Frequently Asked Questions
Research on cognitive systems has its historical roots in early work on AI and cognitive science, but recently has there arisen a need to remind these communities of the goals, priorities, and criteria that led to their initial progress. For this reason, the Fall Symposium on Advances in Cognitive Systems will favor papers on some topics over those on others. Thus, some authors will be uncertain whether their work is appropriate for presentation at the meeting and inclusion in the associated publication. Here we attempt to anticipate and answer questions about the relevance of various submissions.
Generally speaking, the meeting will attempt to foster research carried out in the original spirit of AI, which aimed to design and implement computer programs that exhibited the breadth, generality, and flexibility often observed in human intelligence. We may contrast this approach with the majority of more recent work, which aims to develop highly efficient or very powerful techniques for a narrow class of constrained tasks.
We can clarify this distinction with some illustrative examples. For instance, methods for deep understanding of language are appropriate topics for the meeting, whereas shallow methods for information retrieval and extraction are not. Similarly, mechanisms that exhibit the flexibility and fluidity of human reasoners are welcome, whereas techniques for efficient theorem proving or constraint satisfaction are less relevant. Heuristic methods for learning rich cognitive structures as rapidly as humans are on topic, whereas algorithms that induce classifiers from large data sets or that learn reactive controllers from many trials are not.
Here are some specific questions and answers. Authors should view them as guidelines rather than as strict rules, especially since a paper's position along some dimension that might argue against its relevance may be offset by its position on others that make it highly relevant.
Do papers have to report an implemented computer program?
No. Submissions may analyze a challenging problem, propose a new representation, present an insightful characterization of human intelligence, or provide formal results about an approach.
If a paper describes an implemented program, must it involve the integration of two or more distinct cognitive abilities?
No. Submissions on integrated systems are very welcome, but a paper may focus a single facet of cognition if it discusses its role within a larger cognitive system.
If a paper describes an implemented program, must it present experimental results with quantitative performance metrics?
No. Submissions should make explicit claims and provide evidence to support them, but such evidence may take many forms, including demonstrations of new functionality and well-reasoned, convincing, verbal arguments.
Will you consider papers that describe computational models of human cognition?
Certainly. Research on cognitive systems aims to reproduce the full range of intelligent behavior, and people remain our only examples of broadly intelligent systems. Many key ideas in AI were first presented in models of human cognition, and we hope to publish some papers that continue that tradition.
If a paper describes a computational model of human cognition, must it present comparisons of the model's predictions to quantitative results from psychological experiments?
Definitely not. Submissions may demonstrate that a cognitive system exhibits the same qualitative capabilities as humans under the same qualitative conditions and constraints.
Do papers have to present approaches or mechanisms that are consistent with knowledge of human cognition?
Definitely not. There may well exist approaches to achieving broadly intelligent systems that do not operate in the same manner as humans, although papers that take this stance may benefit from making these differences explicit.
Will you publish papers that describe a new cognitive architecture or substantial extensions or revisions to an existing one?
Certainly. Research on architectures for intelligence is one of the mainstays of cognitive systems. However, papers on new models stated within an existing architecture should emphasize novel insights (e.g., unexpected abilities or limitations) that have resulted from the exercise.
Will you consider papers on integrated approaches to intelligence that do not involve a cognitive architecture?
Definitely. Although cognitive architectures adopt a tightly integrated approach to intelligence, frameworks that utilize less closely linked modules are also legitimate candidates.
Will you publish papers that present theoretical analyses of problems that cognitive systems must address?
Yes. Submissions that report theoretical analyses are just as welcome as ones that describe experiments with computational artifacts. Both have important roles to play in the study of cognitive systems.
Will you consider papers that propose new tasks or problems that challenge the current abilities of cognitive systems?
Certainly. The community needs challenge problems to drive research in new directions, so we encourage submissions on tasks that existing methods cannot handle, provided they describe the problem clearly and propose ways to evaluate progress.
Will you publish papers that survey existing approaches to a class of problems and their limitations?
Yes, but such submissions will be more compelling if they include careful analyses of the underlying reasons for these limitations and propose alternative ways to address the problem class that would not suffer from the same drawbacks.
Will you publish papers that describe logical approaches to cognitive systems?
Yes. Many central insights about cognitive systems have come from the logical tradition, and submissions cast within this framework are welcome, provided they present material in a manner accessible to the broader community.
Will you consider papers that describe probabilistic or statistical approaches to cognitive systems?
Yes. Approaches that use probabilistic or statistical information as annotations to rich cognitive structures are quite appropriate, provided they present material in a way accessible to the broader community. Submissions that describe purely numerical approaches are not appropriate for the meeting.
Will you publish papers that discuss computational methods for learning?
Definitely. However, the most appropriate submissions will describe approaches that acquire rich cognitive structures at roughly the same rates as humans, possibly taking advantage of prior knowledge to constrain the learning task. Papers on purely statistical learning methods are not appropriate for the meeting.
Will you consider submissions that present incremental improvements on existing computational techniques?
Although we hope to encourage innovative research that moves beyond incremental refinements, we will also consider papers that extend and build on existing methods. However, they should do more than demonstrate minor improvements on standard performance metrics, which is the default in the mainstream literature.
Will you publish papers that describe how to achieve an aspect of cognition or intelligence that has not been studied before?
Certainly. We hope to include some papers on understudied topics like creativity, emotion, personality, and metacognition, provided they present a well-specified approach, provide supported claims, and otherwise address the review criteria.
Will you consider submissions that propose a new way to approach an aspect of cognition or intelligence that has been studied?
Yes. We hope the meeting will encourage submissions of this sort, in that the review criteria do not require demonstrating that a novel approach is better than more established approaches along some performance metric.
We encourage authors to consider these guidelines seriously before submitting their papers. If we determine that your topic or approach is not one of those that fall within the scope of the meeting, then we reserve the right to reject it without review. This does not mean that the paper is low quality, but simply that we believe it would not contribute to the goals of the meeting.
If you have any additional questions that are related to the relevance or appropriateness of certain topics, please send email to the Program Chair, Pat Langley (langley@asu.edu).