Control Mechanisms for Spatial Knowledge Processing in Cognitive / Intelligent Systems

Program

Symposium Description

Crucial aspects of cognitive systems, may they be robots, (software) agents, or humans are (a) spatial knowledge processing and (b) mechanisms for control of information processing. Without the former, cognitive systems would not be able to act in as well as reason and communicate about a world which is inherently spatial. Without the latter all the reasoning faculties and activities of the system would be employed solitarily leading, in the end, to a failure of the system. Consequently, over the last decade(s), there has been a growing interest in the understanding and realization of both aspects in all three types of cognitive systems. Despite the substantial research effort devoted to control mechanisms and spatial knowledge processing as such, control mechanisms for spatial knowledge processing have virtually been neglected.

Regarding control mechanisms both the field of AI and the field of cognitive science have focused on comparable problems / questions. One important topic, for instance, has been whether and how top-down and bottom-up influences can be and are integrated to achieve control of information processing in artificial and natural cognitive systems. In AI it now seems to be generally accepted that to implement satisfactorily flexible and at the same time intelligent information processing it is necessary to take into account both top-down influences like goals and bottom-up influences like environmental stimuli triggering certain processing steps. Similarly, research in cognitive science has shown that control mechanisms in humans can be conceived as being implemented as the interaction of intentions / goals and environmental influences. A second fundamental question in both fields of research is whether the top-down influences are to be conceptualized as being instantiated by a single component of the cognitive system: in cognitive science as well as in AI some approaches argue for a central controller (i.e., a central executive) whereas others favor the view that control emerges---maybe even heterarchically---from the interplay of several functional components.

With respect to spatial knowledge processing, research has focused on the type of representations employed and the processes working on them. Whereas AI research aims at devising new representations and respective processes to most efficiently reason about some particular spatial problem, cognitive science research tries to reveal and discover the representations utilized in human spatial knowledge processing. Significantly, the representations identified by both strands of research are comparable regarding important characteristics. More precisely, representations for spatial knowledge processing are characterized in both fields as being (a) qualitative (i.e., distinguishing conceptual categories rather than measures), (b) fuzzy / imprecise, and (c) analogous (i.e.,---at least some of---the relations holding between the constituting parts of the representation are analogous to the relations that hold between the entities denoted by those parts). Adhering to these characteristics a number of different types of representations for different kinds of spatial knowledge processing have been proposed in modeling and implementing cognitive systems.

Goals of the Symposium

Although spatial knowledge processing as well as control mechanisms in information processing have thus been considered in close detail, they have been considered only independently of each other. Therefore, results about and conceptions of control mechanisms in spatial knowledge processing are hardly available. For example, at the moment it is unclear how the construction of the spatial representations is controlled in natural cognitive systems and, likewise, how the construction ideally should be controlled in artificial cognitive systems. Moreover, in the light of the numerous different representations proposed so far, the question arises by which control mechanisms the employment of the most suitable representation structure can be achieved. The goal of this symposium is to give first answers to these and related questions by bringing together researchers from AI and cognitive science. Since there is considerable correspondence of AI and cognitive science research regarding control mechanisms and spatial knowledge processing, assembling researchers from both scientific communities will stimulate the emergence of new solutions to the existing problems / questions.

Questions to be considered in talks and discussions include, but are not limited to:

• How is the construction of spatial representations controlled?
• How is processing that makes use of spatial representations controlled?
• How is it decided which spatial representations to construct?
• Given several existing spatial representations, how is the selection of one or more of them for processing controlled?
• Can different spatial representations be combined? How is such a combination controlled?
• Are control mechanisms used in spatial knowledge processing the same as in other domains? If not, what are the differences?
• What are efficient ways to realize control in spatial knowledge processing?

The symposium will be scheduled to provide extensive discussion time and group interactions. There will be a series of presentations with significant question-and-answer time, as well as topic-oriented group discussion sessions.

Submission Information

Please email submissions of 3-6 pages (preferably in AAAI format as PDF) to schulth [at] sfbtr8.uni-bremen.de. Submissions can be position statements, work in progress, or completed work. For general information regarding the AAAI Spring Symposium Series see www.aaai.org/Symposia/Spring/spring-symposia.php.

Organizing Committee

Program Committee