Research Roadmap Plan C: The Learning Journey Workshops

Extracting Requirements of a Cognitive Architecture from research in human cognitive development (Peter Ford Dominey)
The Role of Information in Cognition (Jürgen Jost)
Mechanisms of coordination in a cognitive system (Christoph von der Malsburg and Andreas Engel)
Applications, Requirements, Capabilities (Patrick Courtney, Bill Sharpe, and Erik Hollnagel)
Attention, Vision, Robotics, and Cognitive Systems (Markus Vincze)
Affect and Emotion in Cognition (Tom Ziemke)
Cognition and Culture: the Enactive Approach (Bill Sharpe & Fred Cummins)

Learning Journey: Extracting Requirements of a Cognitive Architecture

Venue: Lyons
Date: October 27, 2007
Participants: Giorgio Metta, Patrick Courtney, Peter Ford Dominey, Felix Warneken, Collin Bannard
Organized by: Peter Ford Dominey
Report written by: Peter Ford Dominey

The workshop took place over two days - October 22 and 23. It addressed how research from the developmental cognitive sciences can be analysed in order to extract requirements on cognitive systems. We invited Felix Warneken who is a developmental cognitive scientist from the MPI-EVA in Leipzig, along with Collin Bannard - a develpmental computational linguist, to interact with cognitive roboticists and systems people. The workshop was quite insightful.

The main conclusions are summarized here.

Related to this activity, we present an example case study here

Presentations from the workshop will be available soon.

Learning Journey: The Role of Information in Cognition

Venue: Leipzig
Date: March 28-29, 2008
Main discussants: Olaf Breidbach, Thomas Breuel, Peter Dayan, Sophie Deneve, Christoph von der Malsburg
Organized by: Jürgen Jost
Report written by: Eckehard Olbrich, Nils Bertschinger, Jürgen Jost

Information
According to Shannon, information is reduction of uncertainty about specific signals drawn from a known distribution. Information theory is a tool for deriving bounds and checking for optimality.

Bayesian decision theory and information processing
Bayesian decision is a provably optimal method for taking decisions on the basis of incomplete information. Bayesian parameter estimation avoids the problem of overfitting. Thus, Bayesian schemes yield formal methods against which actual performance of both neural and artificial cognitive systems can be compared.

The question is whether, how, and in particular at which scale (individual synapses or neurons, populations of neurons, entire brains)such schemes are implemented. It is quite possible that the same scheme applies at different scales.

Bayesian approaches raise the question of the priors. These then must be obtained from evolutionary and developmental processes. Likewise, Bayesian decision depends on a loss function that needs to be specified.

In specific applications or simple organisms, the decision function can be directly implemented whereas in more complex contexts, it will be derived from a more general (Bayesian) scheme. Also, slow decisions in novel situations coexist with fast automatic decision schemes in situations that have become standard for an organism.

Constraints
Constraints can force more efficient coding through systematic exploration of regularities, as investigated in statistical learning theory. In principle, as Bayesian schemes are optimal and in particular do not overfit, within such a scheme, constraints are not needed. However, as an optimal Bayesian solution may need an arbitrarily long time and an arbitrarily large amount of resources, constraints cannot be neglected, and it is a challenge to develop Bayesian schemes under time or resource constraints.

Neural networks
Standard artificial neural networks may lack the structure to handle large amounts of input data in a short time and under varying contexts. Therefore, the scheme may have to be enlarged to allow for flexible dynamical links, possibly coordinated by specific cells. Networks have to be preadapted to their environment. They need to be flexible enough to combine invariant pieces in variable combinatorial patterns. Thus, both some prewired hierarchical structure and self-organization are needed to find optimal solutions.

Evolution
Arthropods use relatively simple and universal devices to handle inputs even when the underlying sensory devices systematically change. Thus, the same neural feature can be embedded into different functional contexts in different species. For instance, simple standard motor neuron patterns can be coupled to changing peripheries. Also, evolution works on ontogenetic development. Can such features be also building blocks for artificial cognitive systems?

Learning Journey: Mechanisms of Coordination in a Cognitive System

Learning Journey: Applications, Requirements, Capabilities

Venue: Ecole des Mines, Paris
Date: March 20, 2008
Organized by: Patrick Courtney, Bill Sharpe, and Erik Hollnagel

Executive Summary
A significant research programme in cognitive systems is now underway. This focuses on the developing and technology and the necessary scientific understand to provide significant levels of autonomy and decision making into computer-based systems. Active research approaches in the area range broadly, from traditional rule-based AI, through to connectionist, dynamical and emergent systems and include embodied systems combining computing and robotic systems.

One major practical motivation for the development of cognitive systems is to overcome the problems faced by traditional computer systems in dealing robustly with the uncertainties and changing demands that characterise the real world. Potential applications cited span a very broad range and have included care-giver robots, and easier-to-use interfaces. In order to make link with the developing discipline of cognitive systems, a new level of multidisciplinary dialogue on the centre ground is needed to build the concepts and community of Cognitive Systems. The objective on this activity is to identify stepping stones between applications and research into cognitive systems.

The domains of aerospace and automotive are examined with a view to identifying how the issues of autonomy and decision making are addressed, and trends which call for increased autonomy. Two specific areas are selected for further discussion: context-aware detection; and the coordination of multiple cognitive agents using contracts. Other possible areas are briefly indicated and potential next steps are suggested.

The full report is available here.

Learning Journey: Attention, Vision, Robotics, and Cognitive Systems

Venue: Santorini
Date: 15 May 2008
Participants: Monique Thonnat, Ute Bauer-Wersing, Antonis Argyros, Jannik Fritsch, John Tsotsos, Heiko Wersing
Report written by: Markus Vincze

Goals
The attempt was to discuss where there are areas of imminent need for work or what might be limits to really advance the field. It was a rather free discussion. Given the mix of people subdisciplines such as vision and robotics have been stressed on purpose. The summary below first lists scientific challenges and second recommendations for implementation.

Scientific Challenges

  • Need to understand: there is a lack of understanding the underlying concepts in many areas related to vision and their relation to behaviours, functions and tasks. However, such an understanding is necessary to explain the phenomena and replicate them in an artificial cognitive system.
  • Semantic abstraction: while data can be hierarchically clustered, mined or other nice things done with it, there is still a gap from data to semantics and an abstraction in terms of semantic concepts that relate to data. This also includes a clear definition of semantics. And it applies not only to vision and robotics but all sensors or other parts of cognitive systems.
  • Is engineering different? Building an artificial cognitive system might require also other sorts of “understanding”, namely more specific to showing the replication of what has been understood. Nevertheless, there needs to be a more scientific approach to show this (See next).
  • Experimental discipline: while other disciplines require the replication of experiments until some finding is generally accepted, this seems not to be the case in computer vision, robotics and maybe cognitive systems. There is the need to establish experimental procedures in these fields.
    1. Databases need to be evaluated themselves and clearly stated what they really exemplify before using them as any valuable measure (e.g., Pinto, Cox, DiCarlo 2008).
    2. A hard but maybe fruitful way to explain an achievement or understanding is to convince persons from other disciplines about what we have learned. They are less trapped in the conventions of a community.

Recommendations for Implementation

  • Progress might require rather a specialisation, looking in detail at one part to really understand it, than building complete systems.
  • Proposals tend to promise too much. A sober description of what will be achieved and by what means should be emphasised. This includes honesty to sit back and clearly say what has been achieved at project end rather than present project activity.

Reference
Pinto N, Cox DD, DiCarlo JJ (2008) Why is Real-World Visual Object Recognition Hard? PLoS Comput Biol 4(1): e27. doi:10.1371/journal.pcbi.0040027

Learning Journey: Affect and Emotion in Cognition

Venue: Osaka, Japan
Date: 11-12 July 2008
Organized by: Robert Lowe, Anthony Morse, Tom Ziemke, University of Skövde, Sweden
Report written by: Tom Ziemke

Facts & figures
The workshop on "The role of emotion in adaptive behavior and cognitive robotics" was held July 11-12 in Osaka, Japan, in connection with the SAB 2008 conference. Approximately 25 researchers participated, including three invited speakers (Lola Canamero, Ron Chrisley, Marc Lewis) and six (other) euCognition members (Bartneck, Inderbitzin, Manella, Murray, Raif, Rank) whose participation was supported by euCognition.

Scope of the workshop
Affect and emotion have recently become a hot topic in the study of adaptive behaviour and embodied cognition in both natural and artificial systems. However, the regulatory role of affect/emotion, the relevant underlying mechanisms, and the interaction between affective/emotional and cognitive processes are still not well understood. In order to develop a better understanding of the role of affect/emotion in adaptive behaviour and cognitive robotics this workshop brought together approximately researchers working on affective mechanisms, emotional agents, as well as social interaction & human-robot interaction.

Speakers
Invited talks by Lola Canamero (University of Hertfordshire, UK), Marc Lewis (Unversity of Toronto, Canada), and Ron Chrisley (University of Sussex, UK). Other speakers were Vadim Bulitko, John Murray, Silvia Rossi, Francesco Manella, and Rob Lowe.

Further information
More information about the workshop, including detailed program and proceedings, is available at the workshop website. A special issue with selected extended papers will be published by the ''Connection Science'' journal in 2009.

Learning Journey: Cognition and Culture - the Enactive Approach

Venue: Watershed Media Centre, Bristol, UK
Date: June 2009
Organized by: Fred Cummins and Bill Sharpe

The workshop was scheduled to run in December 2008, but due to the financial situation in euCognition the budget became too constrained for us to be able to commit to the costs in time to secure the participation of some of the key international participants. We have therefore postponed the workshop until June 2009 and will submit a proposal to euCogII. Below is the call that we sent out. We secured commitments to attend from our target of about 25 people from a wide range of backgrounds including cognitive systems, anthropology, media and arts.

A workshop web site has been set up where people can register their interest.

Cognition and Culture: an enactive view
The enactive approach to understanding cognitive systems is relatively new. Within that approach, a set of concepts and a novel vocabulary are being collectively refined that allow new ways of understanding complex systems that arise, are maintained, and die within complex environments, and how meaning arises for agents within them. We believe these tools and this vocabulary may be of use to researchers in allied fields, including but not limited to arts, media and cultural studies, anthropology, economics, and sociology, and will promote mutual learning amongst them all. A central issue of interest that brings these disciplines together is the notion of ‘value’ and how this can be viewed as the dynamic structural relationships between identity and context. Further details of the research orientation are below.

This workshop is intended to start the dialogue, and it is hoped it will be the first of three annual meetings that will play a significant role in creating a new level of interdisciplinary research enquiry. Attendance will be limited to 25 people and we are looking to balance the interest from different disciplines and find people who will have a strong interest in continuing the research conversation after the meeting. The agenda of follow-up activities will be developed by all the attendees as a part of the workshop.

Organisers:
Fred Cummins, Univ College Dublin
Bill Sharpe

Organising Committee:
Paul Appleby, BBC
Judith Aston, School of Creative Arts, Univ of the West of England
Jon Dovey, School of Creative Arts, Univ of West of England
Phil Stenton, Hewlett Packard Labs
David Vernon, Khalifa University, and euCognition coordinator

Research Agenda
The purpose of this workshop is to help to develop a robust vocabulary and set of concepts that are capable of sustaining dialogue between researchers in cognitive systems, cognitive science, anthropology, economics, arts, media, and culture by using the insights and approaches of the enactive approach to cognition.

Over the last two decades it has become clear that cognition is strongly entwined with the physical structure of the body and its interaction with the environment. The physical body and its actions together play as much of a role in cognition as do neural processes, and human intelligence develops through interaction with objects in the environment and it is shaped profoundly by its interactions with other human beings. A body of research has emerged that tackles these questions in ways inspired by the early work of Maturana and Varela and going under the name of enactive cognition. Key points of this approach are:

  • Enactive approaches are intrinsically embodied and the physical instantiation plays a pivotal constitutive role in cognition. A strong consequence of this is that one cannot short-circuit the ontogenetic development (structural development of the individual in its context) because it is the agent's own experience that creates its understanding of the world in which it is embedded.
  • Since cognition is dependent on the richness of the system's action interface and since the system's understanding of its world is dependent on its history of interaction, a further consequence of the enactive approach is that, if the system is to develop an understanding of the world that is human-like, the system requires a morphology that is compatible with a human.
  • Enactive approaches assert that skilled action, rather than propositional knowledge representation, is the foundation underpinning intelligent behaviour. A consequence of this is that skilful learning through doing is at the heart of intelligent systems.
While this agenda is being driven forward from interests in constructing new cognitive systems that can find their place in human society, the language and concepts resonate strongly with interests in arts, media, and culture, anthropology and other disciplines. In particular, it allows a fresh discussion on long standing questions about how we can understand the way in which the arts and culture brings forth meaning and value in people's lives. At the heart of this is the process of making meaning in, and for, a life, rooted in its social nature. This allows an exploration of value that moves effectively into a discussion of art and culture as a mode of collectively experiencing the world and how they are kept perpetually in play at both an individual and societal level. While this has been explored for hundreds of years, there is no widely accepted language available for the task. This workshop is based on the proposition that enactive cognition is a useful bridging framework that will allow concepts such as identity, meaning, value, autonomy, and related ideas to travel between different disciplines in mutually informing ways. In particular we aim to start the dialogue by exploring the following themes:
  • The enaction of meaning is the condition of life. A living being and its environment enact, or bring forth, a world in which there are values ‘for’ the organism – things become food, shelter, threats, lovers, and so on. Thus a living thing casts a web of significance upon its world and together all the members of the system enact their mutual values. These values reflect the significance of a situation for the maintenance and renewal of the individual's identity in a dynamic and reflexive social and cultural context.
  • Every person participates in culture as a condition of being human: culture is composed of webs of significance which are sustained by everyone’s engagement with them; meaning is enacted in the moment, in a life, in a community, in a history, and towards the future.
  • Art can be viewed as a ‘making’ which creates the possibility of new sorts of meaning to arise in human life and for an individual identity to be explored, renewed, transformed, and reconceived. While there are some people whose vocation is the artistic making, we all make our own lives and in this sense are all artists of our own lives.
  • Autonomy is a negotiable concept, and the rational characterization of behaviour of many interacting agents within a system or ecology may require the boundaries of autonomy to be drawn in a variety of ways, no one of which is exclusively privileged.
None of these ideas are original in themselves but what is new is the sustained effort to make systematic connections for mutual benefit between disciplines that do not currently explore them together but which seem to share many fundamental ideas.

Background
This workshop has its origins in an ongoing project of the International Futures Forum for the MMM programme that has been exploring whether the system notions of ecology are useful in understanding how arts are maintained in society. That discussion has started to bring together many research strands that converge on the central questions outlined above.

http://www.internationalfuturesforum.com/projects.php?id=24
http://www.missionmodelsmoney.org.uk