(Assembled hastily by A.Sloman: to be revised and extended)
Until recently most of the work on combining natural and artificial computation (or intelligence) has focused on trying first to identify the mechanisms used in animals and groups of animals and secondly to implement artificial versions of those mechanisms in order to solve engineering problems. Examples of the mechanisms include artificial neural nets, evolutionary algorithms, swarming and flocking mechanisms, artificial immune systems, and various mechanisms thought to be used in natural vision, language, learning, problem-solving, cooperating, and so on. In July 2005 Members of the CoSy robot project organised a two day tutorial on Representation and Learning in Robots and Animals, http://www.cs.bham.ac.uk/research/projects/cosy/conferences/ which had a different aim, namely getting researchers in animal (including human) behaviours of various kinds to talk to designers of AI systems, not about biological mechanisms, but about some of the interesting kinds of competences that have been found in animals. This tutorial was partly inspired by the UKCRC Grand Challenge No 5: 'Architecture of Brain and Mind', summarised in http://www.cs.bham.ac.uk/research/projects/cogaff/gc/ Since then, partly inspired by that meeting and partly independently, other workshops and conferences have been organised bringing designes of robots and other machines together with researchers on animal cognition, and in some cases, brain researchers. I have the impression that this is a growing phenomenon and will have the very important function of contributing to our understanding of requirements for human-like, or animal-like robots, countering the widespred tendency to assume that we know what is required and merely lack knowledge of how to meet the requirements. A paper written with Ron Chrisley published in 2005 enlarged on this and referred to the difficulty of identifying requirements as 'ontological blindness': http://www.cs.bham.ac.uk/research/projects/cogaff/04.html#cogsys More things than are dreamt of in your biology: Information processing in biologically-inspired robots. The need to identify unobvious requirements, and tendency not to notice them can sometimes lead to grossly inadequate proposals for designs for working systems. It is an interesting fact that when AI researchers hear about an animal able to perform some task that would be commonplace for humans they are sometimes inspired to ask: How could it do that? not noticing that the same question arises for humans. So the growing number of meetings in which researchers on animal and robotic intelligence confront each other may have the important effect of drawing attention to detailed requirements for intelligent robots. See also the euCognition research roadmap activity: http://www.eucognition.org/wiki/index.php?title=Research_Roadmap What follows is an incomplete list of events of the sort described above. If you know of any that should be added to this list, please send me details. -------------------------------------------------------------------
o IJCAI-05 GC5-tutorial in Edinburgh REPRESENTATION AND LEARNING IN ROBOTS AND ANIMALS (funded by BT, IBM, Infermed, SSAISB, organised by CoSy) http://www.cs.bham.ac.uk/research/projects/cosy/conferences/
o AISB'06 GC-5 symposium Bristol April 2006, . (funded by euCognition) http://www.cs.bham.ac.uk/research/projects/cogaff/gc/aisb06/ ================ o AAAI Cognitive Robotics Workshop, Boston July 2006 http://www.aaai.org/Library/Workshops/ws06-03.php (also other things at AAAI'06, including the Fellows' symposium.) ================ o COGRIC workshop funded by NSF and EU, August 2006 http://www.cogric.reading.ac.uk/ ================ o The euCognition research roadmap meeting in Munich in January 2007 was also directly relevant: http://www.eucognition.org/wiki/index.php?title=Roadmap_Kick-off_Meeting Also other euCognition events. ================ o International Symposium on Creating Brain-Like Intelligence at Honda Research Institute Frankfurt, Feb 2007 http://www.hri-europe.de/ Unfortunately presentations and workgroup summaries are password protected.
o. Intelligent and cognitive systems working group at the EU-funded Interlink Opening Workshop Nice, 10-12 May 2007 http://www.ercim.org/interlinkworkshops (wrongly says 2006) http://interlink.ics.forth.gr/central.aspx?sId=84I240I746I323I344319 (Cognition Working group, led by Rüdiger Dillman.) The web site is rather thin: no talk titles, abstracts or presentations. There was an interesting contrast between people who were committed to the key role of embodiment in cognition and could demonstrate only fairly trivial systems, and Jim Crowley, working on an 'intelligent room' including an intelligent system composed of TV camera and projector (plus computer) that could treat a sheet of paper on the table as a screen and track it as it moved round, ensuring that what was projected was suitable for normal interactions, and interpreting hand movements/gestures as equivalent to mouse/keyboard actions. It was crucial to Crowley's system that the location and orientation of the paper and requirements for projecting on to it were represented in terms of location in the room, not in terms of sensory and motor relationships (which could be learnt, or derived, and used). ================ o. BBSRC Workshop organised in Birmingham by Dietmar Heinke Closing the gap between neurophysiology and behaviour: A computational modelling approach 31st May-2nd June http://comp-psych.bham.ac.uk/workshop.htm ================ o. In mid June the final workshop of the EU COSPAL project was held in Aalborg http://www.cospal.org/ Cognitive Systems: Perception, Action, Learning ================ o. On 24th-26th June 2007 there was a NSF/EU-funded workshop on Natural and Artificial Cognition (wonac) in Oxford. Details are here: http://tecolote.isi.edu/~wkerr/wonac/ It was explicitly a sequel to both the GC-5 tutorial at IJCAI in 2005, and the symposium at AISB in 2006 (The 2005 tutorial had been attended by managers from both funding agencies, and the 2006 symposium by the EU funders.) WONAC was all organised by Paul Cohen and Alex Kacelnik, both of whom had been speakers at the IJCAI'05 tutorial. Most of the presentations were excellent (though too short). (Abstracts are online and some of the presentations - select the Programme link on 'agenda' page. Things under review are password-protected unfortunately. Not all the presentations are there yet.) (We learnt that the largest living organism on earth is a fungus.) Jackie Chappell and Aaron Sloman were asked to give the first two talks on understanding causation (Kantian and Humean) and on the altricial precocial distinction (better: nature/nurture tradeoffs). Our contributions, expanded after the workshop, including an unfinished post-workshop analysis of varieties of causal competence are here (all pdf): http://www.cs.bham.ac.uk/research/projects/cogaff/talks/wonac/ [Quite a lot of people not at that meeting are doing related work on learning affordances, e.g. Peter Gorniak MIT, the Stanford 'grasping' group, Yiannis de Miris, Manuela Viezzer, .... But all such systems are still far behind parrots, or human toddlers.] ================ o The recent euCognition meeting on architectures is relevant: 29th June 2007 http://www.eucognition.org/six_monthly_meeting_3.htm Cognitive Architectures Six of the presenters had been involved in the AISB'06 meeting we organised. ================ o A conference with a strong philosophical component in Bristol 1-3 July 2007 (mixed funding, I think.) http://www.bris.ac.uk/philosophy/department/events/PAC_conference/index.html/Conference.htm Perception, Action and Consciousness: Sensorimotor dynamics and dual vision The list of speakers and talks spanning philosophy, psychology, neuroscience and AI is here: http://www.bris.ac.uk/philosophy/department/events/PAC_conference/index.html/Conference.htm/Conference_Programme.pdf This conference attempted to bring together A: people (e.g. Milner and Goodale) who thought there were two main visual streams (ventral and dorsal) only one of which is associated with visual consciousness, though the other is involved in control of action (a distinction that has been very confusingly/inaccurately described in terms of 'what vs where', and 'what vs how', misleading hordes of people who don't read the originals and/or don't think hard about what could actually work.) B: people (e.g. O'Regan and Noë) who argue that perceptual consciousness is very closely related to sensorimotor contingencies, and therefore the action subsystems (dorsal) must be involved. Plus a bunch of others commenting on the disagreement and related issues. A poster presented by A.Sloman, modified after the meeting is here: http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#pac07 Nobody addressed the special problems of perceiving 3-D structures, and understanding the causal relations between shape, spatial relations, materials, and forces that could be applied and physical changes. Consequently their theories did not have the generality they were aiming for. Kevin O'Regan presented his latest theory of colour, developed with Philipona, which he presented as being a theory about sensorimotor contingencies, i.e. how the responses of the colour receptors change for a given illumination profile according to how the illuminated surface is rotated or moved in space. It soon became clear that he needed to distinguish two things: a. colours of surface as objective (exosomatic) properties defined not by sensory/motor relationships, but by illumination/ reflectance relationships (i.e. relations between inputs and outputs of the *surface*, not the organism) b. the ability to learn how to identify the properties of such surfaces by manipulating them in various ways in various illumination conditions. I.e. what we perceive is something out there, and the sensorimotor relationshiops are just part of a measurement device. (Likewise electrical resistance is (roughly) a relation between voltage and current. How you measure or apply voltages or currents is a separate issue: and that can change over time.) That's a very important general distinction, which has been in the background of much of our discussions. It can now be made clear and explicit. Many of the properties of shape, material, and spatial relations will be of type (a), whereas the robot or animal has to use information of type (b) to drive the process of generating and testing explanatory theories and extending the ontology. The sensorimotor model probably fits parrot competences only for fine-grained posture control and low-level visual control of flying, but all their manipulation of 3-D objects uses properties, relationships and causal interactions between objects independently of how they are sensed and altered. The sensorimotor contingencies relevant to particular configurations may need to be learnt at first but may later be derived from an understanding of geometry, motion, etc. ================ o The CoSy project organised a MeetingOfMinds workshop in Paris, 16-18 September (closed unfortunately). The invited speakers, from psychology and biology, the CoSy speakers, and other information can be found here: http://www.cs.bham.ac.uk/research/projects/cosy/conferences/mofm-paris-07/ Papers, presentations and other materials provided after the event can be found here http://www.cs.bham.ac.uk/research/projects/cosy/conferences/mofm-paris-07/latest.html
(To be added.)
School of Computer Science
The University of Birmingham