Abstract for AISB 2000: How to Design a Functioning Mind
AUTHOR: David W. Glasspool
Imperial Cancer Research Fund and Institute of Cognitive
Neuroscience, University College London
TITLE: The Integration and control of behaviour: Insights
from neuroscience and AI
The integration of a large set of disparate cognitive processes
into a single, coherently acting, purposive agent is an important
problem for the project of creating a functioning mind. Clues to
the way this integration is achieved in the human mind have
emerged from cognitive psychology and neuroscience.
Interestingly, and perhaps tellingly, the picture which is
emerging mirrors solutions (driven primarily by engineering
issues) to similar problems in the rather different domains of
mobile robotics and intelligent agents in AI.
In this talk I shall argue that a dialogue between AI and
neuroscience on the problem of the control and integration of
behaviour will benefit both fields. Many aspects of higher level
control in human cognition remain obscure, although a number of
theories have provided a general outline of the processes
involved. However approaches from AI and robotics can point the
way to the appropriate decomposition of apparently opaque
psychological processes. In turn, the increasingly detailed
picture of human executive function emerging from neuropsychology
can provide both a rich context for and a reality check on
theories of behaviour integration and control in AI.
A number of psychological theories of the higher level control of
behaviour have been proposed. Perhaps the most dominant is Norman
and Shallice's (1986) two-layer framework, informed largely by
evidence from everyday lapses in action control and from
neuropsychological syndromes. The lower level component,
contention scheduling (CS), coordinates the emergence of coherent
purposeful behaviour from a number of independently operating
proto-behaviours, or schemas for action. Behaviour results from
interactions amongst schemas and with the environment, and in
response to goals deliberately imposed by the higher level
component, the supervisory attentional system (SAS). The CS
component is well specified and has been simulated in detail
(Cooper and Shallice, in press). The SAS however remains obscure
and is specified only at a gross level (Shallice and Burgess,
The shadowy nature of the SAS is testament to the difficulty of
"reverse engineering" processes of such scope and complexity in
human psychology. There is, however, a striking similarity
between psychological theories of the Norman and Shallice type
and some theories of action control which have emerged from the
fields of mobile robotics and intelligent agent design. The
notions of the emergence of behaviour from primitive processes
operating in parallel, and of a slower, deliberative system
imposing supervisory control at a higher level of abstraction,
are familiar to roboticists for example. However I shall
illustrate the general point by demonstrating a mapping between
what is known of the Norman and Shallice SAS and a particular
intelligent agent architecture (Das, Fox, Elsdon & Hammond,
1997). The mapping makes possible both an outline specification
and a preliminary computational model for the SAS. The
computational simulation demonstrates the combined SAS-CS system
operating in concert, with a number of independent lower level
processes coordinated in pursuit of a common goal. Even in its
current simplified form the simulation allows predictions to be
made about human executive function.
Cooper, R. & Shallice, T. (In press). Contention Scheduling and
the control of routine activities. Cognitive Neuroscience.
Das S K, Fox J, Elsdon D, Hammond P (1997) A Flexible
architecture for autonomous agents. Journal of Experimental and
Theoretical Artificial Intelligence, 9, 407-440.
Norman, D. A., & Shallice, T. (1986). Attention to action: Willed
and automatic control of behaviour. Reprinted in revised form in
R. J. Davidson, G. E. Schwartz, & D. Shapiro (Eds.) 1986.
Consciousness and self-regulation, Vol. 4 (pp. 1-18). New York:
Shallice, T. and Burgess, P. (1996). The domain of supervisory
processes and temporal organization of behaviour. Philosophical
Transactions of the Royal Society of London B. 351, 1405-1412.
David W. Glasspool
Jan 1993-Feb 1998 University College London
Ph.D. Psychology (Modelling serial order in
Oct 1990 - Dec 1991 University of Manchester
M.Sc. Cognitive Science
Oct 1984 - Jul 1987 University of Durham
B.Sc. Hons. Computing with Electronics, II(i)
Feb 1998-Present Research Fellow, Advanced Computation
Laboratory, Imperial Cancer Research Fund,
Sept 1992 - Feb 1998 Research Fellow, Department of Psychology,
University College London.
1995 - Present Member of the Institute of Cognitive
University College London.
CURRENT RESEARCH AREAS
1. The neuropsychology of human executive function (With T. Shallice,
UCL, R. Cooper, Birkbeck College London, and J. Fox, ICRF).
2. Modelling human reasoning with qualitative logical systems (with J.
3. The neuropsychology of serial behaviour (with T. Shallice, UCL).
RECENT RELEVANT PUBLICATIONS
Glasspool, D. W. & Fox, J. (1999). Understanding probability words by
constructing qualitative mental models. In M. Hahn and S. C. Stoness
(Eds.) Proceedings of the 21st Conference of the Cognitive Science
Society. New Jersey: Lawrence Erlbaum Associates. pp. 185-190.
Glasspool, D. W., Shallice, T. & Cipolotti, L. (1999).
Neuropsychologically plausible sequence generation. In D. Heinke, G. W.
Humphreys & A. Olson (Eds.) Connectionist Models in Cognitive
Neuroscience. London: Springer-Verlag. pp. 40-51.
Glasspool, D. W. & Houghton, G. (1997) Dynamic representation of
structural constraints in models of serial behaviour. In J. Bullinaria,
D. Glasspool & G. Houghton (Eds.) Connectionist Representations.
Proceedings of the 4th Neural Computation and Psychology Workshop.
London: Springer-Verlag. pp. 269-282.
Shallice, T., Glasspool, D., & Houghton, G., (1995). Can
neuropsychological evidence inform connectionist modelling? Analyses
from spelling. Language and Cognitive Processes 10, 195-255.