Artificial Intelligence Development Analysis: Is Man-Machine Mixing the Final Stage?

[AI Development Analysis: Is Man-Machine Mixing the Final Stage? 】 One of the important directions of artificial intelligence research is to draw on the achievements of cognitive science and computational neuroscience. This will enable computers to guide themselves to higher levels through intuitive reasoning and experience learning. The current machine intelligence is still computer-centric and does not achieve the "people-centered" that people hope. How to introduce human cognition model into machine intelligence, so that it can reach the level of human intelligence in reasoning, decision-making, memory, etc., is the focus of current scientific discussion.

Will artificial intelligence (AI) replace humans? Stephen Hawking, a well-known physicist in the United Kingdom, and Bill Gates, the richest man in the United States, once gave a positive answer to this question, but scientists in the field of artificial intelligence do not think so.

“As an approach that can lead multiple disciplines and is expected to produce disruptive changes, the effective application of artificial intelligence technology means value creation and competitive advantage. However, there are many fragile and dynamic human society. With regard to open issues, artificial intelligence is still incapable. From this point of view, no intelligent machine can replace human beings.” In an academic salon held in Xi’an on July 31, academician of Chinese Academy of Engineering and director of China Automation Association Long Zheng Ning said that it is necessary to introduce human cognitive ability or human cognition model into the artificial intelligence system to develop new forms of artificial intelligence. This is "mixed intelligence."

"This type of AI or machine intelligence will be a viable and important growth model." Zheng Ning pointed out that intelligent machines and various types of intelligent terminals have become human companions, and the interaction and mixing of people and intelligent machines is the future of society. Development patterns.

Recently, under the unified arrangements and requirements of the Party Central Committee and the State Council, the "New Generation Artificial Intelligence Development Plan" (hereinafter referred to as the "Plan") has made a comprehensive deployment of the artificial intelligence development strategy in China, and clarified the general requirements for the development of artificial intelligence in China and " Three-step strategic goal. Among them, in the five important directions for the "planning" deployment, the "mixed and enhanced intelligence of man-machine collaboration" is impressively on the list.

What is hybrid intelligence?

How to define mixed intelligence? “In the literal sense, hybrid intelligence should be a mixture of different types of intelligence. What kind of intelligence needs to be mixed? Artificial intelligence, human intelligence, and intelligence in the natural world.” Chen Junlong, dean and professor of the University of Macau’s School of Technology, told reporters, Although artificial intelligence is more efficient than humans in the fields of search, calculation, storage, and optimization, its advanced cognitive functions such as perception and reasoning are far from the human brain.

“After all, machine perception and reasoning are still designed by humans.” Chen Junlong said that as the intelligence that nature has endowed with humanity, in the field of brain neuroscience, it is still a long way for people to fully understand the working principle of the human brain.

"The current artificial intelligence system relies on a large number of sample trainings to complete 'supervised learning' at different levels, and true general intelligence will dexterically 'unsupervised learning' based on experience and knowledge accumulation. A simple combination of artificial intelligence computational models or algorithms is impossible to obtain a universal artificial intelligence.” Zheng Nanning explained that, therefore, “the hybrid enhanced intelligence of human-computer collaboration is a typical feature of the new generation of artificial intelligence”.

Zheng Ning pointed out that even though artificial intelligence systems in specific areas such as Google's AlphaGo, IBM's Deep Blue, and Watson rely on powerful computational capabilities to make significant advances in challenging human intelligence, these systems are still unable to achieve a higher level through their own thinking. The intelligence - they still have a gap with general artificial intelligence with a high degree of self-learning ability.

"However, the great success of artificial intelligence in these specific fields has provided us with an important reference and new method for the research and development of new generation artificial intelligence." Zheng Nanning divided the form of mixed intelligence into two basic realization forms: "People Hybrid Enhanced Intelligence in Loops and Hybrid Enhanced Intelligence Based on Cognitive Computing.

Two types of hybrid intelligence

"Man-in-a-loop hybrid enhanced intelligence" is the introduction of people's role into the intelligent system, forming a human-in-loop hybrid intelligence paradigm. In this paradigm, people are always part of this kind of intelligent system. When the output confidence of the computer in the system is low, people actively intervene in adjusting parameters to provide reasonable and correct problem solving, and constitute a feedback loop to improve the level of intelligence.

“Introducing the role of humans into the computational circuits of intelligent systems can tightly couple high-level cognitive mechanisms for analyzing and responding to fuzzy and uncertain problems with machine intelligence systems, making them adapt to each other and work together to form a two-way relationship. Information exchange and control make people's perception and cognitive ability combined with the computer's powerful computing and storage capabilities to form a '1+1>2' intelligent enhanced intelligence form," explains Zheng Nanning.

The "combined enhanced intelligence based on cognitive computing" refers to the introduction of bio-inspired intelligent computing models in artificial intelligence systems, and builds a hybrid enhanced intelligence based on cognitive computing.

"This kind of hybrid intelligence is intelligent software or hardware that enhances the perception, reasoning, and decision-making capabilities of computers by mimicking the function of biological brains to more accurately build intelligent computational models like human brain perception, reasoning, and response incentives, and in particular build cause and effect. New computational frameworks for models, intuitive reasoning, and associative memory.” Zheng Nanning said that for current artificial intelligence, solving certain problems that are intellectually challenging for humans may be relatively simple, but solutions are accustomed to humans. The problem is very difficult.

For example, he said that few three-year-old children can play Go (unless specially trained), but all three-year-old children can recognize their parents, and do not need to be trained in annotated face data sets.

“One of the important directions of artificial intelligence research is to draw on the research results of cognitive science and computational neuroscience to enable computers to guide themselves to higher levels through intuitive reasoning and experience learning.” Xue Jianru, Professor, Institute of Artificial Intelligence and Robotics, Xi'an Jiaotong University The reporter told reporters that the current machine intelligence is still computer-centric and does not achieve the "people-centered" that people hope. How to introduce human cognition model into machine intelligence, so that it can reach the level of human intelligence in reasoning, decision-making, memory, etc., is the focus of current scientific discussion.

Is man-machine mixing the final stage?

The reason why AI failed to meet people's wishes is that the biggest problem is that scientists do not have a unified understanding of the human brain's cognitive model, or the models obtained are all very partial understanding of the brain. But the brain is multi-layered, integrated, and has a variety of organisms that are coupled. At present, there is no way to get a unified, universal architecture.

Based on this, Luo Yan, a researcher at Microsoft Research Asia, believes that "hybrid intelligence may not be a transitional stage in the AI ​​development process, and may be the final stage of AI - we cannot allow the machine to learn a lot and learn In a strong state, AI may end up being a mixed state of man-machine." She turned and said: "There is nothing wrong with this state."

In this regard, the National Thousand Talents Program expert, Hu Binluo, Dean of the School of Information Science and Engineering at Lanzhou University, questioned that he considered the mixed intelligence as the final stage of AI to be “extreme”.

"It is really difficult to define, or should be expressed, that we should not worry that one day the machine will completely replace human beings." Luo Hao replied.

Zheng Nanning added that the development of artificial intelligence will always be accompanied by the evolution of human society. Human evolution will continue to advance the development of artificial intelligence technologies and theories, and the development of artificial intelligence technologies and theories will continue to urge humanity to become more civilized. The times go to progress. “In other words, even if humans have disappeared on Earth, artificial intelligence can also go to very deep places in the universe.”

Gong Yihong, deputy director of the National Engineering Laboratory for Visual Information Processing at Xi'an Jiaotong University, further suggested that the first problem that needs to be solved is the interaction between people and machines. "At present, with the development of technologies such as voice recognition and touch screen, we have made some progress in human-computer interaction, but this is far from enough, and human-computer interaction needs more efficient."

Gong Yihong said that the efficiency of information transfer between people and machines is still very low, and far from achieving real human-machine collaboration and mutual promotion. "The path of information transfer is a key issue of mixed intelligence and must be addressed in the future.