Lead: In recent years, with the development of science and technology, artificial intelligence has made remarkable achievements, but still no breakthrough. Intelligent human-machine integration is the future direction of development of artificial intelligence. In this paper, Reflections on the integration of human intelligence, briefly introduced the origins of artificial intelligence, artificial intelligence proposed future direction of human-computer fusion of intelligence, and intelligent human-machine integration concepts were explained; then analyzes the today intelligent human-machine integration difficulties faced; Finally, the integration of intelligent human-computer exploration methods, namely depth situational awareness, and to try to establish a theoretical framework for intelligent human-machine integration.
First, the origin of artificial intelligence and future direction
Artificial Intelligence (AI) actually originated in Europe, is the original form of philosophy, mathematics manifested form, such as ancient Greek philosophy of "Who am I? ", Laibunizi mathematics "universal rational calculus text +" Wait. 1956 Dartmouth Summer Forum proposed the concept of artificial intelligence (AI) based on the idea of a British mathematician. After six decades of artificial intelligence with machine learning, data mining, development of deep learning technology has made significant progress. During this period of artificial intelligence had three major theoretical ideas were based on neural network on behalf of connectionism, in order to enhance learning as the representative of behaviorism and symbolism of a knowledge map (expert systems), represented by. Recently, the US Department of Defense Agency DARPA advanced technology based on technical analysis to determine the characteristics of AI technology development stage, that the AI has experienced the first wave and the second wave, will usher in the Third Wave: the first wave of AI technology wave begins in the early 1960s, with "hand-knowledge" is characterized by the establishment of a set of logic rules to represent a particular field of knowledge, reasoning for the rigorous definition of the problem, there is no ability to learn, the ability to deal with uncertainty is very weak. The second wave of AI technology wave began in the late 1960s, the "statistical learning" is characterized by the establishment of a statistical model for a specific problem domain, they use big data to their training, have a very low degree of reasoning ability, but not the ability to have context. The third wave of the wave of AI technology to "adapt to the environment" (context-adaptive) is characterized by sustainable learning and interpreted, create a system capable of generating explanatory models for real-world phenomenon can be natural communication between man and machine, system in the face of new situation and tasks can learn and reasoning. AI continued self-learning ability will be the third wave of AI core power technology wave [9].On this basis, we go through thinking and analysis, that the fourth wave of AI technology wave will be "actively adapt to the environment" (Context-Adaptive wider range) is characterized by having a sustainable learning + learning unsustainable and can not be interpreted + explanation for the phenomenon of real + virtual world model system capable of generating a modest initiative explanatory, can be a natural depth exchanges between people and machines, the system in the face of new situation and tasks to achieve human-computer interaction and mutual learning reasoning. Man-machine fusion mutual learning initiative mutual understanding of each other mutual assistance ...... fusion capability will be the fourth wave power core AI technology wave.
Objectively speaking, only human artificial intelligence can be described of a part of a programmable, human and human intelligence, machine (product), the product of the interaction of environmental systems. Intelligent generation mechanism, in short, is the result superposition character (machine is a man-made) environment system interactions, by the people, machines, a variety of environmental change in the state superimposed derived situation, situation and trends (referred to as potential ) together constitute, three changes in the state of good and bad, high and low, there is an inverse cis, reflecting intelligence generation is, machines, environmental systems, concordance resonance extent determined by the size of the potential of people, the three between those who have constructive and destructive interference effects, or to enhance or eliminate, the Trinity is intelligent and strong, over three weak body is smart. How common frequency tuning is the key to integration of human intelligence. Contemporary artificial intelligence developed from the initial automated machine completely artificial compiled into a manual precompiled machine learning, the next development could be a smart fusion is achieved through the human machine cognition, ultimately awakening machine.
Second, intelligent human-machine integration difficulties faced by
" Smart " concept would imply the individual, limited overall, infinite relationship. For smart times come, it was suggested that " need to be considered from a totally different perspective and understanding of the principles of spatial and temporal behavior existed since ancient times ," such as traditional people, goods, environmental relations. When people some intelligence activities, usually carried out in accordance with changes in the external environment, the key point or critical point correction or adjustment, through local and global short, medium and long-term optimization of expectations, real assigning weights to various data information and knowledge processing, and more more programmatic + non-programmed mixing process. The machine intelligence is difficult to realize such a random mixture of resilience, deterministic programming mark more prominent, like " Alpha Dog " ( AlphaGo ) / Yuan / star of such relatively good intelligence system, a clear calculation major win at the border the speed and accuracy, the relative open place game or confrontation in the environment did not show so well in a closed environment, even very bad. True intelligence is not only adaptable, more importantly, is not adaptable, and then create a new possibility, intelligence is probably not simple to adapt, adapt, more importantly, do not conform, adapt, thereby creating a series of new possibilities: free, assimilation, Feng Fu, change, independence. Turing machine only drawback is that stimulus - reaction no choice but to comply without assimilation mechanism.
The world is constituted by a contact or by the property constituted? This is a question worth considering. It should be by both together constitute it! "Tao Te Ching" said The fortieth chapter in " dynamic Counter -;. Road with weak world all things are born, not born " , this sentence is the concentrated expression of this idea, here it comes . " Countering, " with respect to " positive person " , but also " round-trip " meaning; and " weak " relative " strong " , with those who have anti-positive person, it is called yin and yang. Counter - moving, with a very particular here to illustrate words " road " is " yin and yang of that road " . The weak, the strong are the yin and yang. With the yin and yang Road to move, we have interaction.
Information technology is essentially a calculation fact, intelligence is perceived value. Information to the data from the knowledge (structure) cognitive calculation, the information from the knowledge data (deconstruction) is calculated cognition. If the smart as language, artificial intelligence, such as grammar, human intelligence is more like semantics, pragmatics. Grammar rule-based, statistics and probability, and semantic pragmatics is between one kind of people based on the use meaningful elements of the agreement, the subconscious is more prescriptive than the cross-border grammar, flexible, and yet people present its law an effective cognitive rules, so it became the complexity of things. Symbolic characterize normative grammar, semantic context of the natural basis. A group of border territory and a reduced component, but also novel elements to understand one of the difficulties is smart inside and outside the coexistence of more than one intertwined interference and influence. The anywhere, anytime, any information given to people at the right time into, place, manner of information to the hands of the right people is one of the manifestations of intelligence. Global, man machine dimension L, the dimensionality reduction is a human machine; local, and vice versa. Because the global relates to heterogeneous things, non-familial similarity; while the opposite is local. For human intelligence systems, the role of Go is just the local part.
Artificial intelligence bottom diode is 0 , 1 binary logistic, most human intelligence underlying technology polyhydric human intent (non-logical). Art is human intelligence, artificial intelligence can mainly technical. Artificial intelligence is a tool that many people take it as a universal key, but some people imagine it has become all-powerful Monkey King and Santa Claus, while ignoring the role of human wisdom. Human intelligence is a process involving emotional (especially brave) more intellectual energy in the rapidly changing emergency situation, a person dominated by emotions rather than thinking, and thus need to arouse intellectual qualities of courage, then support and maintain the necessary in action reason, in human intelligence, we often see an orderly / creative tension between the disorder, as in many situations, the same thing you see (such as apples or 1 hour) are often different, take the initiative to look at, passive look, semi-active look different. Prejudice often easy to form of artificial intelligence, knowledge extracted from the map rule out a priori and common sense, and introduced it as a model to generate constraints, may make the program run compromised intelligent, so how the human perception of blur, accurate recognition and machine perception, identifying the problem will be combined with a very worth considering.
1, human cognitive inconsistency
Mainly due to the integration of human intelligence is that time and space are difficult and cognitive inconsistency. Information and knowledge people can deal with variation, a thing which is characterized, not only in itself but it is also the fact that other things, the fact has been relative, while data processing machine identifies the relative lack of such variability. More importantly, the people of the time, spatial cognition is to have intentionality, is a subjective expectation of ( Should ), and machine perception of time and space is biased towards formal, is an objective reality ( being ). The two are not on the same dimension, it has a strong inconsistency. Human cognition is focused on the psychological level, is subjective, and the perception of the machine is biased in favor of the physical level, it is the objective. In the cognitive aspects of human learning, reasoning and judgment adaptable, when political reform has also changed, Incident law also changed, and machine learning, reasoning and judgment are specific mechanisms for the designers developed or chosen for a particular space-time tasks, and current space-time job in the user intent is often not entirely consistent, poor variability. This inconsistency both inconsistency subjective expectations and objective data feedback of the machine, including the expected inconsistencies subjective and the objective facts of people.
Many things seemingly illogical questions, such as David and Goliath in many cases, in fact, in essence is a logical question, which is relatively weak in the David and Goliath in the local but often with a strong victory weak, It contains many illogical logic. Also, a lot of logical problems also exist in non-logical problems, such as the number of cases is not a chapter ko, ko on the surface, in fact, the reason is change management, the truth is not complete, is a prerequisite boundary condition constraints, when many of these boundary conditions constraint premise be some minor changes, naturally, can not become a chapter. Thus, logic and non-logic coexist in among things, but also the root causes of order and disorder, in which the organization is the human-machine interaction with smart fusion research priorities, but also intelligent human-machine integration difficulties.
Another key issue is the axiom man-machine fusion and non-fusion axiom hybrid reasoning, intuition and rational combination of decisions. Axiom is the theoretical basis for the development of the history of mathematics, and science in the process of logical deduction is the most core method. Similarly, the process is still running computer in accordance with strict algorithmic language run. But this is different from the human decision-making process, the human ability to think also depends on the reasoning by analogy. Part of the reasoning analogical reasoning non axiom, non axiom reasoning determines the strong perception problem in the case of a weak trend. This learning method relies on prior knowledge, through the use of big data and probability methods to achieve, to achieve the machine's non axiom reasoning is one difference between man and machine, but also an important way to achieve human emotion on the machine. Produced by prior knowledge of human intuition, and rational analysis is the antithesis of intuition. In the data processing machine always rationally, and how to make the machine produce intuitive ability is the key to the smoothness of the man-machine integration. Axiom and non axiomatic reasoning, intuition and rational decision-making will be combined to address important research integration of intelligent man-machine output [1] .
2, intentionality and formalization
British computer scientists, artificial intelligence philosopher Margaret Tang Bo, she made very early artificial intelligence core and bottlenecks that intentionality and combine formal, yet today there is a breakthrough, in fact, this is the man the difficulty fusion intelligence. In the current investment in human-computer application integration product, a clear division of labor between man and machine, but not organically integrated. Human beings can make better predictions on the development trend of information in the environment, resources are not complete, this is because human beings acquired learning can continue to enhance their cognitive ability. The machine does not have the ability to think, and mankind is precisely the ability to cross-cutting combination can be produced by association. So how to make the association machine capability is key to achieving real intelligence.
意向性是对内在的感知的描述（心理过程、目的、期望），形式化是对外在的感知的描述（物理机理、反馈）。人机融合智能及深度态势感知就是意向性与形式化的综合。形式化更多的是倾向于让人们对事物有一个直观的空间上的认知，而把这种空间上的认知延伸到时间上描述，就是意向性。形式化是态，那么意向性就是势。人机融合就是要形成一个对内在外在、主观客观、认知与行为上的感知的整体描述，形成一个可以描述人的心理过程、目的、期望以及机器的物理机理、反馈的模型。
当前智能领域面临的困难是人的意向性与行为的差异程度，行为可以客观形式化，而意向性是主观隐性化的，一个智能系统想要形成和存在，其内部的构件在本性或运行规律上就必须拥有既相互吸引又相互排斥、既靠拢又闪避、既结合又分离、既统合又脱节的能力。人机融合智能中意向性是联结事实与价值的桥梁，形式化可以某种程度地实现这种意向性。
3、休谟之问的伦理问题
人机融合智能的最后一个关键问题是伦理问题。人类价值观的起源是伦理学。人类本身拥有很多伦理道德困境，人工智能的出现也带给了人类对待人工智能伦理问题的思考。与此同时，人机融合只能伦理问题的关键之一是人机融合智能的范畴归属。人机融合智能的伦理问题包括人工智能的伦理以及人机融合后的责任归属，这也是人机融合智能在接下来发展的问题之重。
“休谟问题”说的是从事实推不出价值来，可是，这个世界却是一个事实与价值混合的世界，不知从价值能推出事实吗？汉字就是智能的集中体现，有形有意，如日月人，一目了然；西方的文字常常无形无意，逻辑类推。智能的本质就是把意向性与形式化统一起来， 所以汉字从象形到会意的过程就是人类自然智能的发展简史……汉字的偏旁部首就是一种类的封装，把强相关的字聚在一起。如果说人类造字是语言表征的封装积累，那么，人类造智则是思想意识的拓扑延展。 智能不是百科全书，而是包含不少的虚构和想象，不仅是分类，还要合类，不仅合并同类项，而且要合并异类项，因而，智能产品系统的顶层设计非常重要。 人工智能一般是逻辑（家族相似性）关系，人类智慧常常是非逻辑（非家族相似性）的。未来的智能是在特定环境下人的智能与机器智能的融合，即人机融合智能。人机融合智能不是人工智能，更不是机器学习算法。同样，人工智能、机器学习算法也不是人机融合智能，人机融合智能是人机环境的相互融合，是《易经》中的知几（看到苗头）、趣时（抓住时机）、变通（随机应）。人机融合智能是随动，不是既定，其中的“知己知彼”中的“知”不是简单的态势“感知”，更是态势“认知”。认知是从势到态的过程，感知是从态到势的过程。认知侧重认，信息输入处理输出流动过程；感知侧重感，数据信息的 输入过滤过程，认知涉及先验和经验等过去的感知，所以态势认知包 括了以前的态势感知。人工智能是一把双刃剑，计算越精细准确，危险越大，因为坏人可以隐真示假，进行欺骗，所以人机有机融合的智能更重要。客观而言，当前的人工智能基本上就是自动化+ 统计概率，简单地说，归纳演绎的缺点就是用不完备性解释完备性。
毕加索曾透露：“绘画不是一个美学过程，而是……一种魔法， 一种获取权力的方式，它凌驾于我们的恐惧与欲望之上。”看懂了毕 加索的作品，就能理解毕加索想要表达的“魔法”，并且把它化用到 生活中的其他领域，尤其是智能领域和人机融合智能领域。
需要注意的是，休谟问题至今尚未真正得到解决。正因为“价值”是相对的，因人而异的，所以这一问题也永远不可能真正得到解决，这一点已经在上一节中作出了论述。唯物主义者虽然想把唯物主义贯彻到精神领域，但这是永远也不可能做到的。因为精神和物质，在本质上是完全不同的东西，一个是主观，一个是客观。就如同怀疑论者经常使用的“桶中脑实验”（英国哲学家普南提出，有的版本也翻译为“缸中脑”）描述的那样，我们对于这个世界的认识，其实完全只是一种主观的判断，这种判断和真实的“客观世界”是否一致，我们永远也不可能知道。虽然某些唯物主义者总喜欢用“无数次的实践”来证明主观与客观理论上最终能达到这种一致性，但实际上，“无数次的实践”是不可能做到的。所以说这只是一种空想罢了。
三、人机融合智能的难点：深度态势感知研究
态势感知的定义不在此做赘述。态势感知（situation awareness）一词最早于第一次世界大战中提出，之后在心理学领域中作为“情境意识“被广泛应用，直到1988年Endsley对态势感知的重新定义，以及其在1995年提出的著名的态势感知三级模型[3]，标志着将态势感知迁移到了工程学领域中，再到2003年Wickens提出的基于注意力的态势感知模型（A-SA模型，Attention-SituationAwareness Model）[4]以及Hooey于2010年将态势元素（Situation Element）[5]引入态势感知研究中，标志着态势感知研究由主观数据驱动到客观数据驱动，由定性分析到定量分析的过渡。近年来，随着人工智能相关技术的迅猛发展，网络态势感知（CyberSituation Awareness）成为了网络安全领域的研究热点。态势感知似乎成为了一种研究方法（method），而不是一个可以指导人们认识世界、改造世界的方法论（Methodology）。当前的态势感知理论技术仍然存在很多不足，主要是未将人的心理活动过程与机器的外在表现形式以及环境中的态势要素有机地结合。鉴于此，本文尝试提出了深度态势感知这个概念，具体说明如下：
深度态势感知的含义是“对态势感知的感知，是一种人机智慧，既包括了人的智慧，也融合了机器的智能（人工智能）”，是能指+所指，既涉及事物的属性（能指、感觉）又关联它们之间的关系（所指、知觉），既能够理解弦外之音，也能够明白言外之意。它是在Endsley以主体态势感知（包括信息输入、处理、输出环节）的基础上，对包括人、机（物）、环境（自然、社会）及其相互关系的整体系统趋势分析，具有“软/硬”两种调节反馈机制；既包括自组织、自适应，也包括他组织、互适应；既包括局部的定量计算预测，也包括全局的定性算计评估，是一种具有自主、自动弥聚效应的信息修正、补偿的期望—选择—预测—控制体系。关于深度态势感知的详细解释请参见一文 [7] 。
在维纳出版的著作《控制论——关于在动物和机器中控制和通讯的科学》中，维纳将控制论看作是对机器、生命以及社会的规律进行研究的科学，是研究个体(可能是生物，也可能是机器)在动态环境中怎样保持稳态的过程的科学，控制论的思想和方法对社会科学与自然科学领域的研究产生了深远的影响[8]。在《控制论》一书中，维纳提出“控制的核心是反馈，反馈是人们的目的性行为”。然而，控制论在揭示机器的自然存在时不仅完全屏蔽了社会巨型机——它本身不过是其中的一个时段和一个成分，而且还完全屏蔽了组织生成性这个关键问题，而生成性则是除人造机之外一切物理、生物和社会机器所固有的禀性。
事实上，把生命体特有的“目的性行为”概念用“反馈”这种概念代替，把按照反馈原理设计成的机器的工作行为看成为目的性行为，并未突破生命体（人）与非生命体（机器）之间的概念隔阂。原因很简单，人的“目的性行为”分为简单显性和复杂隐性两种，简单显性的“目的性行为”可以与非生命体机器的“反馈”近似等价（刺激—反应），但复杂隐性的“目的性行为”——意向性却远远不能用“反馈”近似替代，因为这种意向性可以延时、增减、弥聚，用“反思”定义比较准确，但“反思”概念却很难用非生命体的机器赋予（刺激—选择—反应）。“反思”的目的性可用主观的价值性表征，这将成为人机融合的又一关键之处。价值将由吸引子和动机共同构成。反思是一种非生产性的反馈，或者说是一种有组织性的反馈。自主是有组织的适应性，或被组织的适应性。据此我们将Endsley态势感知三级模型和维纳的“反馈”思想结合，提出了一个基于“反馈”的深度态势感知模型
图1：基于“反馈”的深度态势感知模型
深度态势感知理论模型在不同情境下处理信息的方式会有所区别，并且以往关于态势感知的研究都充分说明了态势感知具有实时性，即态势感知会随时间而不停地更新、迭代。所以我们尝试着对态势感知进行细化，并提出了一个基于循环神经网络（RNN）的深度态势感知理论框架，如下图所示：
图2：基于RNN的深度态势感知理论框架
我们将态势感知中的“态”定义为人机环境系统中的各类表征个体状态的主客观数据，即state；“势”定义为事件的发展趋势，即trend；“感”定义为对系统中“态”的觉察，即sense；“知”定义为对“势”的理解。该理论框架就是为了辅助人们更好地“感态”、“知势”。而为了获取数据必然要引入客观数据，根据之前的研究，我们可以将态形式化为显著性，势形式化为价值性，感为反应时，知为准确率。感态着重于时效性，而知势更倾向于有效性。
“我思故我在”，这是笛卡尔二元认知论的起点，也是终点，即唯一确定的事，就是“我”的体验。根据认知科学的解释，由于在大多数情况下人的认知能力是有限的，所以最优化是无法实现的。参与人还必须了解他的目标方程，这就要求另一个庞大的认知性先决条件，如同参与人发现他们所处的环境一样，系统地描述这一目标方程是极其复杂的。知己知彼不可分，不知彼就不能知己，任何事物本身不能解释自己，只有从其它参照物处才能感知、理解、发现、说明、定义自己（我是谁，我从哪里来，我要去哪），进而可以认为：自我是不存在的，没有环境和参照物，自己解释不了自己，如同“我”的概念定义不能为“我就是我”一样。再进一步，自我意识也可能是不存在的，它也是交互的产物，只不过可以穿越时空逻辑关系罢了。实际上，所有的自主系统都是不由自主，只不过显隐程度不同而已。之后笛卡尔将自己的哲学观点形式化为著名的二元直角坐标系。
鉴于笛卡尔的观点，深度态势感知虚实参照系可分为人机不同的态（事物）参照系、势（事实）参照系、感（显著）参照系、知（价值）参照系，当这些虚实参照系大部分一致，亦或没有本质的矛盾时，才有可能产生正确的觉察和决策行为。
只有在把一物与它物区分开来，才会对该物有了认知。只有把一个人的知识或信仰状态与他人的区分开来，才会对一个人有了解。哲学上最难，也是最重要的任务之一，就是明确世界的两类特征，即那些独立于任何观察者而存在的内在特征和那些相对于观察者或使用者而存在的外在特征。例如一个物体有质量（无论对谁而言）与这个物体是浴缸（也可是水缸、饰缸、粮缸）。所以对深度态势感知系统研究的下一步工作，就是将其具体应用到某一或某些情境中，检验其有效性和可靠性。
四、结束语
也许很多人看过《黑客帝国》这部有关人工智能的科幻电影，其英文名称为“Martrix”，即“矩阵”。的确，现在的人工智能相关技术（大数据、机器学习、深度学习等）是以矩阵论、概率论等数学理论为基础发展而来的，并且为人们生产生活提供了便利，甚至一定程度上带动了社会变革。人工智能所取得的成就得益于自17世纪以来400年间人们对数学的不懈追求，但现今人工智能所忽视的，也可能是帮助人们突破当今人工智能瓶颈的，恰恰是几千年来对人们对世界的认知以及对自我反思的研究。所以如何将自然科学与社会科学有机地结合，是下一代人工智能技术的研究重点。
But the presence of man-made machines have no self. Self was born in regular interaction, organization and produce their own existence. Produced no initiative interaction and organization, it is not independent, there is no self, no ego, there can not be a sense of the feeling he had, confidant and know the enemy, you can not contact inductive rational, objective and subjective can not be formed, facts It can not be derived value. Intelligent, although the system is a complex issue and involves a very wide, in essence, is still subjective and objective, emotional and rational, intention and formal unity of opposites (human environment) systems only. Its core values still can not be separated breakthroughs in basic theory, rather than data, algorithms, and calculate the force experiments. Human-computer integration, not just create more advanced machine to design better algorithms, how much data is obtained, but the person's own intellectual transformation of transformation, that is the logical thinking of remodeling and change.
In the vastness of the universe, human beings have, however small drop in the ocean, like a huge element in the matrix. AI limit future where intelligent human-machine integration whether to make the machine break through the bottleneck of self-awareness, everything is unknown. "It's too early to tell" , to conclusions too early.
References : Slightly
This article was published in " AI Magazine" 2019.8