泛逻辑学理论——机制主义人工智能理论的逻辑基础

当前,世界各主要大国都把人工智能作为它们的国家战略。人工智能的发展正在快速改变着人类的生活方式和思想观念。在中国,有一小批研究者20多年来一直在基于辩证唯物主义潜心研究具有普适性的人工智能基础理论,包括智能的形成机制、逻辑基础、数学基础、协调机理、矛盾转化等。终于,他们各自建立了机制主义人工智能理论、泛逻辑学理论、因素空间理论、协调学、可拓学、集对分析等。其中,机制主义人工智能理论是基于智能形成机制的通用理论,它能把现有的结构主义、功能主义和行为主义三大流派有机地统一起来,使意识、情感、理智成为三位一体的关系;因素空间理论是机制主义人工智能理论的数学基础;泛逻辑学理论是机制主义人工智能理论的逻辑基础。本文介绍了泛逻辑学理论的基本思想、理论基础和应用方法,阐明它的理论意义和应用价值。特别需要指出的是,在广义概率论基础上建立的命题泛逻辑（包括刚性逻辑和柔性逻辑）,可看成一个完整的命题级智能信息处理算子库,库中完整地包含了全部18种柔性信息处理模式（包括16种布尔信息处理模式）,可用类型编码<a,b,e>来严格区分,用它可寻找到适合自己的信息处理算子完整簇来使用。在每一个信息处理模式中,各种不确定性的组合状态由不确定性程度属性编码<k,h,β,e>来严格区分,用它可在本信息处理模式的算子完整簇中精确选择具体的算子来使用。这表明柔性信息处理本质上是一把密码锁,它需要专门的密码<a,b,e>+<k,h,β,e>才能正常打开,不能乱点鸳鸯谱。通过只有18种模式,每种模式可以从最大算子连续变化到最小算子,已经证明了没有一个命题算子被遗漏。     

多媒体智能：当多媒体遇到人工智能

过去10年中涌现出大量新兴的多媒体应用和服务,带来了很多可以用于多媒体前沿研究的多媒体数据。多媒体研究在图像/视频内容分析、多媒体搜索和推荐、流媒体服务和多媒体内容分发等方向均取得了重要进展。与此同时,由于在深度学习领域所取得的重大突破,人工智能(artificial intelligence,AI)在20世纪50年代被正式视为一门学科之后,迎来了一次“新”的发展浪潮。因此,一个问题就自然而然地出现了：当多媒体遇到人工智能时会带来什么？为了回答这个问题,本文通过研究多媒体和人工智能之间的相互影响引入了多媒体智能的概念。从两个方面探讨多媒体与人工智能之间的相互影响：一是多媒体促使人工智能向着更具可解释性的方向发展;二是人工智能反过来为多媒体研究注入了新的思维方式。这两个方面形成了一个良性循环,多媒体和人工智能在其中不断促进彼此发展。本文对相关研究及进展进行了讨论,并围绕值得进一步探索的研究方向分享见解。希望可以对多媒体智能的未来发展带来新的研究思路。     

Probabilistic Belief Logic and Its Probabilistic Aumann Semantics

In this paper, we present a logic system for probabilistic belief named PBL,which expands the language of belief logic by introducing probabilistic belief. Furthermore, we give the probabilistic Aumann semantics of PBL. We also list some valid properties of belief and probabilistic belief, which form the deduction system of PBL. Finally, we prove the soundness and completeness of these properties with respect to probabilistic Aumann semantics.     

机制主义人工智能理论——一种通用的人工智能理论

现行人工智能研究取得了许多进展,但存在“深度上浅层化、广度上碎片化和体系上封闭化”的重要缺陷。这不是改进算法或者提高硬件性能所能解决的问题,而是要在科学观方法论上寻找根源。本文依据“科学观→方法论→研究模型→研究途径→基本概念→基本原理”这个顶天立地的研究纲领,总结了信息科学的科学观,提炼了信息生态方法论;在新的科学观和方法论指导下构筑了体现智能生长全过程的研究模型,发现了智能生长的共性机制,确立了机制主义研究途径,进而澄清和匡正了信息（特别是语义信息）、感知、知识、认知、基础意识、情感、理智、综合决策等一系列基础概念,总结了实现信息－知识－智能转换的一组基本原理,创建了机制主义人工智能理论。而且证明了:长期三分而立的结构主义（人工神经网络）、功能主义（专家系统）、行为主义（感知动作系统）三大人工智能理论可在机制主义人工智能理论框架内实现和谐统一;机制主义是生成基础意识、情感、理智三位一体高等人工智能的科学途径;机制主义人工智能理论是通用型的人工智能理论。     

人工智能范式的革命与通用智能理论的创生

人工智能的研究取得了不少可喜的进展,也面临着许多严峻的挑战.为了应对这些挑战,学术界提出了各种各样的研究思路.笔者相信,每种思路都有其合理之处,都有可能获得一定的成效.不过,根据笔者的理解,人工智能面临的最深刻最严峻的挑战,是学科和时代的大转变所带来的大阵痛:人工智能范式的张冠李戴.因此,必须对人工智能的范式实施"正冠...     

人工智能在传感器领域的应用

人工智能作为一门正在迅速发展的学科，已被广泛地应用于传感器领域。人工智能与传感器技术相结合构成新型的智能传感器，提高了传感器的智能水平，是智能传感器发展的一个方向。简要介绍了人工智能的四个分支：模糊逻辑、人工神经网络、专家系统、遗传算法在传感器领域的应用。     

Borel型概率计量逻辑

视全体赋值之集为通常乘积拓扑空间,利用该空间上的Borel概率测度在二值命题逻辑中引入了公式的概率真度概念.该方法既克服了计量逻辑学要求赋值集上的概率测度必须为均匀概率测度的无穷可数乘积的局限,又弥补了概率逻辑学只讲局部而缺乏整体性的不足;证明了计量逻辑学中公式的真度、随机真度以及概率逻辑学中公式的概率等概念都可作为本文提出的概率真度的特例而纳入到统一的框架中,从而实现了计量逻辑学与概率逻辑学的融合与统一;证明了逻辑闭理论与赋值空间中的拓扑闭集是一一对应的以及概率真度函数与赋值空间上的Borel概率测度是一样多的等若干结论;本文的第4节给出了公式的概率真度的公理化定义,证明了公式集上满足Kolmogorov公理的任一[0,1]值函数均可由赋值空间上的某Borel概率测度按本文的方法所表出,从而建立了二值命题逻辑框架下的概率计量逻辑的理论体系.     

Borel probabilistic and quantitative logic

The present paper introduces the notion of the probabilistic truth degree of a formula by means of Borel probability measures on the set of all valuations,endowed with the usual product topology,in classical two-valued propositional logic.This approach not only overcomes the limitations of quantitative logic,which require the probability measures on the set of all valuations to be the countably infinite product of uniform probability measures,but also remedies the drawback that probability logic behaves onl...     

概率逻辑、不确定逻辑和模糊逻辑之比较

通过一个实例分析比较了概率逻辑、主观概率逻辑、不确定逻辑和模糊逻辑的思想方法。提出了自己的观点：基于数据统计的概率逻辑是最科学的。不确定逻辑比主观概率逻辑更科学。当具有不确定性的原子命题具有独立性时,不确定逻辑和模糊逻辑的观点是一致的。而对于处理带有不确定性的相关性命题,不确定逻辑比模糊逻辑更科学。但是模糊逻辑在建立推理理论方面见长。     

因素表示的信息空间与广义概率逻辑

国内外近年来所提出的广义概率逻辑对于人工智能的发展有重要意义。能否反映变换演化的实际场景,使逻辑判断能够灵活变通,这是广义概率逻辑发展的关键。为了解决这一问题,本文的目是以信息空间作为逻辑与实际场景的接口。有了这个接口,逻辑判断就能反映变幻莫测的实际场景。本文的方法是用因素空间来定义表现论域以形成新的信息空间,将谓词中的变元取为因素,在已有的逻辑系统中加上本文所提出的背景公理,所有的推理都是在一定背景之下的推理,不同的背景会推出不同的结论。结果是新的逻辑既能维系Stone表示定理的表现要求,又能变得更加灵活有效。结论能使广义概率逻辑更有效地服务于人工智能。为了配合机制主义人工智能的需要,本文还特别提出了语法-语用对接的方法和目标驱动的逆向推理设想,最后为泛逻辑的3种连续算子对进行了数学证明。     

Managing artificial intelligence projects: Key insights from an AI consulting firm

While organisations are increasingly interested in artificial intelligence (AI), many AI projects encounter significant issues or even fail. To gain a deeper understanding of the issues that arise during these projects and the practices that contribute to addressing them, we study the case of Consult, a North American AI consulting firm that helps organisations leverage the power of AI by providing custom solutions. The management of AI projects at Consult is a multi-method approach that draws on elements from traditional project management, agile practices, and AI workflow practices. While the combination of these elements enables Consult to be effective in delivering AI projects to their customers, our analysis reveals that managing AI projects in this way draw upon three core logics, that is, commonly shared norms, values, and prescribed behaviours which influence actors' understanding of how work should be done. We identify that the simultaneous presence of these three logics—a traditional project management logic, an agile logic, and an AI workflow logic—gives rise to conflicts and issues in managing AI projects at Consult, and successfully managing these AI projects involves resolving conflicts that arise between them. From our case findings, we derive four strategies to help organisations better manage their AI projects.     

辩证逻辑与智能

首先对存在争议的辩证逻辑及其与传统逻辑的关系问题进行了讨论.对与人工智能的限度问题密切相关的歌德尔定理进行了分析,澄清了其中的疑难.对人脑中的信息过程、其与逻辑推理的关系、智能的本质等问题,进行了讨论,并提出了作者的见解.     

重新找回人工智能的可解释性

针对深度神经网络AI研究的可解释性瓶颈，指出刚性逻辑（数理形式逻辑）和二值神经元等价，二值神经网络可转换成逻辑表达式，有强可解释性。深度神经网络一味增加中间层数来拟合大数据，没有适时通过抽象把最小粒度的数据（原子）变成粒度较大的知识（分子），再把较小粒度的知识变成较大粒度的知识，把原有的强可解释性淹没在中间层次的汪洋大海中。要支持多粒度的知识处理，需把刚性逻辑扩张为柔性命题逻辑（命题级数理辩证逻辑），把二值神经元扩张为柔性神经元，才能保持强可解释性。本文详细介绍了从刚性逻辑到柔性逻辑的扩张过程和成果，最后介绍了它们在AI研究中的应用，这是重新找回AI研究强可解释性的最佳途径。     

Managing the risks of artificial general intelligence: A human factors and ergonomics perspective

Artificial General Intelligence (AGI) is the next and forthcoming evolution of Artificial Intelligence (AI). Though there could be significant benefits to society, there are also concerns that AGI could pose an existential threat. The critical role of Human Factors and Ergonomics (HFE) in the design of safe, ethical, and usable AGI has been emphasized; however, there is little evidence to suggest that HFE is currently influencing development programs. Further, given the broad spectrum of HFE application areas, it is not clear what activities are required to fulfill this role. This article presents the perspectives of 10 researchers working in AI safety on the potential risks associated with AGI, the HFE concepts that require consideration during AGI design, and the activities required for HFE to fulfill its critical role in what could be humanity's final invention. Though a diverse set of perspectives is presented, there is broad agreement that AGI potentially poses an existential threat, and that many HFE concepts should be considered during AGI design and operation. A range of critical activities are proposed, including collaboration with AGI developers, dissemination of HFE work in other relevant disciplines, the embedment of HFE throughout the AGI lifecycle, and the application of systems HFE methods to help identify and manage risks.     

Biological and artificial intelligence

The paper discusses the characteristics of Biological Intelligence (BI) and its differences with artificial intelligence. In particular the plasticity of the nervous system is considered in the different forms with special attention to deterministic and localizationist views of the brain vs holistic approaches. When memory and learning are considered the localizationist views do not offer a possible solution to a number of problems while memory may be better conceptualized in terms of categorization procedures and generalizing strategies. Finally, the problem of individual variability, an important feature in terms of BI, is considered. The legitimacy of analogies between BI and AI is discussed and the necessity for an innovative approach to the field of AI is stressed.     

Bee colony optimization for the satisfiability problem in probabilistic logic

This paper presents the first heuristic method for solving the satisfiability problem in the logic with approximate conditional probabilities. This logic is very suitable for representing and reasoning with uncertain knowledge and for modeling default reasoning. The solution space consists of variables, which are arrays of 0 and 1 and the associated probabilities. These probabilities belong to a recursive non-Archimedean Hardy field which contains all rational functions of a fixed positive infinitesimal. Our method is based on the bee colony optimizationmeta-heuristic. The proposed procedure chooses variables from the solution space and determines their probabilities combining some other fast heuristics for solving the obtained linear system of inequalities. Experimental evaluation shows a high percentage of success in proving the satisfiability of randomly generated formulas. We have also showed great advantage in using a heuristic approach compared to standard linear solver.     

有关行为主义人工智能研究综述

通过与传统人工智能的比较，介绍了基于行为的智能模拟技术的发展及现状，并详细评述了行为主义人工智能的研究方向以及在研究过程中涉及到的前沿技术，最后给出了基于行为主义人工智能构建智能主体系统的设计原则。     

A note on web intelligence, world knowledge and fuzzy logic

Existing search engines––with Google at the top––have many remarkable capabilities; but what is not among them is deduction capability––the capability to synthesize an answer to a query from bodies of information which reside in various parts of the knowledge base.

In recent years, impressive progress has been made in enhancing performance of search engines through the use of methods based on bivalent logic and bivalent-logic-based probability theory. But can such methods be used to add nontrivial deduction capability to search engines, that is, to upgrade search engines to question-answering systems? A view which is articulated in this note is that the answer is “No.” The problem is rooted in the nature of world knowledge, the kind of knowledge that humans acquire through experience and education.

It is widely recognized that world knowledge plays an essential role in assessment of relevance, summarization, search and deduction. But a basic issue which is not addressed is that much of world knowledge is perception-based, e.g., “it is hard to find parking in Paris,” “most professors are not rich,” and “it is unlikely to rain in midsummer in San Francisco.” The problem is that (a) perception-based information is intrinsically fuzzy; and (b) bivalent logic is intrinsically unsuited to deal with fuzziness and partial truth.

To come to grips with fuzziness of world knowledge, new tools are needed. The principal new tool––a tool which is briefly described in this note––is Precisiated Natural Language (PNL). PNL is based on fuzzy logic and has the capability to deal with partiality of certainty, partiality of possibility and partiality of truth. These are the capabilities that are needed to be able to draw on world knowledge for assessment of relevance, and for summarization, search and deduction.     

Product creation: An appropriate coupling of human and artificial intelligence

Small batch manufacture dominates the manufacturing sector of a growing number of industrialised countries. The organisational structures and management methods currently adopted in such enterprises are firmly based upon historical developments which started with individual craftsmen. These structures and methods are primarily concerned with the co-ordination of human activities, rather than with the management of theknowledge process underlying the creation of products.This paper argues that it is the failure to understand this knowledge process and its effective integration at aKnowledge Level which presents the real barrier to increased flexibility, not, as is presently perceived, a lack of suitableInformation Level integration. Potential techniques and methodologies for achievingKnowledge Level integration are beginning to emerge from Artificial Intelligence research. Realisation of full Knowledge Level integration will not only require further research into the AI techniques and methodologies involved, but also an understanding of the wider human aspects of their application. Some questions concerning the effective coupling of human and artificial intelligence to achieve Knowledge Level integration of the product creation process are presented.     

An application of artificial intelligence to the choreography of dance

This paper describes an experimental computer program that applies the techniques of artificial intelligence to the creation of dance. Specifically, a user expresses a set of dance rules (in a special English-like rule language) which describes some of the dynamic aspects of a dance. These rules are applied nondeterministically by a rule driver program. The rules themselves are similar to those that form the knowledge base of expert systems. The rule driver embodies a heuristic algorithm of the type found in many artificial intelligence programs.James H. Bradford is an Associate Professor of Computer Science at Brock University. He is an active researcher in the area of Human/Computer Interaction with particular interests in the analysis of speech and the representation of dance.Paulette Côté-Laurence is an Associate Professor of Physical Education at Brock University. Her research interests are in the areas of motor control and the acquisition of dance skills, psychology of rhythm, and dance technology.