(paper invited by the Conference of Professors
and Heads of Computing
as a contribution to the UK Computing Research
Strategy
CPHC Meeting, Manchester, 6-7 January 2000)
Andrzej Bargiela
Department of Computing, The Nottingham Trent University, Nottingham
NG1 4BU
e-mail: 
;
URL: http://www.intelligentmodelling.org.uk/
The essence of simulation and modelling within the framework
of Experimental Applications is the characterisation of real-life objects
by:
- a set of abstract entities,
- a relationship between these entities and
- a set of unique mappings that give the abstract entities a real-world
interpretation.
In this sense systems modelling is one of the basic knowledge-building
processes and simulation is a knowledge-evaluation technique which
enables inquiry into the possibility of knowledge and the limits of this
possibility.
With the increasingly rapid evolution of science and the globalisation
of economy and society, over the last several decades, there is a marked
shift in our concerns with knowledge, our perceptions of problems and attempts
at their solutions. The knowledge is assessed not only on the basis of
facts or truths but also on the basis of what can and cannot be known and
how we know at all. The problems are abstracted as questions of how to
cope with knowledge (or the lack of it); and the solutions are defined
in terms of decision support, information management and prediction of
system behaviour through balancing large amounts of information coupled
with large amounts of uncertainty. With the ever-increasing complexity
of systems that are of scientific and social relevance, this trend is certain
to continue into the future and the proposed strategic directions for simulation
and modelling research highlight the most promising avenues towards answering
this challenge.
1. Abstraction.
Abstraction forming is a basic scientific tool for coping with complexity
of systems. Typically, complex systems are considered at various degrees
of "resolution" with small-scale models involving detailed mathematical
description (possibly based on the laws of physics) and the progressively
larger-scale models taking into account a decreasing number of degrees
of freedom. There are no general paradigms for appropriate averaging and/or
aggregation to achieve "lower-resolution larger-scale" models. It is usually
the case that the next level mathematical model is determined independently
from the more detailed one. Although, these models frequently capture enough
essential relationships between the entities, the absence of rigorous bridges
connecting them to the more detailed models means that their construction
requires extensive fitting, tedious and expensive compilation of databases
and it results in a limited ability to support ampliative reasoning [3,
7, 9]. Subsumed in these statements is a very difficult scientific challenge.
Consequently, a major research effort is needed in the following directions:
2. Uncertainty processing.
As the systems become more complex, it is increasingly common that
they include human agents as their inherent components. This trend is predicted
to dominate future developments and, as such, it informs the modelling
and simulation research strategy. While the inanimate objects can be adequately
described by reference to the laws of physics (taking into account the
approximations associated with the various levels of abstraction), human
agents can not. This implies that there is a need for a significant strategic
research effort that will put emphasis on the development of modelling
and simulation methodologies that deal explicitly with the presence of
human-induced uncertainty in systems. Apart from its fundamental value
in a full spectrum of industrial applications, such a research will broaden
the formal systems modelling framework to include social, economical and
political systems.
Recent advances in the conceptualization and the measurement of uncertainty demonstrate that, despite the predominance of the probabilistic interpretation of uncertainty, uncertainty should be understood as a multi-dimensional concept. Depending on the mathematical framework employed, uncertainty may be measured in terms of one or more of the five complementary characteristics: - entropy, - dissonance, - confusion, - nonspecificity, and - fuzziness. The judicious use of uncertainty in systems modeling has shown to lead to a reduction of complexity and the enhancement of credibility of models [1, 9, 10, 13-15]. In this context a number of important research directions are proposed:
3. Simulation paradigms and architectures.
3a. Large Scale Adaptive Systems. The experience suggests that many fundamentally important systems are difficult to describe or control using traditional methods. These include natural ecological systems, immune systems, economies and other social systems [8, 12]. One source of difficulty arises from non-linear interactions among system components, which can lead to unanticipated emergent behaviour. A second form of complexity arises from the adaptation i.e. the change of specification, or evolution, of the primitive components of the system. The quantitative, differential equations based, analysis of Complex Adaptive Systems (CAS) inevitably leads to the requirement for the highest-performance computation. The central challenge in such an analysis will continue to be the development of efficient mappings between the problem domain and the terra-scale computer architectures.
A complementary trend in modelling CAS is that of simplifying them to create "artificial worlds" or "agent-based" models. Agent-based models are discrete in most dimensions, typically time, state and update rules and are not intended to capture the quantitative relationships between the system components but rather to support the development of deep intuitions about aspects of CAS [6, 12]. However the full impact of simplifications that are made when designing agents is frequently not understood, so, the corresponding simulation results need to be approached with due caution. The research challenge for the agent-based modelling and simulation paradigm will be to formalise the relationship between the various degrees of simplifications and the system dynamics and to investigate the impact of scaling on the diversity of agents, their mutual interaction and the interaction with system environment.
3b. Standards for global modelling and interoperability. The emergence of standards in distributed communicating objects (e.g. CORBA, Java), distributed simulation (e.g. HLA/RTI) and distributed collaborative modelling (e.g. DEVS/CDM) offers the potential for simulations to be constructed by inter-connecting component models [12, 16]. At the same time, the envisaged change of the perspective on the software design process away from the traditional "finall product engineering" and towards the "managed change" suggests the need for a standardisation of the provision of semantic information in the model interfaces. A co-ordinated research effort is needed to investigate how the enriched semantics of interfaces can assist the component-based modelling, reusability, maintenance and evolution of functionality of global simulations (Web-based/collaborative systems) and also how it can improve the validation and verification of models.
The widespread availability of the Internet together with the availability of mobile communications have created a technological opportunity for the development of large-scale distributed simulations using dynamically changing pools of computing resources. This is frequently referred to as a paradigm shift from distributed to global simulations. Witin this new framework the key issues that will need to be addressed are those of security of transactions, fault tolerance of active computing nodes, management of global computing resources and the communications infrastructure as well as the fundamental issues of formal reasoning about the correctness of agent-based designs.
3c. Simulation and Modelling Interaction. The wide data bandwidth required to communicate the results of complex multi-level simulations motivates an investigation of effective visualisation techniques [5]. While the technological progress will provide an implementational framework, broader research issues, such as what are the tradeoffs between the visualisation detail and the perception and comprehension of simulation results, will need to be addressed. Given the diversity of human agents that are presented with simulation results, an important research challenge will be the development of objective techniques for the assessment of the effectiveness of visualisation.
The integrative role of combined databases and graphical environments, such as offered by the Geographical Information Systems (GIS), is increasingly recognised as a uniquely powerful environment for simulation experimentation with complex systems [16]. The ability to store, manipulate and analyse layers of spatial information that have different resolutions, meets the important requirement of traversing the levels of abstraction of models and the corresponding simulations. The projected research challenges for GIS, in the context of on-line simulations, will include the incorporation of a temporal dimension in GIS data structures and the development of efficient interfaces to distributed simulation processes.
The input side of interaction with simulations is envisaged to assign a growing importance to such modalities as speech, alphanumeric and diagrammatic handwriting and natural language [4]. Future research will need to explore the semantic richness of these media, which add intonation, gestures and context to the standard meaning of the spoken or written words. Central challenges for interpretation include developing representations that enable combining information from different modalities and developing techniques for synchronising different interpretation processes.
The Virtual Reality simulations are arguably the most spectacularly
successful commercial applications of simulation, particularly within the
domain of entertainment. With the advancement of user interface and telecommunications
technologies, that afford increasingly immersive virtual environments,
greater degree of physical realism and the collaborative participation
of users, virtual reality is expanding onto new grounds such as training,
education and virtual manufacturing [5, 12]. It is paradoxical however
that the development of more "natural" and embodied interfaces leads to
"unnatural" adaptations or changes in the user. In the progressively tighter
coupling of user to interface, the user evolves as a cyborg [2]. Rather
than being alarmed by the phenomenon, one should recognise that such a
dilemma was present in the acceptance of the most primitive technologies
such as clothing and work-tools. Consequently, it is envisaged that the
key challenge before the Virtual Reality researchers will be to elucidate
on the "natural" relationship between users and the technology by ballancing
the technological developments with the psychological studies. Some profound
questions about the effect of embodiment on the sensation of physical presence,
social presence, and self-presence in virtual environments and the focus
and stability of our identity will need to be answered.
REFERENCES