Autumn 2011
19th October 2011, 4pm
The cost of Intelligence as a Cybernetic Problem
- John Cummins, DEVA Research
ABSTRACT: As early humans increased in intelligence they may have run into an evolutionary bottleneck. Increased intelligence would probably carry a cost, namely an increase in psychological stress. Chronic stress has an adverse effect on major biological systems and organs. This increased ‘allostatic load’ would impact on longevity, health and reproduction. The talk will suggest how the problem may have been at least partly resolved by evolution. A speculative model (still very much a work in progress) will be described, proposing the existence of a high-level, late –stage comparator in the brain that became subject to an unusual adaptation. The proposed adaptation involved a flexible cognitive bias affecting the accuracy of assessments of control. Somewhat paradoxically, this might have facilitated an increase in the flexibility of high-level problem-solving and other intelligence. The adaptation will be considered in the context of Fodor’s concept of modularity. Some questions and implications for artificial intelligence will also be considered.
BRIEF BIO: John Cummins obtained an LLB in 1977 from Manchester University and qualified as a Solicitor in 1983. After working as a commercial lawyer in the City of London for some years he left law in 1995 to start DEVA Research, commercial undertaking in the field of early-stage life sciences research. Deva has won two DTI ‘SMART’ awards for innovation in anti-infectives. The present talk arises from a personal interest over the last ten years in the evolution of the stress system in our higher primate ancestors and ourselves.
2nd November, 4pm
Rhythm as entrainment: A dynamical, post-cognitivist case study
- Dr. Fred Cummins, University College, Dublin
ABSTRACT: Classical Cognitive Science holds dear to a model of the autonomous individual that serves some purposes, but fails in many cases. I here discuss two ways of viewing rhythm and synchronisation: the classical model, which leans heavily on the notion of prediction by a cognitive system, and a dynamical account, which emphasises embodiment, movement, and coupling among individuals. Only the latter can account for the very tight coupling observed when pairs of speakers read a text in synchrony. The dynamical account also suggests that the uniquely human ability to move in time with music may provide a window onto a basis for skill-sharing, and may thus constitute an important innovation in moving from ape to human.
BRIEF BIO: Dr. Fred Cummins obtained a B.A. in Computer Science, Linguistics and German from Trinity College Dublin in 1991. Postgraduate study took him to the Cognitive Science program at Indiana University. Fred obtained an M.A. in Linguistics in 1996, and a PhD with joint major in Cognitive Science and Linguistics in 1997 at Bloomington. Fred's thesis was an experimental study of English speech rhythm. Fred subsequently completed a one year postdoctoral research at the Department of Lingusitics in Northwestern University, Evanston, IL, and another at the Dalle Molle Institute for Artificial Intelligence in Lugano, Switzerland. In 1999, Dr. Cummins became a lecturer at University College Dublin. From 2000 to 2004, Fred ran a research group in the now defunct Media Lab Europe. Fred's principal professional interests lie in helping to develop a cognitive science that does justice to the reality of subjective experience. In this context he has particular interests in rhythm, speech, language, enaction, ecological psychology, and the metaphysical basis of experience. Dr. Cummins is an unreformed anti-representationalist and thinks in terms of 'dynamic systems theory'.
30 November 2011, 4pm - 5pm
*cancelled and will be rescheduled in the new year.Computational Creativity: Swarms and Paul Jump out of the Box
Mohammad Majid al-Rifaie & Patrick Tresset
In this talk Patrick will introduce the AIkon-II project and Paul the robot, a robotic installation that sketches people's faces.
In his talk, Mohammad, briefly introduces a novel hybrid swarm intelligence algorithm followed by a discussion on the 'computational creativity' of the swarm. The discussion is based on the performance of the swarm through a cooperative attempt to make a drawing. We raise the question on whether swarm intelligence algorithms (inspired by social systems in nature) are possibly capable of leading to a different way of producing 'artworks' and whether the swarms demonstrate computational creativity in a non-representational way.
BRIEF BIO: Mohammad Majid al-Rifaie obtained a BSc in Computing and Information Systems from University of London, Goldsmiths College, External Programme in 2005. His background is in computing, craftsmanship and journalism and his artistic interests focuses on the inter-connections between artificial intelligence, swarm intelligence, robotics and digital art. Postgraduate study took him to do a PhD course which touched upon Aritificial Intelligence, Swarm Intelligence, Cognitive Science and Robotics at Goldsmits, University of London. Mohammad's thesis focuses on the significance of information sharing in population-based algorithms (e.g. Swarm Intelligence). Mr. al-Rifaie's current research interests, in addition to the role of information sharing, lie in understanding the impact of freedom and autonomy in computational creativity.
BRIEF BIO: Patrick Tresset, a French artist/scientist currently based in London, uses what he calls “clumsy robotics” to create autonomous cybernetic entities that are playful projections of the artist. He co-directs the Aikon-II project with Frederic Fol Leymarie at Goldsmiths College, University of London.
The Aikon-II project investigates the observational sketching activity through computational modeling and robotics. The project also provides a rich ground for an artist to examine issues in creativity, and in exploring
robotics systems as a source of potential augmentation of one's creative capacity. The work also seeks to engage with the public in softening the artificial divide between the "two cultures" of contemporary art & science.
7th December, 4pm
Auditory Selective Attention in Real-Room Reverberation
ABSTRACT: Typical listening situations consist of both multiple sound sources and numerous reflecting surfaces, so each sound is accompanied by a multitude of delayed, attenuated copies. This “reverberation” reduces speech intelligibility by masking later arriving portions of the direct sound and degrades the information conveyed by a sound’s temporal envelope. Further, the frequency response of any acoustic environment is typically not flat, creating spectral distortion. But reverberation is not always detrimental. It creates sensations of spaciousness, important in areas like architectural acoustics and music technology. Also, in the case of complex sounds such as speech, early reflections perceptually “fuse” with the direct sound, which can lead to an increase in effective SNR. Moreover, work in the Reading Auditory Lab has provided evidence that the distorting effects of reverberation are ameliorated by perceptual mechanisms which effect constancy. Because the ratio of reflected to direct sound energy increases with distance between source and listener, the distortion of a sound’s temporal envelope increases with distance. Spectral distortion also varies, not only between different environments, but also between different pairs of source-listener positions within the same space. Any detectable difference between competing speech messages will help a listener to track one of them, so it seems reasonable to ask whether these position-specific distortions could serve to aid auditory selective attention when faced with the problem of multiple talkers in a reverberant space. However, the classic inter-aural cues arising from position differences which enable listeners to localise sounds are known to be highly vulnerable to reverberation. So does reverberation help or hinder auditory selective attention? Talker characteristics are the other main source of differences between speech messages, and cues arising from these differences are more robust in reverberation. The series of experiments I will describe in this talk therefore place talker differences in conflict with position differences, while listening in reverberation measured from real rooms, in order to study the effects of realistic reverberation on auditory selective attention.
BRIEF BIO: Simon got his start in auditory psychology in Anthony Watkins’ lab at Reading, working on a project investigating perceptual constancy for spectral envelope distortion. He got his masters in speech and hearing sciences from UCL then completed a PhD in auditory modelling at Sheffield University, supervised by Guy Brown, one of the pioneers of Computational Auditory Scene Analysis (CASA). His PhD work involved using both perceptual experiments and computer modelling to study the role of pitch in concurrent vowel segregation. After returning to Watkins’ lab at Reading a collaborative project between Reading and Sheffield was born. This aims to use the results of perceptual experiments investigating constancy in reverberation when temporal envelopes vary, to inform the development of a computer model for use as a front-end in reverberation-robust artificial listening devices. Most recently, he has also been studying the effects of reverberation on auditory selective attention.