IEEE CIS Newsletter on Cognitive and Developmental Systems


(Pierre-Yves Oudeyer) #1

The IEEE CIS Newsletter on Cognitive and Developmental Systems is produced by the IEEE CIS Technical Committee on CDS.

It was previously named IEEE CIS Autonomous Mental Development Newsletter.
The AMD Newsletter web site was previously hosted at http://www.cse.msu.edu/amdtc/amdnl/ where archives of previous and current issues can also be found.

Aims and scope

This newsletter is a medium to share information across computational developmental sciences research communities. It participates to animate an interdisciplinary research community around modeling sensorimotor, cognitive and social development, with contributions from biology, neuroscience, psychology (cognitive science), computer science, electrical engineering, and mathematics.

The newsletter is open access. It was created in 2004 and is published twice a year.

Each issue features:

  • A scientific interdisciplinary dialog around cognitive and developmental systems. There is first a dialog initiation on an important multidisciplinary topic related to developmental sciences. Then, several answers by researchers from computational, biological and human sciences backgrounds are presented, followed by a synthesis by the dialog proposer. The goal of the dialog is to let researchers express their own point of views in a concise manner and around the same topic. If you wish to propose a new dialog, please contact the editor-in-chief.
  • Committee News: CDS TC events and news
  • General News: announcements of related conferences, workshops, books, journals, etc.
  • The table of content of the latest release of IEEE Transactions on Cognitive and Developmental Systems

The success of this newsletter depends on your support! Please send the material you recommend and your suggestions to the editor.

Editors

Editor in chief
Pierre-Yves Oudeyer, Newsletter Editor (2008 - present)
INRIA Bordeaux - Sud Ouest
351, cours de la Liberation Batiment A29 33405
France
Email: pierre-yves.oudeyer [at] inria [dot] fr

Assistant editor
Fabien Benureau, Editorial Assistant (July 2013 - present)
Cognitive NeuroRobotics Unit,
Okinawa Institute of Science and Technology
1919-1 Tancha, Onna, Okinawa
Japan
Email: fabien.benureau [at] oist [dot] jp

Founder
The Newsletter was founded in 2004 by Juyang Weng, Department of Computer Science and Engineering, Michigan State University, East Lansing, MI 48824, USA.
Email: weng [at] cse [dot] msu [dot] edu

Past Newsletter Editors-in-chief
Shuqing Zeng, GM Research and Development Center, US (2006-2007);
Yilu Zhang GM Research and Development Center (2004-2005);

Past Editorial Assistants:

Adrien Baranes, Inria, France (2008 - May 2013);
Paul Spears, Michigan State University, US (Oct. 2010 - May 2013)
Matthew D. Luciw, Michigan State University, US (2009 - April, 2010)
Zhengping Ji, Michigan State University, US (2008)
Stephanie Beal, Michigan State University, US (2006)
Kim Thompson, Michigan State University, US (2005)
Nicholas Wassil, Michigan State University, US (2004)

Acknowledgements

Funding and support for past and present edition and distribution of this newsletter has been provided by IEEE CIS, Michigan State Univ, US, as well as Inria (France).

Open access newsletters

CDS TC Newsletter Vol. 15, No. 1, 2018

-Dialog initiated by Celeste Kidd: “Curiosity as Driver of Extreme Specialization in Humans”
-Responses by: Elizabeth Bonawitz, Maya Zhe Wang, Brian Sweis, Benjamin Hayden, Susan Engel, Abigail Hsiung, Shabnam Hakimi, Alison Adcock, Moritz Daum, Arjun Shankar, Tobias Hauser, Goren Gordon and Perry Zurn.

  • Editorial note: Curiosity-driven learning is probably one of the most fundamental mechanisms in human learning, and yet it is also probably one of the least understood. Broadly construed as spontaneous exploration and engagement with activities or material without any extrinsic goal (as opposed to searching for information useful for an extrinsic goal), many mysteries remain to be uncovered. What are the causal links between curiosity and learning? How does prior knowledge about a topic or an activity relates to curiosity about this topic? What is the role of curiosity in life-span development? Can human curiosity explain the apparently unique tendency of humans for extreme specialization? Reversely, how do different forms of curiosity (diversive or specific) evolve as children grow up and become adults? While early computational models of curiosity propose theoretical approaches to understand their cognitive mechanisms, how can we understand the affective/ emotional dimensions of curiosity? And how has the linguistic concept of “curiosity” evolved in occidental culture?
    CDSNL-V15-N1.pdf (1.3 MB)

CDS TC Newsletter Vol. 14, No. 2, 2017

-Dialog initiated by Matthias Rolf, Lorijn Zaadnoordijk and Johan Kwisthout: “One developmental architecture to rule them all?”
-Responses by: Niels Taatgen, John Spencer, Gary Jones, Gerard Wolff, Clément Moulin-Frier and Paul Verschure

  • Editorial note: Humans have a unique capability to achieve and learn a wide diversity of skills of all kinds, from low-level sensorimotor skills to very abstract linguistic or mathematical skills. Is it possible to develop theories of how general cognitive architectures can display such a general flexibility for skill learning? This dialog adresses this question, and discusses whether and how it would be useful both epistemologically and in practice to aim towards the development of a “standard integrated cognitive architecture”, akin to “standard models” in physics, or whether focusing on simple and partial models should be a better approach. In particular, this question is discussed in the context of understanding development in infants, and of building developmental architectures, thus addressing the issue of architectures that not only learn, but that are adaptive themselves.
    CDSNL-V14-N2.pdf (976.8 KB)

CDS TC Newsletter Vol. 14, No. 1, 2017

-Dialog initiated by Jun Tani: “Exploring Robotic Minds by Predictive Coding Principle”
-Responses by: Andy Clark, Doug Blank, James Marshall, Lisa Meeden, Stephane Doncieux, Giovanni Pezzulo, Martin Butz, Ezgi Kayhan, Johan Kwisthout and Karl Friston

  • Editorial note: The idea that the brain is pro-actively making predictions of the future at multiple levels of hierarchy has become a central topic to explain human intelligence and to design general artificial intelligence systems. In this issue, Jun Tani, who has been studying recurrent neural networks models of sensorimotor development for the last 20 years, introduces a dialog to ask whether hierarchical predictive coding enables a paradigm shift in development robotics and AI. Responses give their perspectives on this topic. In particular, they discuss the importance of various complementary mechanisms to predictive coding, which happen to be right now very actively researched in artificial intelligence: intrinsic motivation and curiosity, multi-goal learning, developmental stages (also called curriculum learning in machine learning), and the role of self-organization. They also underline several major challenges that need to be addressed for general artificial intelligence in autonomous robots, and that current research in deep learning fails to address: 1) the problem of the poverty of stimulus: autonomous robots, like humans, have access to only little data as they have to collect it themselves with severe time and space constraints; 2) the problem of information sampling: which experiments/observations to make to improve one’s world model. Finally, they also discuss the issue of how these mechanisms arise in infants and participate to their development.
    CDSNL-V14-N1.pdf (1.1 MB)

CDS TC Newsletter Vol. 13, No. 2, 2016

-Dialog initiated by Olivia Guest and Nicolas Rougier: “What is Computational Reproducibility?
-Responses by: Konrad Hinsen, Sharon Crook, Gaël Varoquaux, Todd Gureckis and Alexander Rich, Robert French and Caspar Addyman, and Celeste Kidd.
-Editorial note: Computational models of cognitive and developmental living systems need to address several major challenges in order to achieve scientific impact: reproducibility, replicability, but also reusability in an interdisciplinary community. Indeed, one needs to ensure that models’ implementations and experimentations match their high-level specifications. It is also key to conduct alternative implementations and experimentations to distinguish which aspects of these models are key concepts, and which others are tools for experimenting these concepts. Last but not least, models should be understandable and reusable by other researchers who are not always themselves computational experts, which is facilitated when they are delivered in a way that allows non-experts to directly “play” with these models. CDSNL-V13-N2.pdf.pdf (734.6 KB)

CDS TC Newsletter Vol. 13, No. 1, 2016

-Dialog initiated by John Spencer, Mark Blumberg, David Shenk: “Moving Beyond Nature-Nurture: a Problem of Science or Communication?
-Responses by: Bob McMurray, Scott Robinson, Patrick Bateson, Eva Jablonka, Stephen Laurence and Eric Margolis, Bart de Boer, Gert Westermann, Peter Marshall, Vladimir Sloutsky, Dan Dediu, Jedebiah Allen and Mark Bickhard, Rick Dale and Anne Warlaumont and Michael Spivey.
-Editorial note: The dialog highlights several socio-epistemic factors that make the nature/nurture view so robust, and in particular the fact that several forms of utility of this view may counterbalance its lack of veracity: it facilitates the work of scientists, as it affords experiments where single dependent variables are studied, which are easier and faster to run, explain and publish; it facilitates communication to the larger public as it relies on a stable set of words and concepts that have been popularized by talented scientific writers for decades; it has been used in many studies of human-specific abilities such as language and maths skills that have stimulated wonder about infant capabilities in both scientists and the general public.
While the dynamical systems approach to development has been thriving with many impressive studies showing the importance of multi-factorial multi-scale processes of pattern formation, it has been struggling with several challenges to broaden its impact: explanations are often more complicated as they highlight the complexity of development; experimental paradigms to study dynamical processes have mostly focused on the development of sensorimotor capacities that are common with other animals, rather than on higher-level human cognition; there is a lack of a strong shared vocabulary and ontology for speaking about self-organization, pattern formation and dynamical systems in cognitive development. However, as several authors argue, the scientific landscape is progressively evolving toward a « synthesis on the horizon ». CDSNL-V13-N1.pdf (938.3 KB)

CDS TC Newsletter Vol. 12, No. 2, 2015

  • Dialog initiated by Stéphane Doncieux: “Representational redescription: the next challenge?
  • Responses by: Jessica Kosie and Dare Baldwin, Georges Konidaris, Freek Stulp and Timothy Hospedales, Paul Verschure and Giovanni Pezzulo, Frank Guerin, Paul Abelha and Bipin Indurkhya.
  • Editorial note: This issue’s dialog explores various challenges of how and why capabilities to change representations could be happening in robots. In particular, contributors discuss how new representations can be formed out of the dynamic interaction between learning algorithms, cognitive architecture and their physical and social embodiment. The dialog highlights the need for a multiplicity of processes, happening at different time scales and levels of abstraction, ranging from fast opaque low-level sensorimotor learning to slow, more transparent rule-based learning.
    CDSNL-V12-N2.pdf (1.6 MB)

AMD TC Newsletter Vol. 12, No. 1, 2015

  • Dialog initiated by Janet Wiles: “Will social robots need to be consciously aware?”.
  • Responses by: Axel Cleeremans, Yasuko Kitano, Cornelius Weber and Stefan Wermter, Justin Hart and Brian Scassellati, Juyang Weng, Guy Hoffman and Moran Cerf.
  • Editorial note: Several dimensions of the question stand out. First, as we are very far from understanding what “consciousness” is, it appears that building robots capable of various forms of self- and other- aware- ness, and importantly how they can develop these capabilities progressively, can be very useful in the quest to unveil the underly- ing mechanisms. Second, as consciousness is a multiscale complex systems, multiple approaches and perspectives need to be taken in this process of robot building. Third, when one looks at applications, it is the function, and not the nature, of consciousness which becomes the relevant angle of analysis, and several ethical questions arise.
    AMDNL-V12-N1.pdf (631.1 KB)

AMD TC Newsletter Vol. 11, No. 2, 2014

  • Dialog initiated by Katharina Rohlfing, Britta Wrede and Gerhard Sagerer: “Trained for everything? The challenges of interdisciplinary education
  • Responses by: Giulio Sandino and David Vernon, Franck Ramus and Thérèse Collins, Maha Salem, Juyang Weng, Thomas Schultz, and Christina Bergmann.
  • Editorial note: Understanding sensorimotor, cognitive and social development in animals and robots is a truly interdisciplinary endeavour. Development happens through the pro- gressive organisation of coupled complex systems, at various spatio-temporal scales, and covering a large diversity of levels of abstraction, ranging from coupled mechan- ical dynamics for low-level body control to higher-level conceptual development. As thinkers like Edgar Morin have been arguing for a long time, understanding such complex systems in the living requires the combination of multiple perspectives of analysis, coming from disciplines as varied as neuroscience, psychology, robotics, mathematics, physics, biology, philosophy, linguistics, anthropology, and primatology. Yet, this raises significant challenges in terms of training scientists, both in terms of scientific methods and in terms of career management in academia, as discussed in the dialog of this issue of the AMD Newsletter. AMDNLV11N2.pdf

AMD TC Newsletter Vol. 11, No. 1, 2014

  • Dialog initiated by Katerina Pastra: “***Autonomous Acquisition of Sensorimotor Experiences: Any Role for Language?***”
  • Responses by: Rick Dale, Katharina Rohlfing and her colleagues, Gary Lupyan, Catriona Silvey, Emmanuel Dupoux, and Kerstin Fisher.
  • Editorial note: This dialog discusses the interaction between the formation of linguistic and conceptual/sensorimotor structures, and in particular on the hypothesis that language as a communication system may have evolved as a byproduct of language as a tool for organizing conceptual structures.
    AMDNL-V11-N1.pdf (1.2 MB)

AMD TC Newsletter Vol. 10, No. 2, Fall 2013

  • Dialog initiated by Peter Ford Dominey: “***How are Grammatical Constructions Linked to Embodied Meaning Representations?***”
  • Responses by: Brian McWhinney, William Croft, Alistair Knott, Michael Arbib and Victor Barrès, and Juyang Weng
  • Editorial note: This dialog investigates how structured embodied meanings, such as the temporal enfolding of events, and grammatical constructions to talk about these meanings, can be developed. A particularly illuminating hypothesis is that structured conceptual and linguistic meanings form within the flow of structured social inter- action loops, which themselves act as a frame that guides the infant. This process of learn- ing shall not be conceptualized as a process where building blocks or atoms of conceptual structures are first learnt, then combined, but rather as a process starting from learning how to pragmatically deal with global situa- tions, which are then later on decomposed into subunits. Further, it is actually suggested that linguistic grammatical constructions interact with a sensorimotor system which has itself a full-blown grammatical organization, which opens in itself stimulating research avenues.
    AMDNL-V10-N2.pdf (1.2 MB)

AMD TC Newsletter Vol. 10, No. 1, Spring 2013

  • Dialog initiated by John Weng: “***Modeling AMD and its Application to Assistive Robotics: Closed Skull or Not?***”
  • Responses by: Yiannis Demiris, Adriana Tapus, Manuel Lopes, Katharina Rohlfing and Britta Wrede, Anthony Morse and Yoonsuck Choe
  • Editorial note: In socially assistive robotics, a field of immense societal value, robots need to adapt continuously to the particularities, preferences and evolving needs of the human(s) they were made to help, as well as to cluttered human environments. These requirements meet the objectives of developmental robotics, where methods for lifelong incremental learning are studied, with a strong guiding role for developmental constraints coming both from within and outside the learner. Yet, many important and complicated issues remain open. In particular, while there are arguments stating that concepts and representations in assistive robots should be emergent and not hand tuned by engineers, it is also crucial, in such application fields, that safety, trust, acceptability and usefulness for the humans can be ensured. While in a distant future we can see how these two lines could be converging, in the short term they can be incompatible for real-world applications.
    AMDNL-V10-N1.pdf (3.4 MB)

AMD TC Newsletter Vol. 9, No. 2, Fall 2012

  • Dialog initiated by Denis Mareschal: “***Children’s Natural Learning: Why Development Really Does Matter!***”
  • Responses by: Linda Smith, Annette Karmiloff-Smith, Peter Dayan and Gert Westermann
  • Editorial note: Development happens through the dynamic interaction of biological and physical components at several scales of time and space. These interactions range from the dynamics of gene expression within cells to inter-organ chemical signalling during embryogenesis to the progressive shaping of brain structures through physical and social interaction. What are the adequate levels of abstraction for modelling and understanding such a complex system? Opinions differ on what should or should not be abstracted. But all converge to a needed shift for 21st century developmental science: “innate” or “learned” are outdated concepts that are
    not suited for understanding human behaviour and cognition. There are no “start” or “end”, but only continuous processes of change that grow our bodies, our minds and our cultures.
    AMDNL-V9-N2.pdf (3.2 MB)

AMD TC Newsletter Vol. 9, No. 1, April 2012

  • Dialog initiated by Katharina Rohlfing and Britta Wrede: “***What Novel Scientific and Technological Questions Does Developmental Robotics Bring to HRI? Are we Ready for a Loop?***”
  • Responses by: Yukie Nagaï, Andrea Thomaz, Lakshmi Gogate, Adriana Tapus, Stevo Bozinoski, Anthony Morse, Thomas Hannagan, Rachel Wu, and Helge Ritter
  • Editorial note: New fundamental concepts, experimental methods and techniques have been growing recently about the understanding of developmental social learning in humans and robots. They are at the crossroads of research in developmental robotics, human-robot interaction, and the psychology and neuroscience of human social learning. This newsletter features a stimulating dialog illustrating the high-importance of these concepts and techniques both for building machines capable of learning in interaction with non- engineer humans across an extended time-scale, but also for better understanding human development.
    AMDNL-V9-N1.pdf (4.8 MB)

AMD TC Newsletter Vol. 8, No. 2, October 2011

  • Dialog initiated by Yaochu Jin and Yan Meng: “***Evolutionary Developmental Robotics – The Next Step to Go?***”
  • Responses by: Josh Bongard, Daniel Polani, Stéphane Doncieux, Fumiya Iida, Nicolas Bredeche, Jean-Marc Montanier, Simon Carrignon, and Gunnar Tufte
  • Editorial note: This dialog presents a stimulating exploration of the frontiers of developmental and evolutionary robotics, within the so-called evo-devo approach. It asks questions about the interaction between phylogeny, ontogeny and epigenesis, as well as about the interaction between morphological and mental development.
    AMDNL-V8-N2.pdf (1.3 MB)

AMD TC Newsletter Vol. 8, No. 1, April 2011

  • Dialog initiated by John Weng: “***Are natural languages symbolic in the brain?***”
  • Responses by: Stevan Harnad, Jürgen Schmidhüber, Aaron Sloman, Angelo Cangelosi, and Yuuya Sugita and Martin Butz.
  • Editorial note: This dialog directly addresses a quite controversial but fundamental issue: is the symbol grounding problem a real problem? Are symbols really fundamental for understanding human cognitive development, or are they just a modern conceptual invention of modern human culture? Respondants interestingly mix philosophical and mathematical arguments, showing how recent technical advances can illuminate old questions and vice versa, how philosophical theories can either question or sup- port the assumptions and concepts of modern technical approaches.
    AMDNL-V8-N1.pdf (1.3 MB)

AMD TC Newsletter Vol. 7, No. 2, October 2010

  • Dialog initiated by Gianluca Baldassarre and Marco Mirolli: “***What are the key open challenges for understanding autonomous cumulative learning of skills?***”
  • Responses by: Andy Barto, Kevin Gurney and Peter Redgrave, Juergen Schmidhüber, Kenji Doya, Jun Tani and Pierre-Yves Oudeyer
  • Editorial note: This dialog focuses on a topic that has recently gathered a lot of interest in developmental robotics: intrinsic motivation and its relation to open-ended cumulative learning of skills.
    AMDNL-V7-N2.pdf (1.5 MB)

AMD TC Newsletter Vol. 7, No. 1, April 2010

  • Dialog initiated by Angelo Cangelosi: “***The Symbol Grounding Problem Has Been Solved: Or Maybe Not?***”
  • Responses by: Stevan Harnad, Luc Steels, Aaron Sloman, Stephen Cowley, Vincent Müller, Carol Madden, Peter Ford Dominey and Stéphane Lallée
  • Editorial note: This dialog focused on the symbol grounding problem in relation to advances in robotics, machine learning and artificial intelligence.
    AMDNL-V7-N1.pdf (1.7 MB)

AMD TC Newsletter Vol. 6, No. 2, October 2009

  • Dialog initiated by Max Lungarella: “***Developmental Robotics: From Black Art to Discipline Guided by Principles?***”
  • Responses by: wJosh Bongard, Franck Guerin, Chrystopher Nehaniv, Linda Smith, Alex Stoychev, Juyang Weng, and Patricia Zukow-Goldring
  • Editorial note: This special issue features a dialog of essential importance for developmental robotics: Can we identify principles that specifically define developmental robotics? And should we strive to organize research around such principles or rather consider tinkering and ad hoc investigations as a strength or a marker of scientific innovation as argued by philosopher Feyerabend?
    AMDNL-V6-N2.pdf (994.8 KB)

AMD TC Newsletter Vol. 6, No. 1, April 2009

  • Dialog initiated by Kerstin Dautenhahn: “***Are robots Beneficial to Children with Autism?***”
  • Responses by: Tamie Salter, Sarah Parsons, Philippe Gaussier and Pierre Andry, Patrizia Marti, Henrik Hautop Lund, Kazuyoshi Wada
    AMDNL-V6-N1.pdf (883.5 KB)

AMD TC Newsletter Vol. 5, No. 2, October 2008

  • Dialog initiated by Paul Fitzpatrick: “***Which skills most need development?***”
  • Responses by: Charlie Kemp, Hideki Kozima, Matthew Schlesinger, Claus Von Hofsten, Juyang Weng, Lola Cañamero
    AMDNL-V5-N2.pdf (1004.0 KB)

AMD TC Newsletter Vol. 5, No. 1, April 2008

  • Dialog initiated by Juyang Weng and Jay McClelland: “How the Mind Works and How the Brain Develops
  • Responses by: Minoru Asada, Gedeon O. Deák, John K. Tsotsos, Jeffrey L. Krichmar, Olaf Sporns, Paul Werbos
    AMDNL-V5-N1.pdf (616.3 KB)

AMD TC Newsletter Vol. 4, No. 2, October 2007

  • Dialog initated by Minoru Asada: “***Should Robots Develop as Human Infants Do?***”
  • Responses by: Britta Wrede, Katharina J. Rohlfing, Yuki Nagai and Gehard Sagerer, Hideki Kozima , Juyang Weng, Shoji Itakura
    AMDNL-V4-N2.pdf (2.7 MB)

AMD TC Newsletter Vol. 4, No. 1, April 2007

  • Dialog initiated by Pierre-Yves Oudeyer and Frédéric Kaplan: “***How can we assess open-ended development?***”
  • Responses by: Dan Burfoot and Max Lungarella, Christopher G. Prince and Nathan A. Helder, Paul Fitzpatrick, Charlie Kemp, Doug Blank, Jim Marshall, and Lisa Meeden
    AMDNL-V4-N1.pdf (2.0 MB)

AMD TC Newsletter Vol. 3, No. 2, November 2006

  • Dialog initiated by Brian Scassellati: “Is our toolbox full?”
  • Responses by: Chad Jenkins, Stan Franklin, Juyang Weng, M. Xie, Pierre-Yves Oudeyer and Frédéric Kaplan
    AMDNL-V3-2.pdf (2.7 MB)

AMD TC Newsletter Vol. 3, No. 1, April 2006

AMDNL-V3-1.pdf (1.3 MB)

AMD TC Newsletter Vol. 2, No. 2, October 2005

  • Dialog initiated by Stephen Levinson: “***Can Robots Learn Language the Way Children Do?***”
  • Responses by: Stan Franklin, Dave Touretzky, Juyang Weng, Ming Xie, Jayakumar Sadhasivam Kandhasamy, and Kok Heng Leong, Yilu Zhang
    AMDNL-V2-2.pdf (2.1 MB)

AMD TC Newsletter Vol. 2, No. 1, April 2005

  • Dialog initiated by Olaf Sporns: “***What are Potential Principles for the Autonomous Development of Value and Motivation?***”
  • Responses by: Stan Franklin, Jean-Marc Fellous, Pierre-Yves Oudeyer and Frédéric Kaplan, Juyang Weng
    AMDNL-V2-1.pdf (1.5 MB)

AMD TC Newsletter Vol. 1, No. 2, October 2004

  • Dialog initiated by Juyang Weng and Yilu Zhang: “Object Detection and Object Variance in Autonomous Mental Development
  • Responses by: Stephen E. Levinson, Olaf Sporns, Chris Brown, Ming Xie, Zhengyou Zhang
    AMDNL-V1-2.pdf (1.1 MB)

AMD TC Newsletter Vol. 1, No. 1, April 2004

AMDNL-V1-1.pdf (754.5 KB)