The Central Role of Heuristic Search in Cognitive Computation Systems

This paper focuses on the relation of heuristic search and level of intelligence in cognitive computation systems. The paper begins with a review of the fundamental properties of a cognitive computation system, which is defined generally as a control system that generates goal-directed actions in response to environmental inputs and constraints. An important property of cognitive computations is the need to process local cues in symbol structures to access and integrate distal knowledge to generate a response. To deal with uncertainties involved in this local-to-distal processing, the system needs to perform heuristic search to locate and integrate the right set of distal structures. The level of intelligence of the system depends critically on the efficiency of the heuristic search process. It is argued that, for a bounded rationality system, the level of intelligence does not depend on how much search it needs to do to accomplish a task. Rather, the level of intelligence depends on how much search it does not need to do to achieve the same level of performance. Examples were discussed to illustrate this idea. The first two examples show how machines that play games like tic-tac-toe and chess rely heavily on the efficiency of the heuristic search algorithm to achieve better performance, demonstrating the relation of heuristic search and intelligence in a bounded rationality system. The second example shows how humans adapt to different information ecologies to perform information search on the Internet and how their performance improves over time, demonstrating how heuristic search can be improved in an adaptive rationality system. The two examples demonstrate how better search control knowledge and representations of task environment can improve the efficiency of heuristic search, thereby improving the intelligence of the system.     

可信人工智能系统的质量属性与实现: 三级研究

人工智能系统以一种前所未有的方式,被广泛地用于解决现实世界的各种挑战,其已然成为推动人类社会发展的核心驱动力.随着人工智能系统在各行各业的迅速普及,人们对人工智能系统的可信性愈发感到担忧,其主要原因在于,传统软件系统的可信性已不足以完全描述人工智能系统的可信性.对于人工智能系统的可信性的研究,具有迫切的需要.目前已有大量相关研究,且各有侧重,但缺乏一个整体性、系统性的认识.本研究是一项以现有二级研究为研究对象的三级研究,旨在揭示人工智能系统的可信性相关的质量属性和实践的研究现状,建立一个更加全面的可信人工智能系统质量属性框架.本研究收集、整理和分析了2022年3月前发表的34项二级研究,识别了21种与可信性相关的质量属性及可信性的度量方法和保障实践.研究发现,现有研究主要关注在安全性和隐私性上,对于其它质量属性缺乏广泛且深入的研究.对于需要跨学科协作的两个研究方向,需要在未来的研究中引起重视,一方面是人工智能系统本质上还是一个软件系统,其作为一个软件系统的可信值得人工智能和软件工程专家合作研究;另一方面,人工智能是人类对于机器拟人化的探索,如何从系统层面保障机器在社会环境下的可信,如怎样满足人本主义,值得人工智能和社会科学专家合作研究.     

Intelligent control of a rotary kiln fired with producer gas generated from biomass

During the past decade, the academic world has been extremely active in developing new algorithms and theories in the field of artificial intelligence (AI) and intelligent systems. In most cases, however, emphasis has been placed more on theoretical frameworks and mathematical bases than on what the individual AI techniques could offer and on how different techniques could be applied to solve real industrial-scale problems. The reputation of intelligent systems has consequently suffered from an inability to transfer new and sophisticated techniques to industrial applications with identifiable benefits. As a result, although a wide range of intelligent control techniques has been available already for many years, most of the applications in the process industry are based on more conventional techniques. Recently, as awareness of intelligent systems has grown, industrial problems and implementations have fortunately received increasing attention. In this paper, an intelligent supervisory-level system implemented at one of the major Finnish pulp mills to control a lime kiln fired with producer gas generated from biomass is presented. First, the major results of a field study are summarised, with special attention paid to burnt lime quality aspects. Next, a novel linguistic equations approach, which provides flexible methods for both modelling and control, is briefly described. The overall structure and main functions of the developed control system are then described with the main emphasis on the control of temperature and lime quality. Finally, the results obtained during the extended testing period of the system are presented and discussed.     

Artificial Driving Intelligence and Moral Agency: Examining the Decision Ontology of Unavoidable Road Traffic Accidents through the Prism of the Trolley Dilemma

The question of the capacity of artificial intelligence to make moral decisions has been a key focus of investigation in robotics for decades. This question has now become pertinent to automated vehicle technologies, as a question of understanding the capacity of artificial driving intelligence to respond to unavoidable road traffic accidents. Artificial driving intelligence will make a calculated decision that could equate to deciding who lives and who dies. In calculating such important decisions, does the driving intelligence require moral intelligence and a capacity to make informed moral decisions? Artificial driving intelligence will be determined by at very least, state laws, driving codes, and codes of conduct relating to driving behaviour and safety. Does it also need to be informed by ethical theories, human values, and human rights frameworks? If so, how can this be achieved and how can we ensure there are no moral biases in the moral decision-making algorithms? The question of moral capacity is complex and has become the ethical focal point of this technology. Research has centred on applying Philippa Foot’s famous trolley dilemma. We claim that before applications attempt to focus on moral theories, there is a necessary precedent to utilise the trolley dilemma as an ontological experiment. The trolley dilemma is succinct in identifying important ontological differences between human driving intelligence and artificial driving intelligence. In this paper, we argue that when the trolley dilemma is focused upon ontology, it has the potential to become an important elucidatory tool. It can act as a prism through which one can perceive different ontological aspects of driving intelligence and assess response decisions to unavoidable road traffic accidents. The identification of the ontological differences is integral to understanding the underlying variances that support human and artificial driving decisions. Ontologically differentiating between these two contexts allows for a more complete interrogation of the moral decision-making capacity of the artificial driving intelligence.     

Interacting with the real world: design principles for intelligent systems

The last two decades in the field of artificial intelligence have clearly shown that true intelligence always requires the interaction of an agent with a real physical and social environment. The concept of embodiment that has been introduced to designate the modern approach to designing intelligence has far-reaching implications. Rather than studying computation alone, we must consider the interplay between morphology, materials, brain (control), and the environment. A number of case studies are presented, and it is demonstrated how artificial evolution and morphogenesis can be used to systematically investigate this interplay. Taking these ideas into account requires entirely novel ways of thinking, and often leads to surprising results.This work was presented, in part, at the 9th International Symposium on Artificial Life and Robotics, Oita, Japan, January 28–30, 2004     

Approximate dynamic programming approach for process control

We assess the potentials of the approximate dynamic programming (ADP) approach for process control, especially as a method to complement the model predictive control (MPC) approach. In the artificial intelligence (AI) and operations research (OR) research communities, ADP has recently seen significant activities as an effective method for solving Markov decision processes (MDPs), which represent a type of multi-stage decision problems under uncertainty. Process control problems are similar to MDPs with the key difference being the continuous state and action spaces as opposed to discrete ones. In addition, unlike in other popular ADP application areas like robotics or games, in process control applications first and foremost concern should be on the safety and economics of the on-going operation rather than on efficient learning. We explore different options within ADP design, such as the pre-decision state vs. post-decision state value function, parametric vs. nonparametric value function approximator, batch-mode vs. continuous-mode learning, and exploration vs. robustness. We argue that ADP possesses great potentials, especially for obtaining effective control policies for stochastic constrained nonlinear or linear systems and continually improving them towards optimality.     

When Colossus and Xactimate are not exact: how computerized claims adjusting software has not changed the landscape of insurance litigation

The insurance claim adjusting process – the process by which an insurance company decides how much it will pay on a claim – is moving from a pen-and-paper task to a computer automated system. Computerized claims adjusting software is very efficient, but is not very transparent, and this mystifying process has created new legal concerns among plaintiffs' lawyers and claims adjusters. This article demonstrates how traditional legal claims can be strengthened by this move to a computerized regime. Rather than trying to find inventive new claims or procedures for bringing their claims, potential plaintiffs should look to the ways in which an automated system can improve their claims for bad faith insurance adjusting or unfair trade practices. Although many insurance companies have begun using the same adjusting software, plaintiffs' claims are not appropriate for an antitrust action and cannot be brought as a class action. Although technology has made this seem like a modern problem, this article finds that established law is well suited to handle plaintiffs' complaints.     

工业控制系统的信息物理因果流模型

工业控制系统向智能控制的发展随着人工智能的因果革命也应该进行因果建模的思考。通过提出工业过程控制系统的单、多层信息物理结构,引入信息、物理因果流,建立过程控制系统的信息物理因果流模型,为依据因果关系设计工业控制系统、分析控制运行机制、实施系统故障预测和监控等提供理论基础和描述框架。     

Reactive process control using a blackboard architecture

A software architecture to engineer complex process control applications must combine into the same paradigm efficient reactive and real-time functionalities and mechanisms to capture dynamic time-pressured intelligent behaviors, and must provide convenient high level tools to free the programmer from having to think at an unappropriate level of detail. We implement such characteristics into a blackboard framework that builds the basic abstract elements of reactive behavior and the blackboard computational model on top of low level real-time operating system functions. Under this approach, the engineer gets a powerful and flexible high level medium to map a complex system design that requires artificial intelligence techniques, like intelligent monitoring, and reactive planning and execution, with fully support for real-time programming. The paper also reviews other alternatives which have been explored in the past recent years for implementing complex reactive planning and execution systems.     

Impact of virtual channels and adaptive routing on applicationperformance

Research on multiprocessor interconnection networks has primarily focused on wormhole switching, virtual channel flow control, and routing algorithms to enhance their performance. The rationale behind this research is that by alleviating the network latency for high network loads, the overall system performance would improve; many studies have used synthetic workloads to support this claim. However, such workloads may not necessarily capture the behavior of real applications. In this paper, we have used parallel applications for a closer examination of the network behavior. In particular, the performance benefit from enhancing a 2D mesh with virtual channels (VCs) and a fully adaptive routing algorithm is examined with a set of shared-memory and message passing applications. Execution time and average message latency of shared memory applications are measured using execution-driven simulation and by varying many architectural attributes that affect the network workload. The communication traces of message passing applications, collected on an IBM-SP2, are used to run a trace-driven simulation of the mesh architecture to obtain message latency. Simulation results show that VCs and adaptive routing can reduce the network latency to varying degrees depending on the application. However, these modest benefits do not translate to significant improvements in the overall execution time because the load on the network is not high enough to exploit the advantages of the network enhancements. Moreover, this benefit may be negated if the architectural enhancements increase the network cycle time. Rather, emphasis should be placed on improving the raw network bandwidth and faster network interfaces     

AI,anthropocentrism, and the evolution of ‘intelligence’

Intuitive conceptions guide practice, but practice reciprocally reshapes intuition. The intuitive conception of intelligence in AI was originally highly anthropocentric. However, the internal dynamics of AI research have resulted in a divergence from anthropocentric concerns. In particular, the increasing emphasis on commonsense knowledge and peripheral intelligence (perception and movement) in effect constitutes an incipient reorientation of intuitions about the nature of intelligence in a non-anthropocentric direction. I argue that this conceptual shift undermines Joseph Weizenbaum's claim that the project of artificial intelligence is inherently dehumanizing.     

Artificial intelligence test: a case study of intelligent vehicles

To meet the urgent requirement of reliable artificial intelligence applications, we discuss the tight link between artificial intelligence and intelligence test in this paper. We highlight the role of tasks in intelligence test for all kinds of artificial intelligence. We explain the necessity and difficulty of describing tasks for intelligence test, checking all the tasks that may encounter in intelligence test, designing simulation-based test, and setting appropriate test performance evaluation indices. As an example, we present how to design reliable intelligence test for intelligent vehicles. Finally, we discuss the future research directions of intelligence test.     

Lightweight monitoring of MPI programs in real time

Current technologies allow efficient data collection by several sensors to determine an overall evaluation of the status of a cluster. However, no previous work of which we are aware analyzes the behavior of the parallel programs themselves in real time. In this paper, we perform a comparison of different artificial intelligence techniques that can be used to implement a lightweight monitoring and analysis system for parallel applications on a cluster of Linux workstations. We study the accuracy and performance of deterministic and stochastic algorithms when we observe the flow of both library‐function and operating‐system calls of parallel programs written with C and MPI. We demonstrate that monitoring of MPI programs can be achieved with high accuracy and in some cases with a false‐positive rate near 0% in real time, and we show that the added computational load on each node is small. As an example, the monitoring of function calls using a hidden Markov model generates less than 5% overhead. The proposed system is able to automatically detect deviations of a process from its expected behavior in any node of the cluster, and thus it can be used as an anomaly detector, for performance monitoring to complement other systems or as a debugging tool. Copyright © 2005 John Wiley & Sons, Ltd.     

MINE VENTILATION EXPERT SYSTEM

A growing trend in the area of artificial intelligence is the development of software that simulates the decision-making process of human experts. The purpose of this paper is to advance the use of available tools from the area of artificial intelligence, and in particular expert systems, into the field of mine ventilation management. The authors have developed an expert system (i.e., a smart manager) that controls the operation of a mine ventilation network. The expert system is rule based and hence can be (independently) enhanced in an incremental fashion. It serves as a smart interface between the monitoring and control components of a mine ventilation system.     

Distributed intelligence for multi-camera visual surveillance

Latest advances in hardware technology and state of the art of computer vision and artificial intelligence research can be employed to develop autonomous and distributed monitoring systems. The paper proposes a multi-agent architecture for the understanding of scene dynamics merging the information streamed by multiple cameras. A typical application would be the monitoring of a secure site, or any visual surveillance application deploying a network of cameras. Modular software (the agents) within such architecture controls the different components of the system and incrementally builds a model of the scene by merging the information gathered over extended periods of time. The role of distributed artificial intelligence composed of separate and autonomous modules is justified by the need for scalable designs capable of co-operating to infer an optimal interpretation of the scene. Decentralizing intelligence means creating more robust and reliable sources of interpretation, but also allows easy maintenance and updating of the system. Results are presented to support the choice of a distributed architecture, and to prove that scene interpretation can be incrementally and efficiently built by modular software.     

面向边缘计算应用的宽度孪生网络

边缘计算是将计算、存储、通信等任务分配到网络边缘的计算模式. 它强调在用户终端附近执行数据处理过程, 以达到降低延迟, 减少能耗, 保护用户隐私等目的. 然而网络边缘的计算、存储、能源资源有限, 这给边缘计算应用的推广带来了新的挑战. 随着边缘智能的兴起, 人们更希望将边缘计算应用与人工智能技术结合起来, 为我们的生活带来更多的便利. 许多人工智能方法, 如传统的深度学习方法, 需要消耗大量的计算、存储资源, 并且伴随着巨大的时间开销. 这不利于强调低延迟的边缘计算应用的推广. 为了解决这个问题, 我们提出将宽度学习系统(Broad learning system, BLS)等浅层网络方法应用到边缘计算应用领域, 并且设计了一种宽度孪生网络算法. 我们将宽度学习系统与孪生网络结合起来用于解决分类问题. 实验结果表明我们的方法能够在取得与传统深度学习方法相似精度的情况下降低时间和资源开销, 从而更好地提高边缘计算应用的性能.     

人工智能技术在矿山智能化建设中的应用初探

介绍了人工智能技术的相关概念、发展概述及其在煤炭行业发展中的应用,指出目前人工智能技术在矿山应用只是点状结合和浅度结合,没有实现人工智能技术和矿山某个生产或管理系统层面的深度融合。概述了智能矿山的发展历程,指出智能矿山是人工智能技术、大数据技术、物联网技术和矿山实体的深度融合体,利用智能通信、智能控制和智能计算技术实现数字化矿山的计算、处理,构建数字孪生矿山,通过数字孪生矿山和物理矿山的智能交互演化,达到对煤矿安全、高效、绿色的生产控制。构建了将人工智能技术和矿山深度融合的包括设备层、智能层、应用层的智能矿山三层构架:应用层处于智能矿山的最高层,其中的数字孪生矿山子层相当于“数字大脑”,实现矿山最高层次的智能控制;智能层中的智能体要求子系统不仅仅是应用人工智能技术处理子系统所产生的数据,而是从架构上就要将智能计算、智能通信、智能控制融为一体。展望了智能矿山建设的发展趋势:智能化矿山需要加强人工智能技术和矿山融合度的深入研究,将现有的基于人工智能的故障检测、诊断及超前干预技术应用到机器人系统中,智能计算、智能通信、智能控制融合的巡检机器人将是最早能推广的井下智能体之一;智能化矿山需要进一步加强复杂巨系统建模技术的研究,只有建立了矿山的复杂巨系统模型,才能实现采矿活动和环境的协同互动,实现采煤活动的精准控制,复杂巨系统模型的缺乏将是未来智能矿山建设亟需解决的问题。     

可解释的知识图谱推理方法综述

近年来,以深度学习模型为基础的人工智能研究不断取得突破性进展,但其大多具有黑盒性,不利于人类认知推理过程,导致高性能的复杂算法、模型及系统普遍缺乏决策的透明度和可解释性。在国防、医疗、网络与信息安全等对可解释性要求严格的关键领域,推理方法的不可解释性对推理结果及相关回溯造成较大影响,因此,需要将可解释性融入这些算法和系统中,通过显式的可解释知识推理辅助相关预测任务,形成一个可靠的行为解释机制。知识图谱作为最新的知识表达方式之一,通过对语义网络进行建模,以结构化的形式描述客观世界中实体及关系,被广泛应用于知识推理。基于知识图谱的知识推理在离散符号表示的基础上,通过推理路径、逻辑规则等辅助手段,对推理过程进行解释,为实现可解释人工智能提供重要途径。针对可解释知识图谱推理这一领域进行了全面的综述。阐述了可解释人工智能和知识推理相关概念。详细介绍近年来可解释知识图谱推理方法的最新研究进展,从人工智能的3个研究范式角度出发,总结了不同的知识图谱推理方法。提出对可解释的知识图谱推理研究前景和未来研究方向。     

基于监控大数据平台上的人工智能应用总体设计

针对电网监控业务管理中存在系统运行复杂、日常工作繁杂、管理难度高和人员培训难度大等难题,本文设计了一套基于大数据平台的人工智能电网监控管家管理系统,系统中引入了监控人工智能,通过智能交互中心、智能监视中心和智慧处理中心三大技术实现了系统的全过程设备维护、智能巡检、行为监管、智能提醒服务、记录信息错误提醒、监控信号异常缺陷处置、智能交互、设备信息调试全过程管控、遥控防误操作和人员培训评价,全方位掌握监控人员工作的规范性,对履职能力进行分析评价,提出提升和改进建议,同时系统融入了智能语音处理相互技术、智能图像识别技术、专家库技术,实现技术突破,提高了工作效率,建立了监控信息分析,优化了工作模式、增强了专业管理,为监控工作的顺利、高效完成奠定坚实基础。     

A system for supporting organizations in knowledge-based document preparation

The paper discusses how recurrent organizational activities such as document preparation can be supported by a knowledge-based document preparation tool. REGENT (REport GENeration Tool) is a software environment, which generates documents from reusable document pieces by planning, executing and monitoring the document preparation process in an organizational setting. The documents are constructed from stored document pieces using artificial intelligence methods. A system architecture is developed to enable the document generation process within a broader office automation setting. The report preparation process knowledge is captured in a knowledge representation scheme. A two-phased artificial intelligence problem solving strategy is developed to carry out the reasoning steps when configuring reports from document pieces. The REGENT environment is especially effective when preparing recurrent report types such as the preparation of annual reports. The approach is illustrated with examples gathered during the partial implementation of REGENT at FAW (Artificial Intelligence Research Institute).