A Design Theory for Distributed Interactive Systems

Alistair Sutcliffe
Dept of Computation,
UMIST

1. Introduction

This is the HCI grand challenge which was recognised in the HCI sub group’s report
to the CPHC research strategy workshop in January 2000. HCI is multidisciplinary in
nature, hence its challenge concerns the disciplines of psychology and sociology as
well as computer science. HCI is also an engineering design discipline, a perspective
which is shared with software engineering, so the challenge within HCI has
implications for other challenges which may appear in the workshop. The activity of
design is seen as creative and atheoretic by many practitioners but without theory
good design will only spread by examples and hence practice will evolve inefficiently
in small increments. The need for design theory to accelerate progress has been
becoming pressing as the Internet spreads examples of poorly designed user interfaces
throughout the world.

2. Motivation

Whatever the progress Computer Science makes in answering other challenges, to

deliver effective systems into use, and in the majority of cases that means interacting
with people, we need to understand how computational artefacts deliver effective
functionality, interaction and motivation for people. The consequences of not doing so
surround us in everyday systems failure, particularly in government departments.
Some may argue that system failure may be ascribed to ineffective transfer of current
computer science knowledge (viz programming errors in Ariane V, Therac 25);
however, nearly all system failures involve the human element, either in interaction or
in the design process – the London Ambulance Service is the canonical example, the
Cambridge CAPSA system is another case study.

The consequences of inadequate theory-led design of interactive systems and an

engineering process that does not account for the frailties of human reasoning can be
measured in million of pounds. Furthermore, if interactive systems are perceived by
users and policy makers to be poor quality and even worse value for money, they will
inevitably question the value of computer science research and its influence on
commercial systems. Even if we solve the technology transfer problem for other areas
of computer science, failure to solve the HCI challenge will lead to the whole
discipline being tarred with the same brush of the users’ perception of poor quality.
The user interface is, after all, the component they see, feel and interact with.

3. Background

HCI has been aware of the design theory challenge for some time, even since the
conceptual framework of HCI was elaborated by Long and Dowell in 1989. They
argued that HCI needed to advance beyond its current ‘craft’ level of practice, (still
largely the state of practice today) to an engineering approach back by theory and the
sciences of psychology and sociology. Theoretical investigations made a promising
start in the application of ICS psychological theory to design developed by Barnard
and co workers in the Amodeus projects. Unfortunately that momentum has not been
sustained. Progress in the USA and elsewhere has been modest. Preminent among the
cognitive theories is ACT-R from John Anderson (CMU), others of note are EPIC,
and LICAI while GOMS is perhaps the only exemplar of a theory that has achieved
modest application. A major weakness of these theories is their scope of application
which is limited to single person interaction with simple user interfaces. This hardly
scales to the current perspective of collaborative, ubiquitous and mobile technology
where user interfaces are everywhere, sometimes invisible, and support individuals,
groups and communities.

This discrepancy has been recognised in HCI, leading to interest in looser theoretical
frameworks (Activity theory, flow theory); unfortunately such frameworks only offer
interesting concepts that might be applied to design. Theory that explains, and more
importantly, predicts users’ behaviour when interacting with artefacts and each other
is sadly lacking.

4. The challenge- A Design Theory for Distributed Interactive Systems.

This grand challenge comes in two parts:

(i) the first challenge is to create a theory that can predict how people will behave
when interacting with computer enhanced artefacts, in distributed groups.
(ii) the second consequent challenge is to predict the features of the designed artefact
that will influence users’ behaviour to ensure an effective and satisfying experience.

The design challenge is to operationalise the behaviour theory, which will be

grounded in psychology and sociology, to predict the interactive functionality, user
interface features and dialogue that provides an effective and motivating experience
for groups of users in a context. Although it is unreasonable to expect theory to
predict design in detail, theory should influence the design process as precisely as
possible by bridging artefacts (e.g. principles, models, techniques). Establishing the
connection between behaviour theory and the design process is a key part of the
challenge.

This challenge has some history in HCI since the concept of distributed cognition was
launched by Hutchin’s book ‘Cognition in the wild’ in 1995. The problem is complex,
as befits a grand challenge. The components of the challenge are presented first as the
HCI (multidisciplinary) problem, secondly the design implication, and third links
implied to challenges elsewhere in computer science.

1. Modelling the interaction between organisational structures, cultures, goals, tasks

and roles of people within organisations, with the communication channels between
them and computer enhanced artefacts. Organisations vary from highly trained groups
in the military to more loosely coupled groups (e.g. collaboration between the
emergency services) to informal groups (Internet chat, listservs). We have little
understanding about how groups cohere and interact with each other, let alone how
technology can support them. Answers might be found in social group theory,
organisation theory, and policy adoption models.
Design motivation: to improve functional allocation – the effective and appropriate
partitioning of work tasks between people and computers, with decision support for
shared tasks; requirements definition for CSCW systems; dynamic role allocation in
workflow systems; requirements for computer mediated communication.

Challenges elsewhere: in Software engineering applied to CSCW solving shared

awareness, collaborative interaction protocols (being addressed in UK)

2. Theory that predicts the intent and behaviour of individuals acting in groups within
a particular context. This problem involves supporting situation awareness, a
challenge in the Human factors/Safety Engineering area (see also John McDermid’s
grand challenge), awareness of others, managing discourse and planning common
action. Misunderstandings of the intention of others, and especially the opaque intent
of machines, has been responsible for many accidents, particularly with highly
automated systems (e.g. Airbus flight management systems). Clarks’ theory of
common ground is a promising area being explored by some, while the psychology of
perception, comprehension, and group decision making may also contribute useful
leads.

Design Motivation: design of user interface displays and dialogues that convey the
intent, state and action of others (machine and human) in a predictable and transparent
manner. Presentation of contextual information to enhance human decision making,
especially in highly automated, yet safety critical domains.

Challenges elsewhere: Monitoring and interpreting events in real world, provision of

adequate service quality in distributed multimedia communication.

3. Theory that predicts the discourse structure for effective communication between
people and computers, and between people when mediated by computers. We know
from linguistics and psychology that understanding human communication involves
not only the syntax, semantics and pragmatics of discourse, but also non verbal
communication, shared memory, context, social norms and culture. Little of this
theory has been applied to design of human computer (or computer mediated)
communication. Devices are becoming ubiquitous, intelligent and able to converse in
different modalities, including limited natural language, so we need to understand
conversation with computer enhanced artefacts. In this case the theory spawns sub
challenges in acquiring and modelling the interlocutor, and managing conversation to
obey Grician maxims, i.e. be polite, relevant, do not deceive, etc).

Design Motivation: design of multimodal dialogues for human computer interaction,

design of natural language interaction. While open ended NL dialogue will remain
bound by the domain knowledge acquisition/ machine learning bottleneck (e.g. CYC
– a failed grand challenge), restricted NL and speech is a reality so an interested sub
challenge is to integrate NL and user action with complex graphical and tangible
worlds. Design of appropriate conversation initiative between machines and people is
a pressing need as machines become more intelligent and hence become more equal
partners in collaborative tasks.

Challenges elsewhere: The above motivations imply solutions to the user modelling
problem- how a machine acquires an accurate model of the human interlocutor, short
of boring users by continually asking questions. Although the user modelling
challenge resides in HCI, it is a specialisation of a wider machine learning problem,
how to infer properties of an agent give limited meta knowledge, and low level event
stream input (user interaction).