|
|
UK COMPUTING RESEARCH GRAND CHALLENGES |
|
Machines for intelligent ghosts? |
Grand Challenge 5 (GC-5) Architecture of Brain and Mind Integrating high level cognitive processes with brain mechanisms and functions in a working robot. |
Symposium on AI-Inspired Biology
Held at AISB2010 Convention
This web page is a brief report on the latter.
The first was a tutorial at IJCAI 2005 in Edinburgh. The
second and third were at AISB Conventions, in 2006 and 2010.
The annual AISB Conventions are mainly composed of separately organised
1-Day or 2-Day symposia.
At all three events the majority of talks were by invited speakers or
the organisers.
All have surviving websites with papers and some of the
presentations, hosted at the university of Birmingham.
For full information on GC-5 events since the initial UKCRC GC Conference in November 2002, please see the GC-5 website.
The symposium was mainly concerned with past, ongoing and especially future influences, from AI/robotics and artificial cognition to the study of natural cognition.
This contrasts with research on biologically-inspired AI, which has been a major feature of AI since its inception, especially in the last decade, for instance in the EU Cognitive Systems initiative, begun in 2003 as part of the Framework 6 project, and continued on a much larger scale in Framework 7 (about 400 million euros).
The AIIB call for participation gave particular emphasis to discussion of future prospects, problems, and methodologies for AI-Inspired biology, and the need for collaboration between researchers on natural intelligence (including evolution of diverse forms of intelligence) and artificial intelligence, especially robotics.
More detailed example AIIB topics are listed here.
The AIIB Symposium organisers represented a range of disciplines: Biosciences, AI/Robotics and Philosophy.
There was also a multi-disciplinary panel of reviewers.
In addition to overview presentations by the organisers, there were talks by 8 invited speakers, 7 accepted full papers and a similar number of posters. Prof Margaret Boden was invited to lead the final discussion session (What Next?).
It is intended that speaker presentations will be added to the website, which already includes most of the papers and all the poster abstracts.
Some presentations focused on specific ways in which AI/Robotics could aid, inspire or challenge researchers on natural intelligence, while others addressed more general methodological questions, including questions about limitations of Artificial systems as contributions to the study of natural intelligence.
Attendance fluctuated during the two days, peaking at about 40.
(a) The need for more research on the nature and complexity of structures and processes in the environment that determine evolutionary pressures and also requirements against which both artificial designs and evolved designs need to be evaluated -- e.g. by analysing the problems of perception and action from a robotics viewpoint and by studying examples of convergent cognitive evolution in animals with different morphologies.In all these cases, experience gained in AI/Robotics can be used to suggest new questions to be asked, new ideas for proposed solutions, and in some cases designs that can be implemented to test proposed theories, though some of the issues are well beyond the current state of the art in AI.(b) The need to understand the variety of possible information processing architectures that can integrate diverse, concurrently active, interacting subsystems in natural and artificial systems -- e.g. interactions between perception, motor control, and metacognition, or interactions between visual perception and speech understanding/production subsystems.
(c) The need to understand the varied ontologies, the variety of forms of representation, and mechanisms supporting them, involved in diverse biological processes, including ontologies and representations required for different forms of perception, online control of actions, perception of affordances and other possibilities inherent in the environment, deliberation about possible future sequences of actions, and meta-semantic functions involving representation of cognitive states and processes in oneself and other individuals.
(d) The importance of (amodal) exosomatic ontologies, referring to things in the environment rather than to contents of sensory and motor signals, e.g. for organisms with independently mobile manipulators, and the need to investigate mechanisms for creating and supporting such ontologies.
Examples are ontologies that include strength and compliance of supports of various kinds (e.g. used by orangutans), and different properties of materials composing objects in the environment, e.g. rigidity, elasticity, plasticity, chewability, squishiness, fluidity, etc.
A particular problem is how to represent processes in which all these qualities of materials interact with changing shapes, locations, orientations and trajectories of portions of objects in the environment (including the individual's body-parts).
(e) The potential explanatory role of new computational models of brain-like systems, including causal networks. (Chemical mechanisms were not discussed, but could have been.)
(f) The need to generalise notions of motivation and reward to allow more ways in which biological evolution can produce mechanisms that guide the choices of individuals, on the basis of information that is not available to the individuals but was available during evolution.
Examples include selection by evolution of (a) reward mechanisms and (b) motivations that are triggered by the environment independently of any rewards for the individual.
(g) Ways in which systems, including architectures, can grow themselves under the influence of both the environment and specific genetic mechanisms (some of which do not become active until some time after birth).
Some researchers define "architecture" so as to rule out change, which can block important kinds of research!
(h) the need to improve and extend our ontology for describing complex information-processing systems, including, for instance, improving concepts used for investigating the muddy waters of "emotions" and "consciousness", concepts for describing and comparing architectures, and concepts for thinking about varieties of dynamical system that can play a role in information-processing.
Examples: how to avoid muddles about emotions how to avoid muddles about consciousness.
However this suggestion does not preclude others offering to continue to promote GC-5 as one of the UKCRC Grand challenges.
Last updated: 8 Apr 2010; 11 Apr 2010
Installed: 6 Apr 2010
Maintained by
Aaron Sloman
School of Computer Science
The University of Birmingham