In this section we present the basic postulates concerning efficiency management of tasks realization by a single agent and agent society.
Adaptive Judgement Relative to The Agent Needs Hierarchy
1. By using c-granules an agent control is able to perceive some properties of phys- ical control states. One of the properties of these states are agent needs. Any agent need is represented by a c-granule. Examples of the agent needs concern:
(i) acquisition of resources necessary for agent functionality, (ii) judgement of solutions of problems (e.g., problems related to the construction of an interaction plan leading to satisfaction of another need), (iii) judgement if conditions for the initialization of an interaction plan are satisfied to a satisfactory degree for mak- ing a decision about their initialization, (iv) judgement if results of realization of the interaction plan expected by the agent are satisfactory, and if not the agent decides which actions should be performed for development and initialization of actions responsible for correction.
2. Usually, an agent has c-granules for adaptive judgement of solutions of prob- lems related to the c-granule functionality. They are used for construction, actu- alization, and aggregation of current variants of problem solutions along with judgement on created argumentsforandagainstselection of the variants in the agent activity context. In particular, this judgement may lead to conflict resolu- tion among argumentsforandagainstfor selecting one solution. Changes in the agent activity environment may lead to changes caused by c-granules for adaptive judgement in argumentsforandagainst, their aggregations in the form of solution variants as well the results of aggregation in the form of adapted judgement of resolved issue.
3. For any agent need and any agent time momentt, the c-granule corresponding to the need has a skill for adaptive judgement of such attributes of agent needs as: (i) timeliness and importance att of the agent need (especially important is judgement of the need importance, relative to importance att of other agent needs), (ii) degree of the need realization att (i.e., degree in which the agent is satisfied and/or unsatisfied attfrom the need realization).
4. Agent control has the ability to perceive theagent needs hierarchy. This c-granule is an aggregation of c-granules of needs (representing the properties of perceived needs by agent control), c-granules representing (adaptive) relations among them and adaptive judgements over the actualization of components in the agent needs hierarchy. Agent control has skills for identifying which of the needs from the agent needs hierarchy are currently important. Such needs are aggregated into a c-granule called thecurrent hierarchy of prioritized needs. This c-granule is a part of the agent hierarchy of needs.
Cost/Benefit Analysis and Interaction Plans
1. An agent develops, realizes and adopts plans (represented by c-granules) which on the basis of the agent judgement will increase degrees of the agent needs real- ization (especially the most important needs according to the current prioritized hierarchy of needs). In general, the realization of plans causes some agent costs (often they are related to decreasing degrees of realization of the agent needs).
The agent expects some benefits after the realization of the interaction plan (often they result in increasing degrees of agent needs realization).
2. The agent has some skills for the estimation of costs and benefits for supporting (i) analysis and selection of interaction plans and (ii) judgement performed by an agent control for the selection of plans for realization. Agent hasinteraction plan cost c-granule,interaction plan benefit c-granule, andc-granule for comparison of costs and benefits of interaction plans. The implementation of this approach should make it possible for agents to base judgement on different variants of inter- action plans in a framework related to a well known approach in economy called Cost/Benefit Analysis (http://en.wikipedia.org/wiki/Cost-benefit_analysis).
Swot Analysis and Interaction Plan
1. The realization of interaction plans may be disturbed. As a consequence the expected properties of the interaction plan realization may be different from the real ones. A disruption of the interaction plan isnegativeif at least one of the following conditions is satisfied: (i) “ratio” of total benefits to costs of the interac- tion plan realization (under this disruption) substantially decreases or (ii) the total cost of the interaction plan realization (under this disruption) is not acceptable.
A disruption of the interaction plan ispositiveif it is not negative and at the same time it substantially increases the “ratio” of total benefits to costs of interaction plan realization (under this disruption). Any interaction plan realization should be linked to a prediction of likelihood and consequences of negative and/or pos- itive disruptions. Hence, realization of the interaction plans is related to the risk management encompassing the risk assessment and the risk treatment. The basis for analysis and comprehension by agents of their current situation is the SWOT (Strengths, Weaknesses, Opportunities and Threats) analysis [46–49]. According to the definition, a negative disruption causes negative consequences for the inter- action plan realization. The larger is the likelihood of the negative consequences, the larger is the risk. Hence, the risk of negative disruption in a given interaction plan realization is an aggregation of the negative consequences of this negative disruption and the disruption likelihood. The risk of the interaction plan is an aggregation over the interaction plan of all risks of the negative disruption of this plan.
Corisks and the Efficiency of Interaction Plan
1. Agent may use the strength for utilizing the (not predicted before) opportunities.
Hence, it may occur a positive disruption during interactive plans realization.
This leads to thecoriskconcept, analogously to the risk concept. More formally, corisk of a positive disruption of the interaction plan realization is an aggregation of consequences of this positive disruption and its likelihood, under assumption that they are not changing the risk of negative disruptions below acceptable level.
The corisk of an interaction plan is an aggregation of all corisks of this interaction plan. Agent judgement concerning the developed interaction plans may take into account theinteraction plan efficiency, i.e., an aggregation of the total cost, benefit, risk, and co-risk of the interaction plan.
Realization of the Most Important Tasks of the Agent Based on Perception Processes Leading to Comprehension of Perceived Situations by the Agent
1. Perception of situation by the agent is a process leading to construction by agent of a c-granule representing comprehension of the perceived situation from sensory information. This c-granule is calledsituation comprehension c-granule.
2. The situation comprehension c-granule is an aggregation of relevant c-granules resulting from classification of interactions. This c-granule is representing such computational building blocks(see cited Sect.6.4sentences by Leslie Valiant) for the perceived situation as contexts, SWOT, risks, corisks, prioritized needs or prioritized initialization and realization of interactions in the given situation.
The computational building blocks are used by the agent control for concurrent initialization and realization of possibly efficient interaction plans aiming at, in the agent belief, realization of the agent prioritized needs (including the need of better comprehension of perceived situation).
3. The agent has skills for perception and perception evolving. These skills are used for possibly efficient actualization, improvement, and satisfying of the agent hier- archy of needs which is adaptively changing. Agent is performing these tasks by development, realization, verification (judging), and adaptation of interaction plans—following as much as possible the framework of PDCA cycles (Plan-Do- Check-Act) [50,51]—aiming at construction of the interaction plans which are possibly more and more efficient and are increasingly less risky as well as have possibly more and more high corisk. Agent in searching—as far as possible in PDCA cycles—for ‘optimal’ c-granules expressing relevant features of processes (such as discovery, learning and satisfying the prioritized and adaptively changing agent needs and relations among them) aims at possibly efficient improving these processes. This’optimization’ of c-granules supports the agent in more and more efficient construction, realization, adaptive judgement and adaptation of interac- tion plans that change over time for a more efficient realization of the agent’s priority needs.
Communication Among Agents
1. Communications of agents are realized trough interactions. c-granules are the basic agent constructs for interaction with the environment.
2. Agents are using some specialized c-granules for increasing the efficiency per- formance. Among such c-granules thesemiotic c-granules play the important role. A semiotic c-granule is obtained by aggregation of a c-granule g with another c-granulegcalled thecontext interpretationofg. Semiotic c-granules are supporting the agent’s control in improving the approximation tasks (e.g., identi- fication, specification, and comprehension of the current context of the realization of the agent’s activities) as well as construction and realization of (semi)optimal plans (in a given context). Aprivate agent languageconsists of a distinguished family of semiotic c-granules and is closed with respect to some selected opera- tions of construction of new semiotic c-granules.
3. Agents from agent societies are constructing, using and developing commu- nication skills among agents from these societies. For these purposes they are constructing, using, and developingcommunication languagesrelevant for their needs. In communication processes, agents with two roles are fundamental:
senderand receiver. For example, a sender agent is activating some plans of interactions, which leads to producing or distinguishing some hunks, called the sender artefacts. Receiver agent is perceiving, judging and sometimes storing co-existing situations, and behavioral patterns of the sender-agent together with created or pointed by the receiver-agents’s artefacts. If, for the receiver agent, the co-existence of situations or artefacts is relevant then it is represented as a c-granule, stored, and a special name (and type) to this c-granule is assigned.
Agents may change their roles as sender and receiver; this leads to dialogues.
4. Indialogues agents may learn from each other and/or adjust properties of inter- action plans. This leads to common comprehension of structures and properties ofsocial c-granules(in particular, common plans realized by the society). Each agent is producing a c-granule and treating it as an interpretation of the social c-granule. It may happen that different agents may have slightly different com- prehension of a given social c-granule. Then agents may try through dialogues to reach a consensus about comprehending this social c-granule (or in some cases to modify it).
Agent Team Cooperation in Problem Solving
If the society of agents is satisfying a collection of conditions following from the specificity of the class of problems to be solved, then this society may be able to undertake and realize constructive cooperation during problem solving processes.
For cooperation processes the following aspects are important:
1. Thequality of the common concept ontologyused for initiating and realization of the project for resolving problems from a given class.
2. Relevant selection (especially with respect to initiation and realization) of an adaptive decompositionof a given problem into subproblems to be solved by properly prepared agent subteams.
3. Communication efficiencyamong agents, in particular in the scope of prioritized issues, to be solved.
A society of agents may increase efficiency of problem solving using strategies for discovery and improving (semi)optimal social c-granules relevant for concepts (and relations among them) which efficiently support all the required or predicted problems of concern.
In this section, we have presented a preliminary discussion on basic intuition of c-granule usage in the WisTech framework. From this discussion one can observe how complex the concept ontology of WisTech is. In particular, this concerns the concept of c-granule as well as diversity of c-granules. Taking into account the necessity of judgement under uncertainty, this ontology should encompass the concepts necessary for specification, realization, verification and development of such processes, as:
(a) identification of the current situation and characterization of the most important features from the agent needs hierarchical perspective,
(b) discovery of the relevant context of the current situations should be considered (by considering its past, future, risks, costs, benefits as well as likelihoods and vulnerabilities),
(c) similarity of the current situation and plans with the situations and interaction plans analyzed (observed and/or simulated) in the past needs to be recognized, (d) relevant deviations of expected or predicted interaction results from the real ones
need to be identified,
(e) appearance of unexpected possible changes in the recognized situation needs to be counted by the agent.
We have pointed out that in complex real-life projects there is also a need for developing a language used for carrying out the relevant adaptive judgements con- cerning above mentioned concepts and satisfiability degrees; in particular, conflict resolution among the argumentsforandagainst, as well as a language for expressing different patterns used for inducing new concepts and properties of observed phe- nomena, are a few to name. For example, the properties of observed phenomena are useful in further activities and cooperation of societies of agents. This need is very well expressed by Zadeh and Pearl (Sect.1, [40], and Sect.6.3).
Let us also note that the undertaken efforts over the last decades for developing AI techniques based on fully automatic learning; representation and processing of concepts are not satisfactory from the point of view of complex real-life projects.
Classical examples can be found in research related to reinforcement learning [44]
and different variants of natural computing [52]. This follows from the difficulty of coping with complexity and diversity of complex vague concepts which should be efficiently learned (discovered), approximated with satisfactory quality, and effi- ciently processed with proper judgement. Searching spaces for discovery of satisfac- tory approximations of such concepts are so huge that existing data mining methods, other AI methods, as well as the current and expected hardware technology do not allow us for effective searching over such spaces in realistic time.
At the end of the previous section, we have mentioned that the dialogue systems, where users will cooperate with computing devices towards solving problems, can be brought into the progress. However, the existing technology is yet not fully satisfac- tory for developing such systems. We expect, that further development of the ideas presented in this chapter, concerning interactive computations based on c-granules, will help in realization of this goal.
4 Interactive Granular Computing (IGrC)
The essence of the proposed approach is to develop BDT based on IGrC [2, 38, 42, 53–55]. In this sense IGrC creates the basis for BDT. The approach is based on foundations for modeling IGrC relevant for BDT in which computations are progressing through interactions [26]. In IGrC interactive computations are per-
formed on c-granules linking, e.g., information granules [17] with spatiotemporal physical objects, called hunks [2,56].
Infogranules are widely discussed in the literature. They can be treated as speci- fications of compound objects which are defined in a hierarchical manner together with descriptions regarding their implementations. Such granules are obtained as the result of information granulation [32]:
Information granulation can be viewed as a human way of achieving data compression and it plays a key role in implementation of the strategy of divide-and-conquer in human problem-solving.
Infogranules belong to those concepts which play the main role in developing foundations of Artificial Intelligence (AI), data mining, and text mining [17]. They grew up as some generalizations from fuzzy set theory, [30,32,33], rough set theory, and interval analysis [17]. In GrC, rough sets, fuzzy sets, and interval analysis are used to deal with vague concepts.
However, the issues related to the interactions of infogranules with the physical world, and their relationship to perception of interactions in the physical world are not well elaborated yet [26,57]. On the other hand, in [58], it is mentioned that:
[...] interaction is a critical issue in the understanding of complex systems of any sorts: as such, it has emerged in several well-established scientific areas other than computer science, like biology, physics, social and organizational sciences.
Computations of agents proceed by interaction with the physical world and they have roots in c-granules [2]. Any c-granule consists of three components, namely soft_suit, link_suit and hard_suit. These components make it possible to incorporate abstract objects as infogranules from the soft_suit as well as physical objects from hard_suit. The link_suit of a given c-granule is used as a kind of c-granule interface for handling interaction between soft_suit and hard_suit (see Fig.2). One can relate this to the statementhttp://www.en.wikipedia.org/wiki/Embodiment:
[...] Embodied agent, in artificial intelligence, an intelligent agent that interacts with the environment through a physical body within that environment
Calculi of c-granules are defined by elementary c-granules (e.g., indiscernibility or similarity classes). Then with the help of the calculi it is possible to generate new c-granules from the already defined ones (see Fig.2, where the presented c-granule produces new output c-granules from the given input c-granules).
The discussed c-granules may represent complex objects. In particular, agents and their societies can be treated as c-granules too. An example of c-granule representing a team of agents is presented in Fig.3 where some guidelines for implementation of AI projects in the form of a cooperation scheme itself among different agents responsible for relevant cooperation areas is illustrated [2]. This cooperation scheme may be treated as a higher level c-granule. We propose to model a complex system as a society of agents.
Moreover, c-granules create the basis for the construction of the agent’s language of communication and the language of evolution. The hierarchy of c-granules is illustrated in Fig.4.
Fig. 2 General structure of a c-granule [36]
Fig. 3 Cooperation scheme of an agent team responsible for relevant competence area [2]
…
meta-agent language & emotion
c-granules society of agents language & emotion
c-granules agent language & emotion
c-granules agent thought & emotion
c-granules
sensor and actuator c-granules
complexity of c-granules
concept hierarchy c-granules
Fig. 4 Hierarchy of c-granules [36]
An agent operates on a local world of c-granules. The control of an agent aims at controlling computations performed on c-granules, from the respective local world of the agent for achieving the target goals. Actions from link_suits of c-granules are used by the agent’s control in exploration and/or exploitation of the environment on the way to achieve their targets. C-granules are also used for representation of perception by agents concerning the interactions with the physical world. Due to the limited ability of agent’s perception usually only a partial information about the interactions of the physical world may be available to the agents. Hence, in particular the results of performed actions by agents cannot be predicted with certainty. For more details on IGrC based on c-granules the readers are referred to [2].
One of the key issues of the approach related to c-granules presented in [2], is a kind of integration between investigations of physical and mental phenomena. This idea of integration follows from the suggestions presented by many scientists. For illustration, let us consider the following two quotations:
As far as the laws of mathematics refer to reality, they are not certain; and as far as they are certain, they do not refer to reality.
—Albert Einstein [59]
Constructing the physical part of the theory and unifying it with the mathematical part should be considered as one of the main goals of statistical learning theory.
—Vladimir Vapnik ([57], p. 721)