.S 10
.ps +4
.vs +4
.ce 4
.nf
\fBThe Application of
Intelligent Knowledge Based Systems
in
Building Design
.sp 3
.ps +2
.ce
\fICase for Support\fR
.sp 4
.fi
.ps -4
.vs -4
.nr Hu 1
.HU "SUMMARY"
Over the years there have been 
many attempts to introduce
sophisticated computer-based appraisal techniques
to the building design profession.
Most of these foundered due to
the high level of expertise required
from the user in order to apply the packages
and the necessity for the user
to adhere to the appraisal methodology
built into the package's interface.
The proposed research will tackle 
these problems directly,
using some of the techniques
developed by the ikbs 
(intelligent knowledge based systems) community.
The target application chosen is
the field of
dynamic energy modelling,
in particular the ESP package
developed at ABACUS.
However,
by taking full account
of the model-independent Energy Kernel System
currently being proposed by the
energy modelling community,
and by segregating the specific domain knowledge,
the results of the research
will be easily applied to other appraisal packages.
.P
In order to recast the user interface
in the terminology appropriate to the designer,
and still provide the sort of
power and sophistication that current software can deliver,
some form of
.I "intelligent front end"
(ife) is required.
The proposed
.I ife
will contain an encapsulation
of both the energy modelling domain knowledge and
the knowledge required to use the package
in a design environment.
It must be capable of supporting
multiple design methodologies and
must not constrain a designer in any way,
ideally, conversing with him
in much the same way as a consultant
providing a similar appraisal service.
.P
The functions to be handled by the proposed
.I ife
include:
.BL
.LI
Conversing with the user in the appropriate terminology
.LI
Planning how to use the package to achieve the users objectives
.LI
Collecting and organising the description of the building
.LI
Generating the necessary data and control input for the package
.LI
Storing both the domain knowledge and
the strategy knowledge to be used to drive the package.
.LE
.P
As shown in Figure 1,
the system will be built from cooperating modules
organized around a communications module,
the 'blackboard'. 
Although all but the package handling modules 
will build on existing research from related fields,
there are still a number of problems to be overcome,
particularly in getting 
such disparate modules to coexist harmoniously.
.P
The major objective of the proposal
is to provide a clearer understanding
of the principles underlying
.I ife
design and implementation
in the architectural CAD field
It is anticipated that
at the conclusion of the project,
there will exist a prototype system 
which could be used to demonstrate
the potential benefits of applying
intelligent knowledge based techniques
to engineering design problems.
Much of the prototype will be
in a form suitable for inclusion
into the Energy Kernel System,
where it could also act as a testbed for
further exploration of issues such as
user modelling, design methodology, etc,
all of which are vital in providing
adequate support for engineering designers.
.bp
.nr Hu 1
.HU BACKGROUND
.P
Over the years there have been 
many attempts to introduce
sophisticated computer-based appraisal techniques
to the building design profession.
There are two main incentives for 
providing a designer with such appraisal packages.
Firstly, buildings are typically very complex mechanisms
involving many highly technical fields -
for example, movement, energy and cost appraisal, etc.
In these fields,
the manual methods or rules of thumb currently used
cannot cope with the complexity of modern buildings
and only computers can provide the
accuracy and flexibility required.
Secondly, particularly at the earlier stages of the design process,
there is a need for rapid feedback 
as to the consequences of alternative design decisions.
The present system of specialist consultants,
while adequate for the final specification
and detail design
of the appropriate subsystems,
fails to provide this immediate 'ad hoc' advice.
.P
Most of the currently available packages are failing
to be adopted as standard design tools,
mainly because of shortcomings in the user interface.
These shortcomings derive predominately from
the conflict between the necessity for the packages to be
powerful, comprehensive and rigorous 
to deal adequately with the problems,
while, at the same time, to be
simple, straightforward and intuitive
to facilitate user interaction.
This situation is exacerbated by the
divergence of the conceptual framework
of the design oriented user
and the technically oriented developer
of the appraisal package.
To complete this confusion,
there is the subtly different terminology
of the scientific, engineering and design professions.
The current, and laudable, trend
towards 'user friendly' interfaces
carries the risk of negating the
power and flexibility of the actual package
by restricting the interaction
to the 'lowest common denominator' user level.
.P
Appendix A contains a discussion of 
the limitations of existing user interfaces,
in the context of the requirements
of users of building energy appraisal packages.
Two of the major fundamental problems,
the quantity and nature of the data being manipulated
and the expertise and conceptual outlook required of the user,
apply to a greater or lesser extent 
to all appraisal packages.
Although overcomming these will ultimately require
truly intelligent systems,
recent advances in IKBS
(Intelligent Knowledge Based Systems)
and MMI
(Man Machine Interface) techniques
offer some hope of alleviating them.
.nr Hu 1
.HU "Context of Proposal"
The difficulties mentioned above
are exemplified by ESP,
a dynamic energy simulation system
developed by the ABACUS CAD unit at
the University of Strathclyde.
ESP is a large, state-of-the-art package
which predicts the environmental performance
of any proposed or existing building
from its basic thermophysical properties.
As such, it represents one extreme 
of the energy appraisal software,
requiring high levels of expertise
and a familiarity with building thermodynamics.
Thus it is an ideal candidate
with which to explore the application of
IKBS and MMI techniques to its user interface.
.P
Although it is intended to use ESP
as the target appraisal package,
the recent proposal by the energy modelling community
to develop a Energy Kernel System
for developing the next generation of models
offers a mechanism to make the results
of this research available for
a wide range of models.
To achieve this,
the same software methodology
as is being proposed for the Energy Kernel System
will be used in this research
and all knowledge specific to ESP
will be clearly identified and segregated.
.nr Hu 1
.HU "PROPOSED RESEARCH TOPICS"
There are basically two areas whose improvements
would greatly enhance the usefulness of appraisal packages,
such as ESP,
as a design tool.
.P
The first is the level of expertise
required to drive the system.
In order to select only the necessary and sufficient 
input data and to interpret the output data,
requires a lot of expertise in the energy modelling field.
Some means of reducing this level of knowledge
to that of the normal designer
would have enormous impact on the uptake
of energy modelling by the design profession,
and hence on the quality of the built environment.
This could be achieved by an Intelligent Front End,
to reduce both the quantity of data required
from the user,
and the number of errors that get through
to the simulation stage.
The
.I ife
would act as a consultant to the user,
taking what data the user offers,
conducting a dialogue to gather further information
and identify the user's objectives,
and then performing the simulation.
The system would contain a model of the user,
that is information about his level of competence,
modus operandi, working terminology, etc.
This, updated during the course of the session,
would be used to restrict
any querying of the user to the appropriate level.
To supplement the data gained from the user,
the system would use a knowledge base 
of energy modelling data and relationships
supplied by an expert in the field.
It could be tailored,
firstly during installation for the 
practice in which it will be used,
and then, perhaps automatically,
for the individual user.
.P
The second, related, potential area for improvement
is in the manner of communication with the user.
The current generation of small but powerful workstations
makes it possible to
apply the latest MMI techniques
to provide users
with the powerful interaction tools
that current appraisal packages demand. 
One such technique is multi-windowing,
where multiple outputs are displayed on the screen
in a manner analogous to sheets of paper on a desk.
It allows the user to use the machine
for other, possibly unrelated, tasks
without interfering with the dialogue
from the appraisal package,
allowing files/databases to be consulted,
subsidiary analysis to be carried out,
side issues to be investigated, etc.
Other techniques include
the use of spreadsheet-like input incorporating
dynamic (intelligent?) defaults,
graphical feedback of the current state
of the building definition
and animation of the energy flows around the building.
.nr Hu 1
.HU "OVERVIEW OF SYSTEM"
This proposal is to construct an
.I "intelligent front end"
for the building design/modelling package, ESP.
The concepts underlying
.I ife s
are discussed in more detail in Appendix B.
In spite of the recent surge of interest,
there is still rather little
research being carried out in the area,
though related, enabling fields seem to be
producing usable results.
Accordingly, it is proposed to
develop an
.I ife
built mainly on the existing output
from this research.
One objective is
to investigate the problems 
connected with 
.I ife
design and implementation,
particularly with regard to the demands placed on the sub-systems.
The second objective is to produce a prototype
intelligent interface that could encourage
the uptake of routine building performance appraisal by designers.
The prototype could also provide
a testbed for further research into
such issues as user modelling, design methodologies,
building description,
performance assessment methodologies, etc.
More generally,
the results of the research should be useful to
the application of IKBS techniques 
to wider engineering design problems.
It is important that this prototype
is structured in such a fashion
that it can easily be assimilated
into the proposed Energy Kernel System.
To achieve this,
a large measure of independence
from specific packages and/or
design methodologies is required.
.nr Hu 2
.HU "Outline of proposed system"
The previous section indicated
the tasks to be handled by the
.I ife .
The proposed system, see \fBFigure 1.\fP,
can be considered as consisting of the following functions:
.BL
.LI
User modelling: to track the users mental design process
and ensure the system responds appropriately.
.LI
User dialogue handling: to converse with the user 
in a suitable (eventually natural) language,
making maximum use of current graphical MMI techniques.
.LI
Plan recognition & Planning: to identify
the users objectives, ie plans,
and decide on the most appropriate simulation scheme.
.LI
Building modelling: to organize the data describing 
the building under consideration.
.LI
Back-end handling: to drive the package, in this case ESP,
and feed it the necessary data.
.LI
Knowledge base: to contain the necessary strategy knowledge
about designing and about using ESP as well as
the energy modelling domain knowledge.
.LE
.so ife.dg1
.P
It should be appreciated that
the construction of a system
embodying all of these components
at a state-of-the-art level
would be a mammoth task.
Accordingly, it is proposed to concentrate
on the creation of an adequate framework,
with the supporting modules being fleshed out
later in the project.
One component of the project will focus on
designing a flexible and powerful overall structure
and implementing the Blackboard and Knowledge Base.
The second component will focus on the remaining modules,
commencing with the Dialog and Back-end Handlers.
.P
By far the most interesting of the modules described above
is User Modelling.
Unfortunately,
modelling users conceptualizations is very difficult,
mainly because the mental processes concerned
are not well understood,
but also because identifying internal concepts
from external behaviour is error-prone even for humans -
eg. decide whether someone views a building as
an envelope subdivided into rooms
or an aggregation of rooms.
A further problem arises when the user switches viewpoints,
or operates with two different, but complementary conceptualizations.
For these reasons,
user modelling is seen by the Artificial intelligence community
as one of its most challenging research topics.
Therefore, rather than jeopardize the production of the prototype,
user modelling will be treated as a sub-project
to be explored in parallel with,
but independently from,
the rest of the system modules
and only a simplified version will be built into the prototype.
.nr Hu 1 
.HU "SYSTEM DESCRIPTION"
This functional division leads to the notion 
of implementing the system as a suite
of independent modules communicating via
the knowledge base, ie a sort of 'blackboard' model.
The major advantage of this approach
is that, after specification of the knowledge base,
the development of the other modules
can proceed independently.
This will largely prevent implementational difficulties
arising in one module affecting the rest of the system.
It will also allow the modules
to be carried through at different levels,
eg sophisticated user dialogue handling but rudimentary planning.
This is more attractive given that
several of these modules have already been investigated
by other researchers in related fields,
see Appendix B,
so that, building on this established base,
it should be possible to get a prototype
working in a reasonable timescale.
It also fits in well with the
software development methodology
currently being proposed by the energy modelling community
for the new Energy Kernel System.
This, together with the isolation of
design and domain knowledge into the knowledge base,
opens up the possibility of using the
.I ife
with other appraisal packages and design methodologies.
.nr Hu 3
.HU "Blackboard:"
This module is the communications center of the system.
The other modules examine the blackboard
for information they can use,
and post their results back on it.
This scheme facilitates multiple use of information,
eg one user statement posted by the dialogue handler
may be useful, simultaneously, for
the user model, knowledge base, planner and building model.
Although some of the newer shells
becoming available in the USA
may be suitable,
It is anticipated that the system
would have to be implemented from scratch,
probably in Prolog.
.HU "Knowledge Base:"
The crucial area is the knowledge base,
upon which everything else hangs,
and the design/implementation of this constitutes
the major part of the work proposed.
This will contain the knowledge about
the application (design);
the domain (energy modelling);
the tools (ESP);
and the user,
all stored explicitly and independently
to facilitate modification or replacement.
At the moment,
it is intended to base this
on the conceptual graph approach of Sowa.
.HU "Dialogue Handler:"
Initially this will deal with a restricted command language, designed
to give the user the opportunity
to volunteer information,
abort or redirect the system's line of inquiry
and give "I don't know" replies.
Although classified as dialogue,
this module will handle non-textual input as well
to allow the latest MMI techniques to be employed.
It is intended to use a state transition network
to control the dialogue,
thus providing maximum flexibility
and facilitating dialogue control by the user model.
Although it is 
not
proposed to do so under this grant,
this could eventually be extended to handle
a subset of English
by borrowing techniques from 
research into natural language,
and the
.I ife
will be structured accordingly.
.HU "User Model:"
The user model,
in the context of this
.I ife ,
is the computers internal representation
of the users's mental processes,
ie. the computers understanding of what the user is thinking,
based on knowledge of the user and on
his past and present actions/statements.
This is discussed in more detail in Appendix B.
However, as discussed above,
it is proposed to restrict this module initially
to classifying the user in one
of a small number of categories,
and controlling the user dialogue accordingly.
For example, an 'architect' 
at an 'early design' stage would not be asked
to provide information about the plant control strategy,
whereas an 'energy modeller' would be expected
to provide information about the
desired timestep control mechanism.
The decision will be based on a user database,
initialised by querying the user,
and modified should the dialogue appear to be breaking down,
eg. the user is giving either too many 'don't know' answers
or a lot of unsolicited information at an inappropriate level.
Later in the project
this will be expanded to allow for 
varying levels of user competence
in different aspects of the appraisal process.
Independently of this pragmatic development,
the more idealized version of a user model
will be researched.
.HU "Plan Recognition:"
Since a prerequisite for this module
is a comprehensive user model,
no attempt will be made to address this issue
at this time.
The user will be required to state his objectives
in the predefined terms that the planner requires,
eg overheating analysis, solar gain analysis, etc.
However, when the dialogue handler is extended
to allow natural language,
then some form of plan recognition would become essential,
and therefore the system,
and in particular the planner,
will be constructed such that
an appropriate module can be added.
.HU "Planning:"
In order to avoid
major modifications to ESP,
planning will initially be restricted to 
selection of one out of a number of predefined processes .
For example,
if the user indicates a desire
to investigate the possibility of overheating,
the planner will initiate the relevant simulations
and displays the necessary output results
in separate windows.
As the research develops and the knowledge base expands,
the amount of predefinition will be reduced
until the planner is actually using simulation primitives
and modelling knowledge from the Knowledge Base.
It is at this stage that the prototype
could be moved into the proposed Energy Kernel System
and the requirement for ESP as the Back-end removed.
.HU "Building Model:"
This module will depend rather critically
on the design of the knowledge base module.
At present, it is envisaged as 
a straightforward database,
containing the building's
geometry, construction, occupancy and system data
as input by the user or as supplied by the knowledge base.
This will adhere to the current input formats
as required by ESP.
Ultimately,
this facility will be provided as part of the Energy Kernel System.
.HU "Back-end Handling:"
As indicated under the planning module,
the emphasis here is to minimize the 
changes in ESP.
Initially this will simply extract the data
from the building model and planner,
creating the data and control files to drive ESP
in a 'batch' mode.
This module will be the only one
specifically tailored to ESP,
so that the lessons learned from constructing the system
would be more easily generalized for wider 
engineering design applications.
At the end of the project,
this back-end will be reformulated
to fit into the Energy Kernel System.
.P
It should be clear from the above that
in order to produce a working system,
a number of research-level problems have to be overcome.
However, given the modular nature of the proposed system
and the initially restricted scope of
some of these modules,
it should be possible to have a working prototype
within the timescale of the project.
Although the main work,
and probably the prototype system,
will have to be done
before the Energy Kernel System becomes available
it is intended that software development methodology
used by the Kernel will be adopted
so that the results of this research
can be easily integrated with other Kernel developments.
.nr Hu 1
.HU "RELATED WORK"
There has been rather little published in
the field of
.I "intelligent front ends"
as such.
Most of the basic work seems to have been carried out
under the cognitive science, mmi and intelligent tutoring banners.
Recently there has been a surge of interest in
.I ife "s"
and the Alvey IKBS Directorate has set up
a research theme in the area and run two workshops.
Most of the work reported was on the periphery,
but two relevant research projects running
are
.BL
.LI
Edinburgh: an
.I ife
to an ecological modelling expert system.\*([.Uschold84\*(.]
.ds [F Uschold84
.]-
.ds [T An Expert System for Ecological Modelling
.ds [A Uschold, M
.ds [B Alvey IKBS Research Theme Workshop: Intelligent Front Ends 1
.as [B " Cosners House, Abingdon, England  26-27 September 1983
.ds [E A Bundy
.ds [I IEE
.ds [C Hitchin, Herts. SG5 1SA, UK.
.ds [D 1984
.ds [P 9
.nr [P 0
.ds [K IFE  intermediate representation
.nr [T 0
.nr [A 0
.nr [O 0
.][ 3 article-in-book
.LI
Imperial College/NAG: an
.I ife
to a statistical package, GLIM.\*([.Nelder84\*(.]
.ds [F Nelder84
.]-
.ds [T An IFE for GLIM
.ds [A Nelder, J
.ds [P 12
.nr [P 0
.ds [B Alvey IKBS Research Theme Workshop: Intelligent Front Ends 1,
.as [B " Cosners House, Abingdon, England  26-27 September 1983
.ds [E A Bundy
.ds [I IEE
.ds [C Hitchin, Herts. SG5 1SA, UK.
.ds [D 1984
.ds [K IFE  statistical assistant
.nr [T 0
.nr [A 0
.nr [O 0
.][ 3 article-in-book
.LE
.P
There has also been a similar workshop held
at the AAAI-86 conference.\*([.Neches86\*(.]
.ds [F Neches86
.]-
.ds [T AAAI-86 Workshop on Intelligence in Interfaces -
.as [T Philadelphia, USA, 14 August 1986
.ds [A Neches, R.
.as [A " and Kaczmarek, T. eds.,
.ds [R "
.ds [I Information Sciences Institute, University of California
.ds [C Marina del Rey, California, USA
.ds [D 1986
.ds [K IFE  research summaries
.nr [T 0
.nr [A 0
.nr [O 0
.][ 3 book
One other directly relevant piece of work is:
.BL
.LI
Stanford HPP: SACON, a knowledge based consultant for structural analysis.\*([.Bennet79\*(.]
.ds [F Bennet79
.]-
.ds [T SACON: A Knowledge Based Consultant for Structural Analysis
.ds [A Bennet, J S
.as [A " and Englemore, R
.ds [J Proc. IJCAI-79
.ds [C Tokyo, Japan
.ds [I International Joint Conference on Artificial Intelligence
.ds [D 1979
.ds [P 47-49
.nr [P 1
.ds [K IFE  production rules MYCIN
.nr [T 0
.nr [A 0
.nr [O 0
.][ 1 journal-article
.LE
.P
Most of the literature is rather general and speculative\*([.Bundy84,\|O'Keefe84\*(.]
.ds [F Bundy84
.]-
.ds [T An Architecture for Intelligent Front Ends
.ds [A Bundy, A
.ds [B Alvey IKBS Research Theme Workshop: Intelligent Front Ends 2
.as [B " University of Sussex, UK  10-1 July 1984
.ds [E A Bundy
.ds [I IEE
.ds [C Hitchin, Herts. SG5 1SA, UK.
.ds [D 1984
.ds [K IFE
.nr [T 0
.nr [A 0
.nr [O 0
.][ 3 article-in-book
.ds [F O'Keefe84
.]-
.ds [T Towards a Statistical Assistant
.ds [A O'Keefe, R
.ds [B Alvey IKBS Research Theme Workshop: Intelligent Front Ends 1
.as [B " Cosners House, Abingdon, England  26-27 September 1983
.ds [E A Bundy
.ds [I IEE
.ds [C Hitchin, Herts. SG5 1SA, UK.
.ds [D 1984
.ds [P 10
.nr [P 0
.ds [K IFE  mbase
.nr [T 0
.nr [A 0
.nr [O 0
.][ 3 article-in-book
with little hard information.
.nr Hu 1
.HU "RESEARCH EXPERIENCE"
The proposed research will require a blend of expertise 
in both the application area,
advanced energy simulation,
and the ikbs field.
The two cooperating groups will bring
the necessary expertise to the project.
.nr Hu 3
.HU "IKBS, Rutherford Appleton Lab:    "
Mr Damian Mac Randal 
is currently a member of the Intelligent Knowledge Based Systems section
at the Lab.
Prior to joining the group,
he was an active researcher
in the field of Computer Aided Building Design, CABD,
and has worked at ABACUS on the ESP system.
The IKBS group at RAL was created in response to the Alvey programme's
initiative in ikbs and
is widely involved with most aspects of artificial intelligence.
It currently provides technical support for Alvey research groups
and runs an artificial intelligence support group for
the SERC Environment Committee.
Its position in RAL gives it access to 
a large pool of experience in the computing field,
and in particular the MMI area.
.HU "ABACUS, Strathclyde University:    "
Professor Tom Maver is the director of this unit and
has an established international reputation
in the field of CABD.
He is actively involved with research into,
and the teaching of, CABD in the design profession and
is the author of a book on the subject.
Dr. J. Clark is a SERC Advanced Fellow, and
has been at the forefront of energy modelling research
for a number of years.
He is the progenitor of the ESP system and
has is the author of a book on energy simulation in building design.
The ABACUS unit has been in existence since the early seventies and 
has carried out research spanning the complete spectrum
of building cost/performance appraisal.
These activities have resulted in several sophisticated
simulation packages which are well established in
certain sections of design practice.
The proposed 
.I ife
research extends its existing concern with making
such sophisticated computer-based appraisal packages
more accessible to the design profession.
.nr Hu 1
.HU   "REQUIREMENTS"
Briefly, funding is requested for:
.BL
.LI
One RA for 1 year to work in the IKBS unit at RAL
and one RA for 2 years to work at ABACUS.
.LI
Travel funding is requested for 
(1) coordination of the work at the two sites,
(2) visits to groups working either in the 
.I ife
field, or in the related fields,
(3) and attendance at the major conferences,
both to maintain good contacts with other workers in the field
and to disseminate the results arising from the research.
.LI
No further capital equipment is required
as existing workstations at each unit are adequate and sufficient.
However, there will be some requirement at Strathclyde for
the maintenance of these workstations
and the consumables associated with them.
.LE
.P
A detailed breakdown of the resources required
is given in Appendix C.
.nr Hu 1
.HU "EXPECTED BENEFITS"
It has been found that,
for all but the most knowledgeable and experienced designers, 
assistance with current appraisal packages
is essential.
One of the advantages of an
.I ife
in the field of energy modelling
is that, for typical designers,
any help at all
in dealing with the varying 
scientific, engineering and design vocabularies
would be very valuable.
It would substantially increase the uptake
of sophisticated appraisal packages.
This will help improve the quality of the built environment
by allowing the designer to eliminate
poor performance features at the design stage.
.P
The main outcome of the proposed research
will be a clearer understanding of
.I ife 
design and implementation principles.
This should allow similar interfaces
to other existing packages to be constructed quickly
and opens up the possibility of
truly integrated multi-criteria design appraisal.
This would allow the designer
to more easily handle the various trade-offs
necessary in modern buildings.
.P
It is also anticipated that the prototype system
would be sufficiently robust to allow
its impact on the design profession
from the user interface point of view,
to be evaluated.
As well as giving useful feedback
about the prototype system,
it would provide a testbed
for further research into user modelling
and design methodology.
.P
One of the spin-offs from this research
will be a number of small, useful user interface 'objects'
for inclusion in the Energy Kernel System.
This is seen as the best method
of encouraging the uptake
of the results of this research 
by other workers in the energy modelling community.
.bp
.nr Hu 1
.HU "APPENDIX A   Limitations of Existing User Interfaces"
If sophisticated appraisal systems
are to be accepted and used by designers,
they will have to be 
\s-2(1)\s+2 available on affordable machines
\s-2(2)\s+2 easy to use
\s-2(3)\s+2 and operating at a technical level commensurate
with the designers knowledge.
The first point is rapidly being solved with the development,
and continuing enhancement,
of workstations like the Sun and Perq.
The other two points address Man-Machine Interface
(MMI) and IKBS considerations respectively.
.P
Before dealing with the details of the user interface,
the various classes of users,
and the facilities they require,
have to be identified.
In the field of building energy simulation,
experience to date 
tends to classify users into two categories,
the 'expert' and the 'novice'.
.VL 7 
.LI expert:
generally computer and energy literate,
spending a lot of time using the system
either as a research tool into energy modelling
or as a design tool investigating building behaviour.
.LI novice:
generally using the computer and/or package infrequently
either for appraisal of a building's performance or
to better understand the effect of a particular design decision.
Note that the novice is usually
an expert in his own 'design' field ,
the term 'novice' being applied only in the context of energy modelling.
.LE
.P
The requirements of these two classes of user differ substantially.
The expert, as usual,
is concerned with speed and flexibility of input,
direct control of the operation of each module of the system
and access to all the resultant output
in a structured but flexible manner.
The novice, also as usual,
wants a clear and coherent interface
where the system provides guidance
as to the current options and their implications,
error trapping and easy error recovery,
and a concise summary of the results
in an easily understood manner.
.P
As well as distinct classes of user,
there are 3 distinct facets
of the appraisal process,
each raising its own set of problems
for the user.
These are 'data input', 'simulation control'
and 'output analysis'.
.VL 7
.LI input:
One of the basic difficulties
facing current designers is
the sheer quantity of data
required to describe/manipulate 
a computer based building model.
Not only is gathering this data a time consuming task,
but frequently the data has not yet been specified,
as is the case at the early design stages.
Also, due to the complex interrelationships,
ensuring the integrity of the data
can demand very high levels of understanding
of the underlying thermodynamics and
the simulation techniques.
.P
This creates two problems for the user.
Firstly, if the data requested is not available,
no help is provided to generate a sensible default.
Secondly, without a good knowledge of
the simulation mechanism,
the importance, and hence the required accuracy,
of an individual piece of data
is very difficult to judge.
.P
As well as the question of 'what',
there is also the problem of 'how'
to input such a large quantity of 
highly inter-related data.
The various factors associated with
data acquisition,
together with the users'
often idiosyncratic conceptualization
of its inter-relationships,
tend to conflict with the rigid 
question/answer style of input
common to many of today's programs.
.LI control:
Generally, control of the appraisal
does not require much sophisticated
user interaction.
At this stage the major difficulty
faced by the user is the selection
of the simulation parameters to
produce a sufficient quality and quantity of output
to allow a worthwhile appraisal of the building -
that is, what is the most meaningful performance assessment
methodology.
For the novice, a lack of understanding
of the implications of the selections being made
can lead to confusion, or even erroneous deductions,
due to the inadequacy of the output data.
.LI output:
It is here that the requirements of the novice and expert differ most.
The expert will be trying to detect
patterns in, and relationships between,
different building parameters,
in order to build up a picture
of the dominant energy flowpaths.
To do this,
all the data generated by the simulation
has to be available and capable of
being displayed in juxtaposition with any other data.
The novice, on the other hand,
merely wishes for a concise summary
of the building performance,
preferably in terms of those variables
most meaningful to the design team and client.
Unfortunately, due to the primitive nature of the system's output,
the novice may experience difficulty
in relating poor performance 
to the design decisions that caused the problem,
or, indeed, even to the possible design modifications 
that could improve this performance.
.pl 11.95i
.nr L 11.95i
.bp
.nr Hu 1
.HU "APPENDIX B   Intelligent Front Ends"
In the above section,
the idea of an 
.I "intelligent front end"
was introduced.
One of the main conclusions so far from research into 
.I ife "'s
for other systems
is that they are an intricate synthesis of
user model, mmi techniques, contextual knowledge 
and the back-end interface.
The limited experience to date has generated 
a number of tentative guidelines, 
the main ones being:
.BL
.LI
In order to be of much assistance,
a front-end must be able to interact at the user's level.
This implies that it should have a model of the
user's knowledge and problem solving strategies
in order to extract all the implicit information
and unstated assumptions from the users statements.
Considering the misunderstandings that can arise
in man-man communication,
the difficulty of this task will be appreciated.
However, another, more tractable, function of the user model
is to track the users mental model
of the system and react accordingly.
This, of course, must adapt
as the user progress from novice to expert.\*([.Ross84\*(.]
.ds [F Ross84
.]-
.ds [T The Virtues and Problems of User Modelling
.ds [A Ross, P.
.ds [B Colloquium on Intelligent Knowledge Based Systems -
.as [B " the Path to User Friendly Computers
.ds [I Institute of Electrical Engineers, Savoy Place, London WC2R 0BL
.ds [R Digest 1984/104
.ds [D Dec 1984
.ds [P 5/1
.nr [P 0
.ds [K IFE  user modelling  intelligent tutoring  blackboards
.nr [T 0
.nr [A 1
.nr [O 0
.][ 3 article-in-book
However, most of the user's progress comes through
building this mental model of the complete system:
back-end, front-end and even the hardware.
Therefore a good front-end must be careful 
not to hinder or sidetrack this model building
by changing the characteristics of the interface
as the session proceeds.
Instead, it should be capable of hastening the process
by firstly hiding itself, the hardware and
any other situational aspect not directly relevant
to the user's task,
and, secondly, by presenting the underlying concepts
of the back-end in an easily assimilable form.
Where the user's model of the back-end system is
deficient or wrong it should take the appropriate steps
to counteract or remedy his misconceptions.
At this level,
the
.I ife
is encroaching on the
field of intelligent tutoring.
.LI
The context in which the system is being used
and the underlying motivation of the user
must be taken into account.\*([.Kidd84\*(.]
.ds [F Kidd84
.]-
.ds [T Man Machine Interface for an Expert System
.ds [A Kidd, A L
.as [A " and Cooper, M B
.ds [B Expert Systems 83, Proc. 3rd BCS Conf. on Expert Systems,
.as [B " Cambridge  UK,  Dec 1983
.ds [P 8-14
.nr [P 1
.ds [I British Computer Society, 13 Mansfield St., London W1, UK
.ds [D 1984
.ds [K ES  al/x  knowledge acquisition  graphical user interface  explanation
.nr [T 0
.nr [A 0
.nr [O 0
.][ 3 article-in-book
./	ds [F Moralee84
./	]-
./	ds [T Intelligent Front Ends
./	ds [A Moralee, S
./	ds [B Alvey IKBS Research Theme Workshop: Intelligent Front Ends 1
./	as [B " Cosners House, Abingdon, England  26-27 September 1983
./	ds [E A Bundy
./	ds [I IEE
./	ds [C Hitchin, Herts. SG5 1SA, UK.
./	ds [D 1984
./	ds [P 6
./	nr [P 0
./	ds [K IFE
./	nr [T 0
./	nr [A 0
./	nr [O 0
./	][ 3 article-in-book
Evidently, the type of help proffered by the 
.I ife
will be qualitatively different if the user
is a student in an School of Engineering,
a technician in an architectural practice
or a researcher in energy simulation.
In the student case,
the emphasis would probably be
on teaching building thermodynamics
as an intelligent 'tutor';
in the technician case,
the emphasis would be on presenting
the system as an friendly? 'consultant';
while in the research context,
the requirement is for a 'colleague'
looking over your shoulder to help spot errors.
These different scenario are sufficiently distinct,
that it would be preferable to deal with them
by using different front ends,
specialized for the context in which they are to be used.
However, even in the more limited scenarios,
the
.I ife
should be able to establish the user's objectives
and generate a suitable problem-solving strategy
without the user having to spell out
every step of the process.
.LI
One of the things which distinguish an 
.I ife
from other pieces of software
is the need to interface to a back-end.
This requires it to perform the mapping
from the user's model of the back-end
to the back-end's model of the user.\*([.Bundy84\*(.]
.ds [F Bundy84
.]-
.ds [T An Architecture for Intelligent Front Ends
.ds [A Bundy, A
.ds [B Alvey IKBS Research Theme Workshop: Intelligent Front Ends 2
.as [B " University of Sussex, UK  10-1 July 1984
.ds [E A Bundy
.ds [I IEE
.ds [C Hitchin, Herts. SG5 1SA, UK.
.ds [D 1984
.ds [K IFE
.nr [T 0
.nr [A 0
.nr [O 0
.][ 3 article-in-book
The interface to the user is a fairly conventional,
if difficult, application of established mmi techniques.
However, to interface to a back-end
requires a lot of knowledge about its function and operation.
At the extreme,
the 
.I ife
has to know as much about the
methods of solving the user's problem
as the back-end.
Abstracting this knowledge can be as difficult
as deriving an expert system to replace the back-end.
.LI
Not only does the 
.I ife
need to have a
good interface to the user,
it is itself part of the user interface to the back-end.
Hence, it must take account of
the usual mmi criteria,
and support a sufficiently rich dialogue
to avoid constraining the user's
problem-solving activity.
.LE
.P
Several of the components of
.I ife "'s"
have already been investigated,
either in their own right
or as part of other research,
eg. plan formation,\*([.Tate79\*(.]
dialogue handling,\*([.Alty84\*(.]
.ds [F Alty84
.]-
.ds [T Use of Path Algebras in an Interactive Adaptive Dialogue System
.ds [A Alty, J L
.ds [B Alvey IKBS Research Theme Workshop: Intelligent Front Ends 2,
.as [B " University of Sussex, UK  10-1 July 1984
.ds [E A Bundy
.ds [I IEE
.ds [C Hitchin, Herts. SG5 1SA, UK.
.ds [D 1984
.ds [K MMI  CONNECT  interface  transition network  production rules
.nr [T 0
.nr [A 0
.nr [O 0
.][ 3 article-in-book
.ds [F Tate79
.]-
.ds [T Generating Project Networks
.ds [A Tate, A.
.ds [J Proc. IJCAI-77,
.as [J " Cambridge, Mass. USA
.ds [I International Joint Conference on Artificial Intelligence
.ds [D 1979
.ds [P 888-893
.nr [P 1
.nr [T 0
.nr [A 1
.nr [O 0
.][ 1 journal-article
user modelling in tutoring systems.
Although, obviously,
there still remains a lot of fundamental research
to be carried out in these fields,
the results already achieved appear
to provide the basic building blocks for an
.I ife .

        Damian Mac Randal              Royal Snail:  Informatics Div,
                                                     Rutherford Appleton Lab,
 JANET:  damian@vd.rl.ac.uk                          Chilton,    Didcot,
  ARPA:  @nss.cs.ucl.ac.uk:damian@vd.rl.ac.uk        Oxon.  OX11 0QX,    U.K.
BITNET:  damian%vd.rl.ac.uk@ukacrl.bitnet      fax:  445808  telx: 83159
  UUCP:  {....!mcvax}!ukc!rlvd!damian        phone:  +44 235 21900 x5160

