**Table 2.**

*Selected definitions for intelligent systems [15].*

*Technology, Science and Culture - A Global Vision*

**2.1 Defining intelligent systems**

of these architectural paradigms.

**Architectural paradigm Definitions**

with the ability to:

of competencies and behaviors demonstrated by the system.

maximize their probability of achieving a goal [11]. However, some commentators suggest that intelligent systems are simply what it says on the tin: systems with some degree of intelligence. In this respect, some, relatively simple intelligent systems, may not be perceived to be AI. This is, most likely, a reflection of the so-called AI effect, that is, if an AI can successfully solve a problem, it is no longer part of AI [12, 13]. Kotseruba and Tsotsos [14] suggest that instead of looking for a particular definition of intelligence, it may be more practical to explore systems against the set

Like many topics, there are wide and narrow definitions of intelligent systems. As the AI domain has evolved, it has fragmented into a variety of subfields which may or may neither be useful for those approaching intelligent systems for the first time nor necessarily informing a general definition. Nonetheless, a starting point is needed. For the purposes of this chapter, we define intelligent systems as systems

"…act appropriately in an uncertain environment, where appropriate action is that which increases the probability of success, and success is the achievement of

Langley [16] notes that the fragmentation of the AI domain has led to the proliferation of three primary architectural paradigms—multiagent systems, blackboard systems, and cognitive architectures. **Table 1** summarizes the main characteristics

For the purposes of this chapter, we focus on Albus and Meystel's [15] reference architecture for intelligent systems, known as the real-time control system (RCS) architecture. RCS has evolved from a robot control schema to one for intelligent system design and has continued to expand while maintaining the validity of its core

• Modules communicate directly with each other

• No direct communication between modules

Cognitive architectures • Short-term and long-term memories that store the agent's beliefs, goals,

embody the architecture's assumptions

• Architecture specifies inputs and outputs of each module and protocols for

• Modules read and alter a shared memory of beliefs, goals, and short-term

• Representation and organization of structures embedded in memories • Functional processes that operate on structures including performance and

• A programming language to construct knowledge-based systems that

• Architecture places no constraints on how each component operates

Multiagent systems • Distinct modules for different facets of an intelligent system

Blackboard systems • Distinct modules for different facets of an intelligent system

communication

structures

and knowledge

learning mechanisms

behavioral subgoals that support the system's ultimate goal." ([15], p. 8).

**2.2 Toward a general intelligent system architecture**

**54**

**Table 1.**

*Architectural paradigms in AI [16].*

operational focus, shorter time horizon, and a greater emphasis on detail. In this way, sensors at the lowest level of the hierarchy process data locally and relatively regularly, while data from the aggregate sensor network are transmitted up the hierarchy for processing over longer time horizons and on a more global basis. Similarly, the concept of focused attention recognizes that in an uncertain and dynamic environment, limitations in the computation capacity of a given node require focus only on "what is important and ignoring what is irrelevant" ([19], p. 16).

The RCS reference architecture has been designed to integrate concepts from not only other areas of computer science but also control theory, operations research, amongst others [20]. As technology has evolved, so have expectations for what can be achieved in and with intelligent systems. Two related paradigms that have gained significant traction in intelligent systems design in the last two decades are self-organization and self-management, both of which are seen as solutions for managing extreme complexity. De Wolf and Holvoet [21] define self-organization as "a dynamical and adaptive process where systems acquire and maintain structure themselves, without external control." They summarize the essential characteristics of self-organizing systems as:


While self-organization has its roots in emergence, they differ in how robustness is achieved. Interest in emergence in computing can be traced back to Turing [22] who noted that "global order arises from local interactions." Emergence can be defined as follows:

"A system exhibits emergence when there are coherent emergent at the macro-level that dynamically arise from the interactions between the parts at the micro-level. Such emergent are novel with regards to the individual parts of the system." ([21], p. 3).

The essence of the difference between emergence and self-organization is directional. Emergence arises from micro to macro, whereas self-organization is determined from macro to micro. While these seem like discrete paradigms, in reality they can be used together.

The related paradigm of self-management has its roots in autonomic computing and IBM's conceptualization of autonomic computing as "computing systems that can manage themselves given high-level objectives from administrators" [23]. Kephart and Chess [23] further elaborated the essence of autonomic computing systems through four aspects of self-management—self-configuration, self-optimization, self-healing, and self-protection—underpinned by the use of control or feedback loops that collect details from the system and act accordingly, anticipating system requirements and resolving problems with minimal human intervention [24]. In this way, these design principles are consistent with Albus and Meystel [15]. **Table 3** summarizes the definitions of the four aspects of self-management.

**57**

**Table 4.**

represent significant research.

*Design principles for intelligent systems [25].*

*Toward the Intelligent Internet of Everything: Observations on Multidisciplinary Challenges…*

of the system adjusts automatically and seamlessly. Self-optimization Components and systems continually seek opportunities to improve their own

Self-configuration Automated configuration of components and systems follows high-level policies. Rest

Self-protection System automatically defends against malicious attacks or cascading failures. It uses early warning to anticipate and prevent system-wide failures. Self-healing System automatically detects, diagnoses, and repairs localized software and hardware

Understanding by building models and abstracting general principles from the

Taking into account the concepts of emergence and autonomic computing, Pfeifer et al. [25] set out a series of design principles for intelligent systems, albeit in the wider context of robotics. These principles, summarized in **Table 4** below, pertain to both the design procedure and the design of agents, and in themselves

It would be wrong to limit the discussion of intelligent systems architecture to RCS and the paradigms and design principles above; however, a more comprehensive

*DOI: http://dx.doi.org/10.5772/intechopen.83691*

**Concept Definitions**

**Concept Definitions**

constructed model

performance and efficiency.

adaptivity)

problems.

*Self-management aspects of autonomic computing (adapted from [23]).*

required

stimulation

processes

Cheap design Exploitation of niche and interaction; parsimony

interesting ways

Emergence Systems should be designed for emergence (for increased robustness and

Diversity-compliance Solving the tradeoff between exploiting the givens and generating diversity in

Time perspectives Three perspectives required: (i) state-oriented ("here and now"), (ii) ontogenetic

Frame of reference Three aspects must be distinguished—(i) perspective—agent, observer or designer,

Three constituents Definitions of ecological niche (environment), tasks, and the agent itself are always

and largely coupled through interaction with the environment

Redundancy Partial overlap of functionality in different subsystems based on different physical

Ecological balance Balance in complexity of sensory, motor, and neural systems; task distribution between morphology, materials, and control

Value Driving forces; developmental mechanisms; self-organization

Complete agent Embodied, autonomous, self-sufficient, and situated agents are of interest

("learning and development"), and (iii) phylogenetic ("evolutionary")

(ii) behavior versus mechanisms—behavior should be the result of systemenvironment interaction, and (iii) complexity of agent-environment interaction

Intelligence is emergent from parallel, asynchronous, partly autonomous processes,

Behavior sensor-motor coordinated with respect to target; self-generated sensory

Design procedure principles

Agent design principles

Parallel, loosely coupled processes

Sensor-motor coordination

Synthetic methodology

**Table 3.**

*Toward the Intelligent Internet of Everything: Observations on Multidisciplinary Challenges… DOI: http://dx.doi.org/10.5772/intechopen.83691*


#### **Table 3.**

*Technology, Science and Culture - A Global Vision*

of self-organizing systems as:

defined as follows:

ity they can be used together.

conditions to promote a specific function.

behavioral characteristics of its constituent parts.

outside the boundaries of the system.

operational focus, shorter time horizon, and a greater emphasis on detail. In this way, sensors at the lowest level of the hierarchy process data locally and relatively regularly, while data from the aggregate sensor network are transmitted up the hierarchy for processing over longer time horizons and on a more global basis. Similarly, the concept of focused attention recognizes that in an uncertain and dynamic environment, limitations in the computation capacity of a given node require focus

The RCS reference architecture has been designed to integrate concepts from not only other areas of computer science but also control theory, operations research, amongst others [20]. As technology has evolved, so have expectations for what can be achieved in and with intelligent systems. Two related paradigms that have gained significant traction in intelligent systems design in the last two decades are self-organization and self-management, both of which are seen as solutions for managing extreme complexity. De Wolf and Holvoet [21] define self-organization as "a dynamical and adaptive process where systems acquire and maintain structure themselves, without external control." They summarize the essential characteristics

1.Increase in order—an increase in order (or statistical complexity), through organization, is required from some form of semiorganized or random initial

2.Autonomy—this implies the absence of external control or interference from

3.Adaptability or robustness with respect to changes—a self-organizing system must be capable of maintaining its organization autonomously in the presence of changes in its environment. It may generate different tasks but maintain the

4.Dynamical—self-organization is a process from dynamism toward order.

While self-organization has its roots in emergence, they differ in how robustness is achieved. Interest in emergence in computing can be traced back to Turing [22] who noted that "global order arises from local interactions." Emergence can be

"A system exhibits emergence when there are coherent emergent at the macro-level that dynamically arise from the interactions between the parts at the micro-level. Such emergent are novel with regards to the individual parts of the system." ([21], p. 3). The essence of the difference between emergence and self-organization is directional. Emergence arises from micro to macro, whereas self-organization is determined from macro to micro. While these seem like discrete paradigms, in real-

The related paradigm of self-management has its roots in autonomic computing and IBM's conceptualization of autonomic computing as "computing systems that can manage themselves given high-level objectives from administrators" [23]. Kephart and Chess [23] further elaborated the essence of autonomic computing systems through four aspects of self-management—self-configuration, self-optimization, self-healing, and self-protection—underpinned by the use of control or feedback loops that collect details from the system and act accordingly, anticipating system requirements and resolving problems with minimal human intervention [24]. In this way, these design principles are consistent with Albus and Meystel [15].

**Table 3** summarizes the definitions of the four aspects of self-management.

only on "what is important and ignoring what is irrelevant" ([19], p. 16).

**56**

*Self-management aspects of autonomic computing (adapted from [23]).*


#### **Table 4.**

*Design principles for intelligent systems [25].*

Taking into account the concepts of emergence and autonomic computing, Pfeifer et al. [25] set out a series of design principles for intelligent systems, albeit in the wider context of robotics. These principles, summarized in **Table 4** below, pertain to both the design procedure and the design of agents, and in themselves represent significant research.

It would be wrong to limit the discussion of intelligent systems architecture to RCS and the paradigms and design principles above; however, a more comprehensive review is outside the scope of this chapter. In their recent survey of over 40 years of research related to cognitive architectures, Kotseruba and Tsotsos [14] identify over 84 architectures including 49 that are still actively developed. These architectures can and have been classified using a wide range of taxonomies including representation and information processing type they implement as well as the capabilities, properties, and evaluation criteria for the various architectures as represented in publications [14].

## **3. Intelligent methods**

By design, intelligent systems typically acquire and must model and analyze big data, data characterized by its scale and complexity in volume, velocity, variety, variability, and veracity [26]. In addition, and in particular in the context of the Internet of Everything, they must deal with the behavior of unpredictable entities, in this case people, in uncertain and dynamic environments. Depending on the criticality of a decision, for example, in healthcare monitoring scenarios, intelligent systems may need to make trade-offs between the deliberation time, the cost, and the quality of results [27]. While traditional mathematical approaches to modeling and analysis were able to arrive at appropriate decisions when used in relatively simple intelligent systems, it was at a cost, both in terms of time and computation. Sometimes approximate reasoning is good enough and indeed preferable.

From the early 1980s onwards, researchers sought to formalize the cognitive processes of humans and particularly human tolerance for imprecision and uncertainty, so-called soft computing or computational intelligence (CI) [28]. Such techniques are now used widely, independently, and together in intelligent systems to support better and quicker decision-making. Some of the most common categories of intelligent methods include expert systems, artificial neural networks, fuzzy set systems, rough set theory, and evolutionary computing [29, 30]. These are briefly summarized in **Table 5**. It is important to note that these methods or techniques may be implemented discretely or together. Indeed, hybrid intelligent systems are an increasing feature of intelligent systems research.

As discussed earlier, not all intelligent systems are as advanced as others and vary in terms of their reasoning and ability to adapt and evolve. Kotseruba and Tsotsos [14] identify three paradigms in cognitive architectures—symbolic (cognitivist), emergent (connectionist), and hybrid. In the symbolic paradigm, concepts are represented using symbols that can be manipulated using predefined instructions and then implemented as if-then rules applied to the symbols representing the facts known about the world. Symbolic paradigms are particularly useful for planning and reasoning; they are more intuitive, easier to understand, and as a result, more common. However, the symbolic paradigm can have difficulty with perceptual processing and adapting to dynamic and uncertain environments [14]. In contrast, the less common emergent approach deals with changing environments through massively parallel models but, as a result, loses transparency. These emergent approaches include many of the so-called intelligent methods presented in **Table 5** below. Unsurprisingly therefore, hybrid architectures are increasingly the most common [14].

Exploring the pros, cons, and state of the art of this plethora of techniques is beyond the scope of this chapter. However, while deep learning and artificial intelligence research are producing impressive results on a case by case basis, there are a number of bottlenecks not least the human element. Intelligent systems require, if not demand, a knowledge database comprising human domain and task knowledge to make sense of this data and ensure meaningful insights are generated. Not only is access to domain experts a bottleneck, there is a dearth of data scientists to capture this knowledge and design and implement models based on it.

**59**

*Toward the Intelligent Internet of Everything: Observations on Multidisciplinary Challenges…*

similar outputs to other neurons [32]

Expert systems Expert systems are a branch of artificial intelligence aiming to simulate the knowledge

Rough set theory Computing systems based on rough set theory implement a purely mathematical

acquisition process and reasoning of domain experts in order to complete complex tasks [31]. It typically comprises a knowledge base, inference engine, and user interface

Artificial neural networks (ANN) are a computing system that aims to replicate the structure and function of human neurons. The ANN typically consists of a number of interconnected processors, the artificial neurons, which receive inputs similar to the electrochemical impulses that biological neurons receive from other neurons and sends

Fuzzy systems represent an extension of traditional expert system techniques as they leverage mathematical theory of fuzzy sets in order to deal with the typical real-world uncertainty and simulate the normal (and less precise) human reasoning [32]

approach to deal with uncertainty and imperfect knowledge. The key advantage of this type of systems is that they do not need any prior or further information about data [29] Rough set theory is mostly used for enabling automated classifiers and for attribute

Evolutionary computing algorithms, also known as genetic algorithms, are designed to

Genetic algorithms tend to provide satisfactory results for very complex task; therefore, they represent a valuable alternative when suboptimal results are acceptable [30]

deal with randomness, nonlinear, and multidimensional functions [29]

*DOI: http://dx.doi.org/10.5772/intechopen.83691*

**Concept Definition**

Artificial neural networks

Fuzzy set systems

Evolutionary computing

*Summary of selected intelligent methods.*

**Table 5.**

**4. Toward an intelligent Internet of Everything**

selection [33]

tures, adaptive infrastructure, and privacy by design.

**4.1 Ubiquitous sensing**

The architecture for a future Internet of Everything has yet to be determined. Indeed, the very concept of the Internet of Everything lacks definition. Yet, despite this, there is what one might term an "irrational exuberance" about the benefits and value it might generate. We propose an extended definition of the Internet of Everything that includes intelligence at its core. Thus, we define an Intelligent Internet of Everything as a system of systems that connect people, processes, things, data, and social networks and through intelligent systems proactively create new value for individuals, organizations, and society as a whole. It assumes a vision of ambient intelligence (see **Figure 1**) where we live, work and interact in, and with a digitally infused environment that "proactively, but sensibly, supports people in their daily lives" ([34], p. 15). Achieving this vision presents significant opportunities and challenges, not least advances in ubiquitous sensing, cognitive architec-

An Intelligent Internet of Everything built on intelligent systems assumes not only ubiquitous computing but a ubiquitous sensing network. Ubiquitous sensing provides rich multimodal sensing abilities to the ubiquitous computing infrastructure without increasing the burden on the users [35]. The last two decades have seen a rapid advance in the sophistication of sensors including intelligent sensors, *in vivo* sensors, and sensors that can increasingly mimic the biological senses. Like intelligent systems, intelligent sensors can take actions based on data from the environ-

The development of the next generation of sensors is closely linked to innovation in nanomaterials and nanotechnology methods which both enable embedded

ment in which they operate rather than merely measuring a signal [36].

*Toward the Intelligent Internet of Everything: Observations on Multidisciplinary Challenges… DOI: http://dx.doi.org/10.5772/intechopen.83691*


#### **Table 5.**

*Technology, Science and Culture - A Global Vision*

**3. Intelligent methods**

review is outside the scope of this chapter. In their recent survey of over 40 years of research related to cognitive architectures, Kotseruba and Tsotsos [14] identify over 84 architectures including 49 that are still actively developed. These architectures can and have been classified using a wide range of taxonomies including representation and information processing type they implement as well as the capabilities, properties, and evaluation criteria for the various architectures as represented in publications [14].

By design, intelligent systems typically acquire and must model and analyze big data, data characterized by its scale and complexity in volume, velocity, variety, variability, and veracity [26]. In addition, and in particular in the context of the Internet of Everything, they must deal with the behavior of unpredictable entities, in this case people, in uncertain and dynamic environments. Depending on the criticality of a decision, for example, in healthcare monitoring scenarios, intelligent systems may need to make trade-offs between the deliberation time, the cost, and the quality of results [27]. While traditional mathematical approaches to modeling and analysis were able to arrive at appropriate decisions when used in relatively simple intelligent systems, it was at a cost, both in terms of time and computation.

Sometimes approximate reasoning is good enough and indeed preferable.

therefore, hybrid architectures are increasingly the most common [14].

this knowledge and design and implement models based on it.

Exploring the pros, cons, and state of the art of this plethora of techniques is beyond the scope of this chapter. However, while deep learning and artificial intelligence research are producing impressive results on a case by case basis, there are a number of bottlenecks not least the human element. Intelligent systems require, if not demand, a knowledge database comprising human domain and task knowledge to make sense of this data and ensure meaningful insights are generated. Not only is access to domain experts a bottleneck, there is a dearth of data scientists to capture

an increasing feature of intelligent systems research.

From the early 1980s onwards, researchers sought to formalize the cognitive processes of humans and particularly human tolerance for imprecision and uncertainty, so-called soft computing or computational intelligence (CI) [28]. Such techniques are now used widely, independently, and together in intelligent systems to support better and quicker decision-making. Some of the most common categories of intelligent methods include expert systems, artificial neural networks, fuzzy set systems, rough set theory, and evolutionary computing [29, 30]. These are briefly summarized in **Table 5**. It is important to note that these methods or techniques may be implemented discretely or together. Indeed, hybrid intelligent systems are

As discussed earlier, not all intelligent systems are as advanced as others and vary in terms of their reasoning and ability to adapt and evolve. Kotseruba and Tsotsos [14] identify three paradigms in cognitive architectures—symbolic (cognitivist), emergent (connectionist), and hybrid. In the symbolic paradigm, concepts are represented using symbols that can be manipulated using predefined instructions and then implemented as if-then rules applied to the symbols representing the facts known about the world. Symbolic paradigms are particularly useful for planning and reasoning; they are more intuitive, easier to understand, and as a result, more common. However, the symbolic paradigm can have difficulty with perceptual processing and adapting to dynamic and uncertain environments [14]. In contrast, the less common emergent approach deals with changing environments through massively parallel models but, as a result, loses transparency. These emergent approaches include many of the so-called intelligent methods presented in **Table 5** below. Unsurprisingly

**58**

*Summary of selected intelligent methods.*
