数字化转型网人工智能专题
与全球关注人工智能的顶尖精英一起学习！数字化转型网建立了一个专门讨论人工智能技术、产业、学术的研究学习社区，与各位研习社同学一起成长！欢迎扫码加入！
本文为数字化转型网(szhzxw.cn）为人工智能/AI专题系列文章之一，本文讲述了人工智能的应用领域、强人工智能和弱人工智能。
一、人工智能的应用领域
1. 问题求解
人工智能的第一大成就是下棋程序，在下棋程度中应用的某些技术，如向前看几步，把困难的问题分解成一些较容易的子问题，发展成为搜索和问题归纳这样的人工智能基本技术。今天的计算机程序已能够达到下各种方盘棋和国际象棋的锦标赛水平。但是，尚未解决包括人类棋手具有的但尚不能明确表达的能力。如国际象棋大师们洞察棋局的能力。另一个问题是涉及问题的原概念，在人工智能中叫问题表示的选择，人们常能找到某种思考问题的方法，从而使求解变易而解决该问题。到目前为止，人工智能程序已能知道如何考虑它们要解决的问题，即搜索解答空间，寻找较优解答。
2. 逻辑推理与定理证明
逻辑推理是人工智能研究中最持久的领域之一，其中特别重要的是要找到一些方法，只把注意力集中在一个大型的数据库中的有关事实上，留意可信的证明，并在出现新信息时适时修正这些证明。对数学中臆测的题。定理寻找一个证明或反证，不仅需要有根据假设进行演绎的能力，而且许多非形式的工作，包括医疗诊断和信息检索都可以和定理证明问题一样加以形式化，因此，在人工智能方法的研究中定理证明是一个极其重要的论题。 数字化转型网www.szhzxw.cn
3、自然语言处理
自然语言的处理是人工智能技术应用于实际领域的典型范例，经过多年艰苦努力，这一领域已获得了大量令人注目的成果。目前该领域的主要课题是：计算机系统如何以主题和对话情境为基础，注重大量的常识——世界知识和期望作用，生成和理解自然语言。这是一个极其复杂的编码和解码问题。
4、智能信息检索技术
信息获取和精化技术已成为当代计算机科学与技术研究中迫切需要研究的课题，将人工智能技术应用于这一领域的研究是人工智能走向广泛实际应用的契机与突破口。
5、专家系统
专家系统是目前人工智能中最活跃、最有成效的一个研究领域,它是一种具有特定领域内大量知识与经验的程序系统。近年来，在“ 专家系统”或“ 知识工程”的研究中已出现了成功和有效应用人工智能技术的趋势。人类专家由于具有丰富的知识，所以才能达到优异的解决问题的能力。那么计算机程序如果能体现和应用这些知识，也应该能解决人类专家所解决的问题，而且能帮助人类专家发现推理过程中出现的差错，现在这一点已被证实。如在矿物勘测、化学分析、规划和医学诊断方面，专家系统已经达到了人类专家的水平。成功的例子如：PROSPECTOR系统（用于地质学的专家系统）发现了一个钼矿沉积，价值超过1亿美元。DENDRL系统的性能已超过一般专家的水平，可供数百人在化学结构分析方面的使用。MY CIN系统可以对血液传染病的诊断治疗方案提供咨询意见。经正式鉴定结果，对患有细菌血液病、脑膜炎方面的诊断和提供治疗方案已超过了这方面的专家。 数字化转型网www.szhzxw.cn
二、强人工智能和弱人工智能
人工智能的一个比较流行的定义，也是该领域较早的定义，是由当时麻省理工学院的约翰·麦卡锡在1956年的 达特矛斯会议上提出的：人工智能就是要让机器的行为看起来就像是人所表现出的智能行为一样。但是这个定义似乎忽略了强人工智能的可能性（见下）。另一个定义指人工智能是人造机器所表现出来的智能。总体来讲，目前对人工智能的定义大多可划分为四类，即机器“像人一样思考”、“像人一样行动”、“理性地思考”和“理性地行动”。这里“行动”应广义地理解为采取行动，或制定行动的决策，而不是肢体动作。
1. 强人工智能（AGI）
强人工智能观点认为有可能制造出真正能推理（Reasoning）和解决问题（解决问题）的智能机器，并且，这样的机器能将被认为是有知觉的，有自我意识的。强人工智能可以有两类：
- 类人的人工智能，即机器的思考和推理就像人的思维一样。
- 非类人的人工智能，即机器产生了和人完全不一样的知觉和意识，使用和人完全不一样的推理方式。
2. 弱人工智能（ANI）
弱人工智能又称“狭义人工智能”，认为不可能制造出能真正地推理和解决问题的智能机器，这些机器只不过看起来像是智能的，但是并不真正拥有智能，也不会有自主意识。
强人工智能的研究目前处于停滞不前的状态下。人工智能研究者不一定同意弱人工智能，也不一定在乎或者了解强人工智能和弱人工智能的内容与差别。就现下的人工智能研究领域来看，研究者已大量造出看起来像是智能的机器，获取相当丰硕的理论上和实质上的成果，如2009年康乃尔大学教授Hod Lipson 和其博士研究生Michael Schmidt 研发出的 Eureqa计算机程序，只要给予一些数据，这计算机程序自己只用几十个小时计算就推论出牛顿花费多年研究才发现的牛顿力学公式，等于只用几十个小时就自己重新发现牛顿力学公式，这计算机程序也能用来研究很多其他领域的科学问题上。
3. 超级人工智能（ASI）
超级人工智能（ASI）是一种超越人类智能的人工智能，可以比人类更好地执行任何任务。ASI 系统不仅能理解人类的情感和经历，还能唤起他们自己的情感、信念和欲望，类似于人类。 尽管 ASI 的存在仍是假设性的，但此类系统的决策和解决问题的能力预计将远超人类。通常，ASI 系统可以独立思考、解决难题、做出判断和做出决定。 数字化转型网www.szhzxw.cn
4. 对强人工智能的哲学争论
“强人工智能”一词最初是约翰·罗杰斯·希尔勒针对计算机和其它信息处理机器创造的，其定义为：
“强人工智能观点认为计算机不仅是用来研究人的思维的一种工具；相反，只要运行适当的程序，计算机本身就是有思维的。”（J Searle in Minds Brains and Programs. The Behavioral and Brain Sciences, vol. 3, 1980） 数字化转型网www.szhzxw.cn
关于强人工智能的争论，不同于更广义的一元论和二元论的争论。其争论要点是：如果一台机器的唯一工作原理就是转换编码数据，那么这台机器是不是有思维的？希尔勒认为这是不可能的。他举了个中文房间的例子来说明，如果机器仅仅是转换数据，而数据本身是对某些事情的一种编码表现，那么在不理解这一编码和这实际事情之间的对应关系的前提下，机器不可能对其处理的数据有任何理解。基于这一论点，希尔勒认为即使有机器通过了图灵测试，也不一定说明机器就真的像人一样有思维和意识。
也有哲学家持不同的观点。丹尼尔·丹尼特在其著作《意识的解释》（Consciousness Explained）里认为，人也不过是一台有灵魂的机器而已，为什么我们认为：“人可以有智能，而普通机器就不能”呢？他认为像上述的数据转换机器是有可能有思维和意识的。 有的哲学家认为如果弱人工智能是可实现的，那么强人工智能也是可实现的。比如西蒙·布莱克本（Simon Blackburn）在其哲学入门教材Think里说道，一个人的看起来是“智能”的行动并不能真正说明这个人就真的是智能的。我永远不可能知道另一个人是否真的像我一样是智能的，还是说她／他仅仅是看起来是智能的。基于这个论点，既然弱人工智能认为可以令机器看起来像是智能的，那就不能完全否定这机器是真的有智能的。布莱克本认为这是一个主观认定的问题。
需要指出的是，弱人工智能并非和强人工智能完全对立，也就是说，即使强人工智能是可能的，弱人工智能仍然是有意义的。至少，今日的计算机能做的事，像算术运算等，在一百多年前是被认为很需要智能的。并且，即使强人工智能被证明为可能的，也不代表强人工智能必定能被研制出来。
数字化转型网人工智能专题
与全球关注人工智能的顶尖精英一起学习！数字化转型网建立了一个专门讨论人工智能技术、产业、学术的研究学习社区，与各位研习社同学一起成长！欢迎扫码加入！ 数字化转型网www.szhzxw.cn
人工智能/AI专题栏目文章（二）：人工智能的应用领域、强人工智能和弱人工智能。翻译：
Application fields of artificial intelligence, strong artificial intelligence and weak artificial intelligence
First, application fields of artificial intelligence
- Problem solving
The first major achievement of artificial intelligence is the chess program, and some technologies applied in the level of chess, such as looking forward a few steps, breaking down difficult problems into some easier sub-problems, and developing into the basic technology of artificial intelligence such as search and problem induction. Today’s computer programs are capable of playing all kinds of square and chess tournaments. However, it has not yet been resolved to include abilities that human chess players possess but cannot yet articulate. Like the ability of chess grandmasters to see the game. Another problem involves the original concept of the problem, in artificial intelligence called the choice of problem representation, people can often find some way to think about the problem, so that the solution is easy to solve the problem. By now, AI programs know how to think about the problem they’re trying to solve, by searching the solution space for the best solution.
- Logical reasoning and theorem proving
Logical reasoning is one of the most persistent areas of AI research, and it is particularly important to find ways to focus only on facts in a large database, keep an eye out for credible proofs, and correct those proofs as new information emerges. A speculative problem in mathematics. Finding a proof or disproof of a theorem not only requires the ability to deduce from assumptions, but also many informal tasks, including medical diagnosis and information retrieval, can be formalized as well as theorem-proving problems. Therefore, theorem-proving is an extremely important topic in the study of artificial intelligence methods. 数字化转型网www.szhzxw.cn
- Natural language processing
The processing of natural language is a typical example of the application of artificial intelligence technology to the practical field. After years of hard work, a lot of remarkable results have been obtained in this field. The main topic in this field is how computer systems can generate and understand natural language based on themes and conversational situations, focusing on a large amount of common sense – world knowledge and expectations. This is an extremely complex coding and decoding problem.
- Intelligent information retrieval technology
Information acquisition and refinement technology has become an urgent research topic in contemporary computer science and technology research. Applying artificial intelligence technology to this field is an opportunity and breakthrough for the wide practical application of artificial intelligence.
- Expert system
Expert system is one of the most active and effective research fields in artificial intelligence. It is a kind of program system with a lot of knowledge and experience in a specific field. In recent years, there has been a trend of successful and effective application of artificial intelligence technology in the research of “expert systems” or “knowledge engineering”. Human experts have a wealth of knowledge, so they can achieve excellent problem-solving skills. Computer programs, if they embody and apply this knowledge, should be able to solve the same problems that human experts solve, and to help human experts spot errors in their reasoning, as has now been shown. For example, in mineral exploration, chemical analysis, planning and medical diagnosis, expert systems have reached the level of human experts. Successful examples include the PROSPECTOR system (an expert system used in geology), which discovered a molybdenum deposit valued at more than $100 million. The performance of the DENDRL system has exceeded the level of the average expert and can be used by hundreds of people in the analysis of chemical structures. The MY CIN system can provide advice on the diagnosis and treatment of blood-borne infectious diseases. As a result of the formal identification, the diagnosis and treatment of bacterial blood diseases, meningitis, has exceeded that of specialists in this field.
Second, strong artificial intelligence and weak artificial intelligence
One of the more popular definitions of artificial intelligence, and one of the earlier definitions of the field, was proposed by John McCarthy of the Massachusetts Institute of Technology at the Dartmouth Conference in 1956: Artificial intelligence is to make machines behave as if they were intelligent as humans. But this definition seems to ignore the possibility of strong AI (see below). Another definition refers to artificial intelligence as the intelligence shown by artificial machines. In general, most of the current definitions of artificial intelligence can be divided into four categories, that is, machines “think like a human”, “act like a human”, “think rationally” and “act rationally”. Here “action” should be understood broadly as taking action, or making a decision to take action, rather than a physical action.
- Strong Artificial Intelligence (AGI)
The strong AI view holds that it is possible to build intelligent machines that can truly reason and solve problems (problem solving), and that such machines can be perceived as sentient and self-aware. There are two types of strong AI: 数字化转型网www.szhzxw.cn
Human-like artificial intelligence, in which machines think and reason just like people do.
Non-humanoid artificial intelligence, in which machines generate perceptions and consciousness that are completely different from those of humans, and use reasoning that is completely different from those of humans. 数字化转型网www.szhzxw.cn
- Weak Artificial Intelligence (ANI)
Weak artificial intelligence, also known as “narrow artificial intelligence,” believes that it is impossible to build intelligent machines that can truly reason and solve problems, but these machines only look intelligent, but do not really have intelligence, and will not have autonomous consciousness.
The research of strong artificial intelligence is currently in a state of stagnation. Ai researchers do not necessarily agree with weak AI, nor do they necessarily care about or understand the content and differences between strong AI and weak AI. In the current field of artificial intelligence research, researchers have built machines that appear to be intelligent, with considerable theoretical and practical results, such as the Eureqa computer program developed by Cornell University professor Hod Lipson and his doctoral candidate Michael Schmidt in 2009. Given some data, the computer program itself, in only a few dozen hours of calculation, deduced the formula of Newtonian mechanics that Newton had spent many years of research to discover, equivalent to rediscovering the formula of Newtonian mechanics in only a few dozen hours, and the computer program can be used to study many other fields of science.
- Super Artificial Intelligence (ASI)
Superartificial intelligence (ASI) is an artificial intelligence that exceeds human intelligence and can perform any task better than humans. ASI systems not only understand human emotions and experiences, but also evoke their own emotions, beliefs, and desires, similar to humans. Although the existence of ASI is still hypothetical, the decision-making and problem-solving capabilities of such systems are expected to be far superior to those of humans. Typically, ASI systems can think independently, solve problems, make judgments, and make decisions. 数字化转型网www.szhzxw.cn
- Philosophical debate over strong AI
The term “strong artificial intelligence” was first coined by John Rogers Hiller for computers and other information processing machines, and is defined as:
“The strong AI view is that computers are not just a tool to study the human mind; On the contrary, given the proper program, the computer itself has a mind.” (J Searle in Minds Brains and Programs. The Behavioral and Brain Sciences, vol. 3, 1980)
The debate about strong AI is different from the broader monistic and dualist debate. The argument goes: If a machine’s only job is to convert encoded data, does that machine have a mind? Hiller doesn’t think that’s possible. He gave the example of a Chinese room to show that if the machine is merely transforming data, and the data itself is a coded representation of something, then the machine cannot have any understanding of the data it is processing without understanding the correspondence between the code and the actual thing. Based on this argument, Hiller argued that even if a machine passes the Turing Test, it does not necessarily mean that the machine really has a mind and consciousness like a human. 数字化转型网www.szhzxw.cn
There are philosophers who take a different view. Daniel Dennett, in his book Consciousness Explained, argues that man is nothing more than a machine with a soul, so why do we think that “man can have intelligence and ordinary machines cannot?” He believes that it is possible for a data conversion machine like the one described above to have a mind and consciousness. Some philosophers believe that if weak AI is achievable, then strong AI is also achievable. For example, Simon Blackburn, in his introductory philosophy textbook Think, argues that a person’s actions that appear to be “intelligent” do not really mean that the person is intelligent. I can never know if another person is really intelligent like me, or if she/he just seems intelligent. Based on this argument, since weak AI thinks it can make a machine appear intelligent, it cannot be completely denied that the machine is really intelligent. Blackburn thinks this is a matter of subjective determination.
It should be pointed out that weak AI is not completely opposed to strong AI, that is, even if strong AI is possible, weak AI still makes sense. At the very least, today’s computers can do things, like arithmetic, that were thought to require intelligence more than a century ago. And even if strong AI is proven to be possible, it does not necessarily mean that strong AI can be developed.
本文由数字化转型网（www.szhzxw.cn）转载而成，来源于MBA智库；编辑/翻译：数字化转型网宁檬树。
