the hidden risks behind it are actually incomprehensible, whether they have a positive or negative impact on people. Furthermore, in the process of transforming from Narrow Artifical Intelligence to strong artificial intelligence to super artificial intelligence, there is an inherent liability risk – who is responsible for possible errors or failures? Further, there are also risks in terms of who can dominate the ability of artificial intelligence and artificial intelligence to be used. In this case, we do feel that artificial intelligence should be used as a tool (or public service for everyone) and reserve a certain degree of decision-making power to humans to help the system deal with rare accidents.
6. True versatility Artificial intelligence is probably a Collective Intelligence. Strong artificial intelligence is likely not to be a single terminal with powerful decision-making functions, but a collective intelligence. Swarm or Collective Intelligence (Rosenberg, 2015; 2016) can be viewed as "a brain of a group of brains." So far, we have only allowed individuals to provide input values, and then we integrate these post-input inputs in an intelligent way of "average emotion." Rosenberg said that existing methods of achieving human collective intelligence do not even allow users to influence each other. They are usually handled in such a way as to allow only non-synchronous effects to occur - which can lead to group bias. On the other hand, artificial intelligence solves such connectivity flaws and creates a unified collective intelligence that is very similar to other species. A good example of nature comes from bees, which make decisions in much the same way that human nerves work. They all use a large number of executable units that run synchronously, integrate noise, trade off alternatives, and ultimately form specific decisions. Rosenberg believes that this decision is ultimately shaped by real-time closed-loop competition on distributed executable units and subgroups. Each subgroup supports a different choice, and the final consensus is not determined by the public through a similar “average sentiment†approach, but with a “sufficient Quorum of Excitation, Rosenberg, 2015â€. The way is determined. The suppression mechanism for the alternative is generated by other subgroups, which can avoid the overall system from reaching a local optimization decision.
7. Artificial intelligence can have unpredictable socio-political effects. The socio-economic impact of artificial intelligence is the unemployment problem. Although this is a very real problem on the one hand (and of course opportunities in many respects), we believe that this issue should also be viewed in other different ways. First, job opportunities are completely destroyed, but they will become different. Because data will be directly accessible and analyzed by individuals rather than businesses, many services will gradually disappear. Moreover, artificial intelligence tends to decentralize knowledge distribution. What we think should be more concerned in this revolution is the double consequences it brings. First, with smarter (artificial intelligence) systems, more and more people will lose their expertise in specific areas. This indicates that artificial intelligence software needs to be designed to integrate a dual feedback system that integrates human and machine processing. Our second concern is related to the first risk mentioned earlier, and we are worried that humans will become "machine technicians." Because everyone thinks that artificial intelligence is better at solving problems and thinks they are more reliable (so we will deploy artificial intelligence systems more). This vicious circle will make us less creative, less creative, less intelligent, and exponentially increase the difference between man and machine. We are experiencing such a system, or we will be smarter when we use it, or we will feel bad when we don't use it. We want artificial intelligence to be more inclined to become the former, rather than bringing a new "smartphone effect" - we will rely entirely on it. Finally, the world is becoming more and more robot-friendly, and humans are also playing the role of connecting robots. Robots (in society) are gradually playing a leading role, and their impact on humans is getting bigger and bigger than humans, which may cause humans to eventually become "faults" in the social system.
In terms of geopolitics, we believe that artificial intelligence will have a huge impact on globalization. It is possible to have an optimized plant operated by an artificial intelligence system to control the operation of the robot, and its site will eventually return to developed countries. Because (in that case) factories in emerging countries will lose the traditional low-cost reasons. We do not know whether this will balance the differences between countries or whether it will increase the existing differences between developed and developing countries.
8. Real artificial intelligence should start asking "why." Until now, most machine learning systems have done a good job of pattern recognition and decision making; and because most programs are hardcoded, they can still be understood. Although we have been able to let artificial intelligence clarify "what" and "how to do", this is already a good achievement, but artificial intelligence still can't understand the "why" behind things. Therefore, we need to design a general algorithm that can physically and mentally build a model of the nature of the world (Lake et al., 2016).
9. Artificial intelligence is advancing privacy protection issues and data leakage prevention issues. Artificial intelligence takes privacy issues to the next level. New privacy protection methods should be invented and adopted. They should be much more complex than simple secure multiparty computing (SMPC) and should be more efficient and faster than homomorphic encryption. Recent research has shown that the Differential Privacy method can solve most of the privacy issues we encounter on a daily basis. However, many companies have gone further, such as Post-Quantum, a network security startup based on quantum computing.
10. Artificial intelligence is changing the Internet of Things (devices). The development of artificial intelligence allows IoT devices to be designed in a fully distributed architecture where each node is able to make its own budget (the so-called boundary calculation). In the traditional centralized model, there is a problem called the server/client model. Each of these devices is connected to a cloud server and is recognized and verified by the cloud server, which results in very expensive equipment costs. However, the IoT network based on the distributed method or the traditional Peer-to-Peer (P2P) architecture can solve this problem, reduce the cost, and avoid the problem of the whole system being damaged due to the effectiveness of one node.
High-Voltage Motor,Oil Water Pump Motor,High Voltage Dc Motor,High Voltage Electric Motor
Shaoxing AnFu Energy Equipment Co.Ltd , https://www.sxanfu.com